Advancing Neurological AI: A Multi-Collection RAG Architecture and Product Requirements for the Neurology Intelligence Agent¶

Author: Adam Jones Date: March 2026 Version: 0.1.0 (Pre-Implementation) License: Apache 2.0

Part of the HCLS AI Factory -- an end-to-end precision medicine platform. https://github.com/ajones1923/hcls-ai-factory

Abstract¶

Neurological disorders represent the leading cause of disability and the second leading cause of death worldwide, affecting over 3.4 billion people -- 43% of the global population. The World Health Organization estimates that neurological conditions account for 443 million disability-adjusted life years (DALYs) annually, with Alzheimer's disease and other dementias, stroke, migraine, epilepsy, Parkinson's disease, and multiple sclerosis among the most prevalent and devastating. Despite rapid advances in neuroimaging, electrophysiology, genomics, and computational neuroscience, clinical neurology remains hampered by profound data fragmentation: imaging findings exist in PACS, EEG data in separate neurophysiology systems, genomic results in laboratory databases, cognitive assessments in paper-based or siloed electronic records, and treatment evidence scattered across thousands of publications and clinical trials.

This paper presents the architectural design, clinical rationale, and product requirements for the Neurology Intelligence Agent -- a multi-collection retrieval-augmented generation (RAG) system purpose-built for clinical neuroscience. The agent will unify 13 specialized Milvus vector collections spanning neuroimaging (structural MRI, functional MRI, diffusion tensor imaging, PET, SPECT), electrophysiology (EEG, EMG/NCS, evoked potentials), neurodegenerative disease management (Alzheimer's, Parkinson's, ALS, Huntington's), cerebrovascular disease, epilepsy, neuro-oncology, multiple sclerosis, movement disorders, headache medicine, neuromuscular disease, neurogenetics, clinical trials, and neuroradiology literature -- alongside a shared genomic_evidence collection containing 3.5 million variant vectors from the HCLS AI Factory genomics pipeline.

The system extends the proven multi-collection RAG architecture established by six existing intelligence agents in the HCLS AI Factory (Precision Biomarker, Precision Oncology, CAR-T, Imaging, Autoimmune, and the planned Cardiology agent), adapting it with neurology-specific clinical workflows, validated assessment scales, cross-modal neuroimaging-genomics triggers, and structured reporting aligned with AAN practice guidelines. Eight reference clinical workflows will cover the highest-impact neurological use cases: acute stroke triage, dementia evaluation, epilepsy focus localization, brain tumor grading, multiple sclerosis monitoring, Parkinson's disease assessment, headache classification, and neuromuscular disease evaluation.

The agent will deploy on a single NVIDIA DGX Spark ($3,999) using BGE-small-en-v1.5 embeddings (384-dimensional, IVF_FLAT, COSINE), Claude Sonnet 4.6 for evidence synthesis, and four NVIDIA NIM microservices for on-device inference. Licensed under Apache 2.0, the platform will democratize access to integrated neurological intelligence that currently requires tertiary academic medical center infrastructure and subspecialty expertise concentrated in a small number of institutions globally.

Table of Contents¶

Introduction
The Neurological Data Challenge
Clinical Landscape and Market Analysis
Existing HCLS AI Factory Architecture
Neurology Intelligence Agent Architecture
Milvus Collection Design
Clinical Workflows
Cross-Modal Integration
NIM Integration Strategy
Knowledge Graph Design
Query Expansion and Retrieval Strategy
API and UI Design
Clinical Decision Support Engines
Reporting and Interoperability
Product Requirements Document
Data Acquisition Strategy
Validation and Testing Strategy
Regulatory Considerations
DGX Compute Progression
Implementation Roadmap
Risk Analysis
Competitive Landscape
Discussion
Conclusion
References

1. Introduction¶

1.1 The Neurological Disease Burden¶

Neurological disorders constitute an unprecedented global health challenge. According to the Global Burden of Disease (GBD) 2021 study and the WHO's 2024 update on neurological conditions:

3.4 billion people (43% of the global population) are affected by neurological conditions
443 million DALYs lost annually to neurological disorders
11 million deaths per year directly attributable to neurological disease (second leading cause of death after cardiovascular disease)
Alzheimer's disease and other dementias affect 55 million people globally, projected to reach 139 million by 2050
Stroke kills 6.6 million people annually and is the leading cause of acquired disability in adults
Epilepsy affects 50 million people worldwide, with 80% in low- and middle-income countries
Parkinson's disease prevalence has doubled in the past 25 years to 8.5 million, the fastest-growing neurological disorder
Multiple sclerosis affects 2.8 million people globally, with incidence rising in every world region
Migraine is the second leading cause of years lived with disability globally, affecting 1.1 billion people
Brain tumors account for 308,000 new diagnoses annually, with glioblastoma carrying a median survival of 15 months

The economic burden is equally staggering. In the United States alone, neurological diseases cost an estimated $789 billion annually in direct medical costs and lost productivity. Alzheimer's disease alone accounts for $345 billion in annual costs, projected to exceed $1 trillion by 2050. Globally, the economic impact of neurological disorders exceeds $2.5 trillion per year.

1.2 The Crisis in Neurological Expertise¶

Unlike cardiology, where advanced imaging and intervention are widely available, neurology faces a severe workforce shortage that compounds the data challenge:

The United States has approximately 19,500 practicing neurologists for a population of 340 million -- a ratio of 1:17,400
The AAN projects a shortfall of 11,000 neurologists by 2031
Average wait time for a new neurology appointment is 4-8 weeks, reaching 6+ months in rural areas
Only ~200 Level 4 epilepsy centers exist in the US, leaving most patients without access to specialized evaluation
Subspecialty expertise (neuro-oncology, movement disorders, neurogenetics) is concentrated in fewer than 100 academic medical centers

This shortage makes AI-assisted clinical decision support not merely convenient but essential. A community neurologist managing 2,500+ patients cannot maintain current knowledge across the breadth of neurology -- from stroke protocols to epilepsy genetics to immunotherapy for MS. An intelligence agent that synthesizes evidence across the full neurological spectrum can function as an always-available subspecialty consultant.

1.3 The Opportunity for Integrated Neurological Intelligence¶

The neurological domain presents unique characteristics that make it exceptionally well-suited for an integrated AI intelligence agent:

Multi-modal data convergence: Neurology integrates structural imaging (MRI, CT), functional imaging (fMRI, PET, SPECT), electrophysiology (EEG, EMG/NCS, evoked potentials), cognitive/behavioral assessment (neuropsychological testing, rating scales), genomics (neurogenetic panels), cerebrospinal fluid analysis, and biomarkers -- all of which must be synthesized for accurate diagnosis and treatment.
Pattern recognition dependency: Many neurological diagnoses depend on recognizing complex spatiotemporal patterns -- EEG seizure localization, white matter lesion distribution in MS, dopaminergic deficit patterns in Parkinson's, brain tumor morphology and enhancement characteristics -- that are well-suited for AI augmentation.
Longitudinal disease tracking: Neurodegenerative diseases unfold over years to decades. Tracking hippocampal volume loss, lesion burden evolution, motor score progression, and cognitive decline trajectories requires systematic longitudinal analysis that AI excels at.
Genomic revolution: Neurogenetics has transformed from an academic curiosity to a clinical imperative. Over 1,000 genes are now associated with neurological disease, with genetic diagnosis changing management in epilepsy (SCN1A/Dravet), movement disorders (GBA/Parkinson's), dementia (APOE, PSEN1/2, MAPT), neuromuscular disease (SMN1/SMA, DMD), and brain tumors (IDH1/2, 1p/19q, MGMT).
Therapeutic pipeline explosion: The neuroscience drug pipeline has expanded dramatically with anti-amyloid antibodies (lecanemab, donanemab), antisense oligonucleotides (nusinersen, tofersen), gene therapies (onasemnogene for SMA, atidarsagene for MLD), and emerging CRISPR-based approaches -- creating urgent need for precision treatment selection.
Strong cross-modal triggers: A neuroimaging finding (e.g., hippocampal atrophy pattern suggestive of genetic frontotemporal dementia) frequently triggers genomic workup (FTD gene panel: MAPT, GRN, C9orf72), which may guide therapy selection and family counseling.

1.4 Our Contribution¶

This paper presents the complete architectural blueprint and product requirements for the Neurology Intelligence Agent, the seventh domain-specific intelligence agent in the HCLS AI Factory platform. Our contributions include:

A 13-collection Milvus vector schema designed for the full spectrum of neurological data: neuroimaging, electrophysiology, neurodegenerative disease, cerebrovascular disease, epilepsy, neuro-oncology, multiple sclerosis, movement disorders, headache, neuromuscular disease, neurogenetics, clinical trials, and literature
Eight reference clinical workflows covering acute stroke triage, dementia evaluation, epilepsy focus localization, brain tumor grading, MS monitoring, Parkinson's assessment, headache classification, and neuromuscular evaluation
A neurology knowledge graph with structured data on 40+ neurological conditions, 20+ neuroimaging protocols, 30+ validated clinical scales, 25+ drug classes, and 50+ AAN/EAN guideline recommendations
Cross-modal triggers linking neuroimaging findings to genomic workup (epilepsy gene panels, dementia gene panels, movement disorder genetics, neuromuscular genetics) via the shared genomic_evidence collection
Clinical decision support engines implementing validated neurological scales (NIHSS, GCS, MoCA, UPDRS, EDSS, mRS, HIT-6) and diagnostic algorithms
A comprehensive product requirements document with user stories, acceptance criteria, and implementation prioritization
Deployment on a single NVIDIA DGX Spark ($3,999), maintaining the platform's commitment to accessible AI

2. The Neurological Data Challenge¶

2.1 Data Fragmentation in Clinical Neuroscience¶

Clinical neuroscience generates data across at least sixteen distinct categories, each with its own structure, vocabulary, source systems, and interpretive frameworks:

Structural Neuroimaging -- Brain MRI sequences (T1, T2, FLAIR, T2*, SWI, post-contrast T1), spinal MRI, CT head (non-contrast, CTA, CTP), measurements (hippocampal volume, cortical thickness, ventricular size, white matter hyperintensity volume, lesion counts and volumes, brain parenchymal fraction).
Functional Neuroimaging -- fMRI (task-based and resting-state connectivity), PET (FDG for metabolism, amyloid PET with florbetapir/flutemetamol/florbetaben, tau PET with flortaucipir, dopamine transporter DaT-SPECT), SPECT (ictal/interictal perfusion for epilepsy).
Diffusion Imaging -- DTI (fractional anisotropy, mean diffusivity, axial/radial diffusivity), tractography (corticospinal tract, arcuate fasciculus, optic radiations), DWI for acute stroke (ADC maps, mismatch assessment), connectomics.
Electrophysiology -- EEG -- Scalp EEG (routine, ambulatory, continuous ICU monitoring, long-term video-EEG monitoring), intracranial EEG (subdural grids, stereo-EEG depth electrodes), quantitative EEG, event-related potentials, spectral analysis, source localization (dipole modeling, LORETA), seizure detection and classification.
Electrophysiology -- EMG/NCS -- Nerve conduction studies (motor, sensory, F-waves, H-reflexes), needle electromyography (insertional activity, spontaneous activity, MUAP analysis, recruitment), repetitive nerve stimulation (NMJ disorders), single-fiber EMG.
Neurodegenerative Disease Data -- Cognitive assessments (MoCA, MMSE, CDR, ADAS-Cog), motor scales (UPDRS, Hoehn & Yahr, ALSFRS-R), biomarkers (CSF Abeta-42, p-tau-181, neurofilament light chain, alpha-synuclein seed amplification), amyloid/tau PET status, genetic risk factors (APOE, LRRK2, GBA, C9orf72, SOD1).
Cerebrovascular Disease Data -- Stroke scales (NIHSS, mRS, ASPECTS), vascular imaging (CTA, MRA, digital subtraction angiography), perfusion imaging (CTP, MR perfusion), hemorrhage grading (ICH Score, Fisher Scale, Hunt-Hess), thrombolysis eligibility criteria, thrombectomy selection (DAWN, DEFUSE-3 criteria), secondary prevention protocols.
Epilepsy Data -- Seizure classification (ILAE 2017), epilepsy syndrome diagnosis, EEG findings (interictal epileptiform discharges, ictal patterns, HFOs), MRI findings (hippocampal sclerosis, cortical dysplasia, tumors, vascular malformations), ASM efficacy and side effects, surgical evaluation data (Wada testing, neuropsychological lateralization, concordance matrix), stimulation device data (VNS, RNS, DBS parameters).
Neuro-Oncology Data -- WHO 2021 CNS tumor classification (molecular-integrated), grading (Grade 1-4), molecular markers (IDH1/2, 1p/19q codeletion, MGMT promoter methylation, H3K27M, TERT, ATRX, BRAF, CDKN2A), treatment protocols (Stupp protocol, CCNU/lomustine, bevacizumab, tumor treating fields), RANO response criteria, survival data.
Multiple Sclerosis Data -- McDonald criteria (2017), MS phenotype (CIS, RRMS, SPMS, PPMS), EDSS scoring, MRI lesion metrics (new T2, enhancing, PRL, cortical lesions, spinal cord lesions, brain atrophy), OCT (RNFL thickness, ganglion cell layer), CSF (oligoclonal bands, IgG index, kappa free light chains), DMT efficacy data (platform therapies through BTK inhibitors), NEDA-3 status.
Movement Disorders Data -- Parkinson's disease (UPDRS parts I-IV, Hoehn & Yahr, dopaminergic medication LEDDs, DBS programming, DaT-SPECT), essential tremor, dystonia (Burke-Fahn-Marsden), Huntington's disease (UHDRS, CAG repeat length, prodromal markers), ataxia (SARA, BARS), functional movement disorders.
Headache Medicine -- ICHD-3 classification, migraine subtypes (with/without aura, chronic, vestibular, hemiplegic), medication overuse headache, cluster headache, trigeminal autonomic cephalalgias, secondary headache red flags, preventive efficacy data (CGRP mAbs, gepants, neuromodulation), disability scales (HIT-6, MIDAS).
Neuromuscular Disease Data -- NCS/EMG patterns (demyelinating vs axonal, generalized vs focal), myasthenia gravis (AChR/MuSK antibodies, QMG score, MGFA classification), GBS (Hughes scale, treatment criteria), CIDP (EFNS/PNS criteria), muscular dystrophies (CK levels, genetic confirmation, functional scales), motor neuron disease (El Escorial criteria, ALSFRS-R, FVC).
Neurogenetics Data -- Mendelian neurological disorders (>1,000 genes), pharmacogenomics (HLA-B*15:02 for carbamazepine, CYP2C19 for clopidogrel), polygenic risk scores (Alzheimer's, Parkinson's, epilepsy), variant interpretation (ACMG criteria adapted for neurological variants), genetic counseling considerations, gene therapy eligibility.
Clinical Trials -- ClinicalTrials.gov neurology entries (12,000+ active/completed), landmark trial results (EMERGE/CLARITY for lecanemab, TRAILBLAZER for donanemab, DAWN/DEFUSE-3 for thrombectomy, ARISE for antisense oligonucleotides), outcome measures, biomarker endpoints.
Neurorehabilitation Data -- Functional assessments (FIM, Barthel Index, mRS at 90 days), rehabilitation intensity and protocols, plasticity biomarkers, assistive technology, cognitive rehabilitation outcomes.

2.2 Why Existing Tools Fall Short¶

Approach	Limitation
PubMed search	Keyword-based; misses semantic connections across neurology subspecialties; no imaging or electrophysiology data integration
UpToDate / DynaMed	Expert-curated but static; no patient-specific reasoning; no imaging AI; no genomic integration; subscription-based
PACS with AI (Aidoc, Viz.ai)	Single-modality (CT for stroke); no EEG, EMG, genomics, or longitudinal tracking; cloud-dependent
EEG platforms (Persyst, BESA)	Electrophysiology-only; no imaging correlation; no literature synthesis; expensive licensing
Genetic platforms (Invitae, GeneDx)	Genetics-only; no imaging or electrophysiology correlation; report delivery, not decision support
General AI assistants	No citation provenance; hallucination risk in high-stakes neurological decisions; no structured data; not guideline-aligned

2.3 The Case for Multi-Collection RAG in Neurology¶

A neurologist evaluating a patient with new-onset seizures must simultaneously consider:

Structural imaging: MRI showing hippocampal sclerosis, cortical dysplasia, or tumor
Electrophysiology: EEG showing temporal intermittent rhythmic delta activity or epileptiform discharges
Functional imaging: PET showing temporal hypometabolism concordant with EEG focus
Genomics: If family history or drug-resistant epilepsy, genetic panel (SCN1A, KCNQ2, CDKL5, TSC1/2)
Guidelines: ILAE seizure classification, AAN/AES treatment guidelines for first seizure
Trials: Evidence for specific ASMs by seizure type and epilepsy syndrome
Clinical scales: Seizure frequency, side effect burden, quality of life measures

No existing tool synthesizes all seven dimensions. A multi-collection RAG architecture enables parallel retrieval across all dimensions with a single query, followed by LLM synthesis into coherent clinical guidance.

3. Clinical Landscape and Market Analysis¶

3.1 Neurology AI Market¶

Metric	Value	Source
Global neuro AI market (2024)	$1.7 billion	Markets and Markets
Projected market (2029)	$4.9 billion	Markets and Markets
CAGR (2024-2029)	23.7%	Markets and Markets
FDA-cleared neuro AI devices (cumulative)	120+	FDA AI/ML database
Active neuro AI clinical trials	380+	ClinicalTrials.gov
Annual neuro AI publications	3,800+	PubMed (2025)
US neurologists	~19,500	AAN Census
Global neurologists	~85,000	WHO estimates
US epilepsy centers (Level 4)	~200	NAEC
US memory disorder clinics	~500	Alzheimer's Association
Annual US neurological costs	$789 billion	AAN Economic Burden Study
Global Alzheimer's drug pipeline	140+ candidates	Alzheimer's Drug Discovery Foundation

3.2 Competitive Analysis¶

Competitor	Strengths	Gaps
Viz.ai (LVO stroke)	FDA-cleared, real-time CT triage, hospital networks	Single use case (stroke), no EEG/EMG, no genomics, no outpatient neuro, SaaS pricing
Aidoc (CT head)	Multi-pathology CT triage (hemorrhage, PE, C-spine)	Imaging-only, no longitudinal tracking, no treatment guidance
Brainomix (e-ASPECTS, CTA)	Stroke imaging AI, European market presence	Stroke-only, no other neurological conditions
Persyst (EEG)	Automated seizure detection, spike detection	EEG-only, no imaging correlation, no genomics, expensive licensing ($30K+)
Natus/Nihon Kohden (EEG)	Comprehensive EEG platforms, ICU monitoring	Hardware-dependent, no AI-driven interpretation, no cross-modal
Biogen Digital Health	MS monitoring, Parkinson's wearables	Disease-specific, not comprehensive, pharma-biased
QMENTA (Neuroimaging platform)	Brain volumetrics, cloud neuroimaging analysis	Research-focused, no clinical workflow integration, cloud-only
Tempus (Neuro-oncology)	Molecular profiling, clinical data	Proprietary, expensive, limited to neuro-oncology

Our differentiation: The Neurology Intelligence Agent will be the only system combining (1) multi-modal neuroimaging AI, (2) EEG/EMG integration, (3) genomic cross-modal triggers, (4) literature RAG with citations, (5) validated neurological scales, (6) guideline-aligned decision support, and (7) on-device deployment -- all open-source at $3,999.

3.3 Target Users¶

User Segment	Use Case	Pain Point Addressed
Community neurologists	Subspecialty-level decision support	4-8 week wait for subspecialty referral
Academic neurology	Research and education	Fragmented data across systems
Epilepsy centers	Pre-surgical evaluation, surgical candidacy	Complex multi-modal concordance analysis
Memory clinics	Dementia differential diagnosis	Distinguishing AD from FTD, DLB, vascular
Stroke centers	Acute triage and secondary prevention	Time-critical decision support
Neuro-oncology programs	Molecular-integrated tumor classification	Rapid molecular result integration
MS centers	Treatment escalation decisions	NEDA monitoring, DMT selection
Movement disorder clinics	Parkinson's vs atypical parkinsonism	Complex differential diagnosis
Clinical trial sites	Patient screening, biomarker endpoints	Manual eligibility assessment
Neurogenetics clinics	Variant interpretation in neurological context	Limited neuro-specific annotation

4. Existing HCLS AI Factory Architecture¶

4.1 Platform Overview¶

The HCLS AI Factory is a three-stage precision medicine platform running on NVIDIA DGX Spark:

Stage 1: Genomics Pipeline (Parabricks + DeepVariant)
    FASTQ -> VCF -> 3.56M annotated variants
         |
Stage 2: RAG/Chat Pipeline (Milvus + Claude)
    Variant interpretation, clinical significance
         |
Stage 3: Drug Discovery Pipeline (BioNeMo + DiffDock)
    Target -> Lead compound -> Docking -> Drug-likeness

Six existing intelligence agents extend this platform:

Agent	Collections	Seed Vectors	Unique Capability
Precision Biomarker	11	6,134	Biological age calculators, biomarker panels
Precision Oncology	10	609	Molecular tumor board packets, trial matching
CAR-T Intelligence	11	6,266	CAR construct comparison, manufacturing optimization
Imaging Intelligence	10	876	NIM inference, DICOM workflows, 3D segmentation
Autoimmune Intelligence	10	~500	Autoantibody panels, flare prediction
Cardiology (planned)	12	~1,530	Risk calculators, GDMT optimization

4.2 Shared Infrastructure¶

All agents share:

Milvus 2.4 vector database (IVF_FLAT, COSINE, 384-dim)
BGE-small-en-v1.5 embedding model (sentence-transformers)
Claude Sonnet 4.6 (Anthropic) primary LLM
genomic_evidence collection (3,561,170 variants, read-only)
Docker Compose orchestration
FastAPI (REST) + Streamlit (UI) pattern
lib/hcls_common shared library (23 modules)

4.3 Proven Patterns Adapted for Neurology¶

Pattern	Proven In	Adaptation for Neurology
Multi-collection parallel search	All agents	13 neurology-specific collections
Knowledge graph augmentation	CAR-T, Biomarker, Cardiology	Neurological conditions, drug classes, genetic disorders
Query expansion maps	CAR-T (12 maps), Cardiology (15 maps)	Neurology terminology (e.g., "seizure" -> epilepsy, convulsion, ictal, epileptiform, paroxysmal)
Comparative analysis	CAR-T, Imaging	"Lecanemab vs Donanemab", "Carbamazepine vs Lamotrigine"
Cross-modal genomic triggers	Imaging (Lung-RADS -> EGFR), Cardiology (DCM -> TTN)	Neuroimaging -> epilepsy/dementia gene panels
FHIR R4 export	Imaging, Cardiology	DiagnosticReport with neuro SNOMED/LOINC codes
NIM inference workflows	Imaging (4 NIMs)	Brain segmentation, volumetrics, lesion tracking
Validated clinical scales	Cardiology (ASCVD, CHA2DS2-VASc)	NIHSS, GCS, MoCA, UPDRS, EDSS
Sidebar guided tour	Imaging, Cardiology	Neurology demo flow

5. Neurology Intelligence Agent Architecture¶

5.1 System Diagram¶

+==========================================================================+
|  PRESENTATION:  Streamlit Neurology Workbench (8529)                     |
|                 10 Tabs | Evidence | Workflows | Imaging | Scales        |
|                 FastAPI REST Server (8528)                                |
+==========================================================================+
                    |                            |
+==========================================================================+
|  INTELLIGENCE:   Neurology RAG Engine                                    |
|                  13-collection parallel search                           |
|                  Knowledge graph (40 conditions, 30 scales, 50 genes)    |
|                  Query expansion (16 maps, neuroscience terminology)     |
|                  Comparative analysis ("Lecanemab vs Donanemab")         |
|                  Clinical scale calculators (NIHSS, GCS, MoCA, etc.)    |
|                  Diagnostic algorithm engine                             |
+==========================================================================+
                    |                            |
+==========================================================================+
|  INFERENCE:      NIM Services (VISTA-3D, MAISI, VILA-M3, Llama-3)       |
|                  8 Clinical Workflows:                                    |
|                  Stroke | Dementia | Epilepsy | Neuro-Onc | MS |        |
|                  Parkinson's | Headache | Neuromuscular                   |
+==========================================================================+
                    |                            |
+==========================================================================+
|  DATA:           Milvus 2.4 (13 neuro collections + genomic)            |
|                  BGE-small-en-v1.5 (384-dim, IVF_FLAT, COSINE)          |
|                  PubMed, ClinicalTrials.gov, AAN/EAN guidelines          |
|                  Curated seed data (imaging, EEG, scales, genetics)      |
+==========================================================================+

5.2 Design Principles¶

Diagnostic rigor: Neurological diagnoses carry life-altering implications (epilepsy, dementia, brain tumor). Every output includes differential diagnosis, red flags, and confidence qualifiers.
Longitudinal awareness: Neurodegenerative diseases require tracking over months to years. Collections and workflows support temporal comparisons (baseline vs follow-up volumetrics, lesion evolution, scale score trajectories).
Multi-modal concordance: Epilepsy and dementia evaluation require concordance across imaging, electrophysiology, and clinical data. The architecture supports concordance matrices.
Genomic integration by default: Every significant neurological finding is checked against the neurogenetics knowledge base for actionable genetic associations.
Graceful degradation: Full functionality in mock mode without GPU or live NIM services.
Pattern consistency: Follows the same FastAPI + Streamlit + Milvus patterns as all other HCLS AI Factory agents.

5.3 Port Allocation¶

Port	Service
8528	FastAPI REST Server
8529	Streamlit Neurology Workbench
19530	Milvus (shared)
8520	NIM LLM (shared)
8530	NIM VISTA-3D (shared)
8531	NIM MAISI (shared)
8532	NIM VILA-M3 (shared)

6. Milvus Collection Design¶

6.1 Index Configuration¶

Parameter	Value
Index type	IVF_FLAT
Metric	COSINE
nlist / nprobe	1024 / 16
Dimension	384
Embedding model	BAAI/bge-small-en-v1.5

6.2 Collection Schemas¶

Collection 1: `neuro_literature` -- ~3,500 records¶

Published neuroscience and neurology research papers, reviews, and meta-analyses.

Field	Type	Description
id	VARCHAR(64)	PubMed ID or unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
title	VARCHAR(500)	Paper title
text_chunk	VARCHAR(8000)	Abstract or text section
year	INT16	Publication year
journal	VARCHAR(200)	Journal name (Neurology, Ann Neurol, Brain, Lancet Neurol, JAMA Neurol, Epilepsia)
neuro_domain	VARCHAR(100)	Neurological subdomain
modality	VARCHAR(50)	Imaging or diagnostic modality if applicable
study_type	VARCHAR(50)	RCT, meta-analysis, cohort, case-control, review
keywords	VARCHAR(500)	MeSH terms and author keywords

Source: PubMed E-utilities with neurology/neuroscience MeSH filters.

Collection 2: `neuro_trials` -- ~600 records¶

Neurology clinical trials from ClinicalTrials.gov and landmark trial results.

Field	Type	Description
id	VARCHAR(64)	NCT number or trial identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
title	VARCHAR(500)	Official trial title
text_summary	VARCHAR(4000)	Trial summary including results
phase	VARCHAR(20)	Phase I-IV
status	VARCHAR(30)	Active, completed, recruiting
sponsor	VARCHAR(200)	Lead sponsor
neuro_domain	VARCHAR(100)	Stroke, epilepsy, dementia, MS, PD, neuro-onc, headache
intervention	VARCHAR(300)	Drug, device, or procedure tested
primary_endpoint	VARCHAR(300)	Primary outcome measure
enrollment	INT32	Number of participants
start_year	INT16	Year trial began
outcome_summary	VARCHAR(2000)	Key results (if completed)
landmark	BOOL	Is this a landmark trial

Source: ClinicalTrials.gov V2 API with neurological condition filters.

Collection 3: `neuro_imaging` -- ~250 records¶

Neuroimaging protocols, findings, and measurements across all modalities.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Finding or protocol description
modality	VARCHAR(50)	mri, ct, pet, spect, dti, fmri
sequence_type	VARCHAR(100)	T1, T2, FLAIR, SWI, DWI, ADC, T1_post, DTI_FA
finding_category	VARCHAR(100)	Atrophy, lesion, enhancement, hemorrhage, infarction, mass
brain_region	VARCHAR(100)	Temporal, frontal, parietal, occipital, cerebellum, brainstem, spinal_cord
measurement_name	VARCHAR(100)	Hippocampal_volume, cortical_thickness, WMH_volume, lesion_count
measurement_value	VARCHAR(50)	Numeric value with units
reference_range	VARCHAR(100)	Normal range per age/sex norms
clinical_significance	VARCHAR(500)	Interpretation guidance
differential_diagnosis	VARCHAR(500)	What this finding suggests

Source: Curated from AAN/ASNR guidelines, neuroradiology references, and brain atlas data.

Collection 4: `neuro_electrophysiology` -- ~200 records¶

EEG patterns, EMG/NCS findings, and electrophysiological data.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	EEG or EMG pattern description
test_type	VARCHAR(50)	EEG, EMG, NCS, SSEP, VEP, BAEP
pattern_name	VARCHAR(200)	Specific pattern (e.g., "3-Hz generalized spike-and-wave")
lateralization	VARCHAR(50)	Left, right, bilateral, generalized, multifocal
brain_region	VARCHAR(100)	Temporal, frontal, parietal, occipital, generalized
clinical_correlation	VARCHAR(500)	Associated conditions
urgency	VARCHAR(20)	Routine, urgent, emergent
classification	VARCHAR(100)	ACNS standardized terminology
differential_diagnosis	VARCHAR(500)	DDx list

Source: Curated from ACNS guidelines, ABEM standards, and EP references.

Collection 5: `neuro_degenerative` -- ~200 records¶

Neurodegenerative disease management: Alzheimer's, Parkinson's, ALS, FTD, Huntington's.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Disease management recommendation
disease	VARCHAR(100)	Alzheimer's, Parkinson's, ALS, FTD, Huntington's, DLB, MSA, PSP, CBD
disease_stage	VARCHAR(50)	Preclinical, prodromal, mild, moderate, severe
biomarker	VARCHAR(100)	Amyloid_PET, tau_PET, NfL, CSF_Abeta42, DaT_SPECT
biomarker_status	VARCHAR(50)	Positive, negative, borderline
drug_class	VARCHAR(100)	Cholinesterase inhibitor, anti-amyloid, dopaminergic, riluzole
drug_name	VARCHAR(100)	Specific medication
clinical_scale	VARCHAR(100)	MoCA, CDR, UPDRS, ALSFRS-R, UHDRS
evidence_level	VARCHAR(20)	Level of evidence
guideline_source	VARCHAR(100)	AAN, NIA-AA, MDS, EFNS
genetic_association	VARCHAR(200)	Associated genes

Source: AAN practice guidelines, NIA-AA criteria, MDS criteria.

Collection 6: `neuro_cerebrovascular` -- ~180 records¶

Stroke, cerebrovascular disease, and neurovascular conditions.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Stroke management or vascular neurology recommendation
stroke_type	VARCHAR(50)	Ischemic, hemorrhagic, SAH, TIA, CVT
vascular_territory	VARCHAR(100)	MCA, ACA, PCA, basilar, PICA, AChA
imaging_finding	VARCHAR(200)	DWI restriction, CTA occlusion, perfusion mismatch, hemorrhage
scale_name	VARCHAR(50)	NIHSS, ASPECTS, ICH_Score, Fisher, Hunt_Hess, mRS
scale_value	VARCHAR(50)	Score value and interpretation
treatment	VARCHAR(300)	tPA, thrombectomy, EVD, surgical evacuation
time_window	VARCHAR(100)	Eligibility time window
eligibility_criteria	VARCHAR(500)	Treatment selection criteria (DAWN, DEFUSE-3)
secondary_prevention	VARCHAR(500)	Long-term prevention strategy

Source: AHA/ASA stroke guidelines, AAN practice parameters.

Collection 7: `neuro_epilepsy` -- ~200 records¶

Epilepsy classification, treatment, and surgical evaluation data.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Epilepsy management recommendation
seizure_type	VARCHAR(100)	Focal_aware, focal_impaired, focal_to_bilateral, generalized_tonic_clonic, absence, myoclonic
epilepsy_syndrome	VARCHAR(200)	JME, CAE, Dravet, West, Lennox-Gastaut, TLE-HS, BECTS
eeg_finding	VARCHAR(300)	Interictal and ictal EEG patterns
mri_finding	VARCHAR(300)	Hippocampal sclerosis, FCD, tumor, vascular malformation
asm_name	VARCHAR(100)	Anti-seizure medication name
asm_class	VARCHAR(100)	Na+ channel, SV2A, GABA, broad-spectrum
asm_first_line	BOOL	Is this first-line for this seizure type
surgical_candidate	VARCHAR(50)	Yes, no, pending_evaluation
genetic_etiology	VARCHAR(200)	SCN1A, KCNQ2, CDKL5, TSC1/2, SLC2A1
evidence_level	VARCHAR(20)	Level of evidence

Source: ILAE classification, AAN/AES practice guidelines, epilepsy genetics databases.

Collection 8: `neuro_oncology` -- ~150 records¶

Brain tumor classification, treatment, and molecular data.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Neuro-oncology recommendation
tumor_type	VARCHAR(200)	WHO 2021 integrated diagnosis
who_grade	VARCHAR(10)	1, 2, 3, 4
molecular_markers	VARCHAR(500)	IDH1/2, 1p/19q, MGMT, H3K27M, TERT, ATRX, BRAF
location	VARCHAR(100)	Supratentorial, infratentorial, spinal, sellar
imaging_characteristics	VARCHAR(500)	Enhancement pattern, diffusion, perfusion, spectroscopy
treatment_protocol	VARCHAR(500)	Surgery, radiation, chemotherapy regimen
response_criteria	VARCHAR(200)	RANO criteria, iRANO for immunotherapy
prognosis	VARCHAR(300)	Expected survival, prognostic factors

Source: WHO 2021 CNS tumor classification, NCCN guidelines, RANO working group.

Collection 9: `neuro_ms` -- ~180 records¶

Multiple sclerosis diagnosis, monitoring, and treatment.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	MS management recommendation
ms_phenotype	VARCHAR(30)	CIS, RRMS, SPMS, PPMS
diagnostic_criteria	VARCHAR(300)	McDonald 2017 criteria met
mri_metric	VARCHAR(100)	New_T2, enhancing, PRL, brain_atrophy, spinal_lesions
mri_value	VARCHAR(50)	Numeric measurement
edss_score	FLOAT	Expanded Disability Status Scale (0-10)
dmt_name	VARCHAR(100)	Disease-modifying therapy name
dmt_category	VARCHAR(50)	Platform, moderate_efficacy, high_efficacy
neda_status	VARCHAR(20)	NEDA-3 achieved, not achieved
escalation_criteria	VARCHAR(500)	When to escalate therapy
monitoring_protocol	VARCHAR(500)	MRI frequency, JCV testing, labs

Source: AAN MS practice guidelines, ECTRIMS/EAN guidelines, McDonald 2017.

Collection 10: `neuro_movement` -- ~150 records¶

Movement disorders: Parkinson's disease, essential tremor, dystonia, ataxia, Huntington's.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Movement disorder recommendation
disorder	VARCHAR(100)	Parkinson's, essential_tremor, dystonia, Huntington's, ataxia, MSA, PSP, CBD, DLB
clinical_feature	VARCHAR(200)	Specific motor or non-motor feature
scale_name	VARCHAR(100)	UPDRS, H&Y, UHDRS, SARA, Burke_Fahn_Marsden
scale_value	VARCHAR(50)	Score and interpretation
medication	VARCHAR(100)	Drug name
medication_ledd	FLOAT	Levodopa equivalent daily dose (mg)
surgical_option	VARCHAR(200)	DBS target (STN, GPi, VIM), FUS, LITT
genetic_association	VARCHAR(200)	LRRK2, GBA, SNCA, PARK2, PINK1, DJ-1
dat_spect_result	VARCHAR(100)	Normal, reduced (pattern description)

Source: MDS clinical criteria, AAN treatment guidelines.

Collection 11: `neuro_headache` -- ~120 records¶

Headache classification, treatment, and prevention.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Headache management recommendation
headache_type	VARCHAR(100)	Migraine_without_aura, migraine_with_aura, chronic_migraine, cluster, TTH, MOH
ichd3_code	VARCHAR(20)	ICHD-3 classification code
red_flags	VARCHAR(500)	SNOOP mnemonic: Systemic, Neurologic, Onset, Older, Previous_hx
acute_treatment	VARCHAR(300)	Triptans, NSAIDs, gepants, ditans, ergots
preventive_treatment	VARCHAR(300)	CGRP mAbs, topiramate, amitriptyline, propranolol, OnabotulinumtoxinA
neuromodulation	VARCHAR(200)	sTMS, eTNS, nVNS, REN
disability_score	VARCHAR(50)	HIT-6, MIDAS score and grade
frequency	VARCHAR(50)	Monthly headache days

Source: ICHD-3 classification, AAN/AHS practice guidelines.

Collection 12: `neuro_neuromuscular` -- ~130 records¶

Neuromuscular disease: neuropathy, myopathy, NMJ disorders, motor neuron disease.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Neuromuscular disease recommendation
disease_category	VARCHAR(100)	Neuropathy, myopathy, NMJ_disorder, motor_neuron_disease
disease_name	VARCHAR(200)	GBS, CIDP, CMT, MG, ALS, SMA, DMD, myotonic_dystrophy
emg_ncs_pattern	VARCHAR(300)	Demyelinating, axonal, mixed, myopathic, NMJ_decrement
antibody	VARCHAR(100)	AChR, MuSK, anti-MAG, anti-GM1, anti-GQ1b
ck_level	VARCHAR(50)	Normal, elevated (range)
genetic_test	VARCHAR(200)	SMN1, DMD, PMP22, MFN2, TTR
treatment	VARCHAR(300)	IVIg, PLEX, rituximab, nusinersen, gene_therapy
functional_scale	VARCHAR(100)	ALSFRS-R, QMG, INCAT, CMTNS, FVC
prognosis	VARCHAR(300)	Expected course and milestones

Source: AAN practice parameters, EFNS/PNS guidelines.

Collection 13: `neuro_guidelines` -- ~200 records¶

AAN, EAN, and subspecialty clinical practice guidelines and practice parameters.

Field	Type	Description
id	VARCHAR(64)	Unique identifier
embedding	FLOAT_VECTOR(384)	BGE-small-en-v1.5 embedding
text_summary	VARCHAR(4000)	Guideline recommendation
guideline_name	VARCHAR(300)	Full guideline title
organization	VARCHAR(50)	AAN, EAN, AES, ILAE, MDS, AHA/ASA, NCCN
year	INT16	Publication year
neuro_domain	VARCHAR(100)	Stroke, epilepsy, MS, dementia, movement, headache, NMD
recommendation_level	VARCHAR(30)	Level A/B/C/U (AAN) or Class I/IIa/IIb/III
key_recommendation	VARCHAR(2000)	Specific recommendation text
clinical_scenario	VARCHAR(500)	When this recommendation applies

Source: AAN Practice Guideline library, EAN guidelines, ILAE/AES guidelines.

Collection 14: `genomic_evidence` -- 3,561,170 records (read-only)¶

Shared genomic variant collection from HCLS AI Factory Stage 1+2.

Purpose: Cross-modal triggers query this collection for neurologically relevant genes (APOE, PSEN1/2, APP, MAPT, GRN, C9orf72, SCN1A, LRRK2, GBA, SMN1, DMD, HTT, etc.).

6.3 Collection Search Weights¶

Collection	Weight	Rationale
Literature	0.14	Largest corpus, broadest evidence base
Guidelines	0.12	Highest clinical authority
Trials	0.10	Primary evidence source
Neuroimaging	0.10	Central to neurological diagnosis
Neurodegenerative	0.08	High-impact chronic diseases
Cerebrovascular	0.08	Acute high-acuity conditions
Epilepsy	0.07	Complex diagnostic workup
Neuro-Oncology	0.06	Molecular-integrated diagnosis
MS	0.06	Growing therapeutic complexity
Movement Disorders	0.05	Subspecialty differential diagnosis
Electrophysiology	0.05	Diagnostic correlate
Headache	0.04	High-volume clinical need
Neuromuscular	0.04	Subspecialty domain
Genomic Evidence	0.01	Supplementary variant context

6.4 Estimated Vector Counts¶

Collection	Seed Records	Post-Ingest Target
neuro_literature	250	3,500+ (PubMed ingest)
neuro_trials	60	600+ (ClinicalTrials.gov)
neuro_imaging	250	250
neuro_electrophysiology	200	200
neuro_degenerative	200	200
neuro_cerebrovascular	180	180
neuro_epilepsy	200	200
neuro_oncology	150	150
neuro_ms	180	180
neuro_movement	150	150
neuro_headache	120	120
neuro_neuromuscular	130	130
neuro_guidelines	200	200
Total (owned)	~2,270	~6,060+
genomic_evidence (read-only)	--	3,561,170

7. Clinical Workflows¶

7.1 Workflow Architecture¶

All workflows follow the established BaseImagingWorkflow pattern: preprocess -> infer -> postprocess -> WorkflowResult. Each workflow supports full mock mode with clinically realistic synthetic results.

7.2 Eight Reference Workflows¶

Workflow 1: Acute Stroke Triage¶

Attribute	Value
Workflow ID	`acute_stroke_triage`
Input	CT head (non-contrast), CTA, CTP, clinical data
Target Latency	< 90 seconds
Models	VISTA-3D (hemorrhage segmentation, infarct core), perfusion analysis
Key Outputs	Stroke type (ischemic vs hemorrhagic), ASPECTS score, LVO detection, perfusion mismatch ratio, NIHSS estimate, tPA/thrombectomy eligibility
Severity Routing	LVO detected or NIHSS >= 6 -> Emergent stroke alert
Cross-Modal Trigger	Cryptogenic stroke + age < 50 -> thrombophilia/CADASIL panel (NOTCH3, COL4A1/2)
Guideline Alignment	AHA/ASA 2019 Acute Ischemic Stroke Guidelines, 2022 Update

Clinical Decision Logic:

Acute Ischemic Stroke Triage:

1. CT Non-Contrast:
   - Hemorrhage? -> YES -> ICH pathway (ICH Score, surgical evaluation)
   - ASPECTS score -> < 6: poor candidate for intervention
                    -> >= 6: proceed to vascular imaging

2. CTA:
   - LVO present? (ICA, M1, M2, basilar)
   -> YES + NIHSS >= 6 + ASPECTS >= 6:
      -> Last known well < 6 hours: Direct to thrombectomy
      -> Last known well 6-24 hours: Apply DAWN/DEFUSE-3 criteria
         DAWN: Clinical-core mismatch (NIHSS >= 10, core < 31 mL for age < 80)
         DEFUSE-3: Perfusion mismatch ratio >= 1.8, core < 70 mL
   -> NO: Medical management (tPA if < 4.5 hours, antiplatelet)

3. tPA Eligibility (< 4.5 hours from last known well):
   - Age >= 18
   - Measurable neurological deficit (NIHSS >= 4 typically)
   - No contraindications (recent surgery, active bleeding, INR > 1.7)
   - BP controllable to < 185/110

4. Hemorrhagic Stroke:
   ICH Score (0-6):
     GCS 3-4: +2, GCS 5-12: +1, GCS 13-15: +0
     ICH volume >= 30 mL: +1
     IVH present: +1
     Infratentorial: +1
     Age >= 80: +1
   30-day mortality: Score 0=0%, 1=13%, 2=26%, 3=72%, 4=97%, 5-6=100%

Workflow 2: Dementia Evaluation¶

Attribute	Value
Workflow ID	`dementia_evaluation`
Input	Brain MRI, cognitive testing scores, biomarkers
Target Latency	< 3 minutes
Models	VISTA-3D (hippocampal volumetry, cortical thickness, ventricular volume, WMH quantification)
Key Outputs	Atrophy pattern classification, hippocampal volume percentile, Fazekas WMH score, differential diagnosis (AD vs FTD vs DLB vs vascular), NIA-AA ATN classification, treatment eligibility
Severity Routing	Rapid progression or unusual pattern -> Urgent neuro referral
Cross-Modal Trigger	Pattern suggestive of genetic FTD -> FTD gene panel (MAPT, GRN, C9orf72); Early-onset AD (<65) -> AD gene panel (PSEN1, PSEN2, APP); DLB features -> GBA
Guideline Alignment	NIA-AA 2024 Revised Criteria, AAN Mild Cognitive Impairment Guidelines

Atrophy Pattern Recognition:

Pattern	Brain Region	Suggestive Diagnosis
Medial temporal atrophy (MTA)	Hippocampus, entorhinal cortex	Alzheimer's disease
Frontal > temporal atrophy	Frontal lobes, anterior temporal	Behavioral variant FTD
Asymmetric left temporal	Left temporal pole, fusiform	Semantic variant PPA
Asymmetric left perisylvian	Left inferior frontal, insula	Nonfluent variant PPA
Posterior cortical atrophy	Parieto-occipital	Posterior cortical atrophy (AD variant)
Midbrain atrophy ("hummingbird")	Midbrain tegmentum	PSP
Pontine/cerebellar atrophy ("hot cross bun")	Pons, cerebellum	MSA-C
Caudate atrophy	Caudate heads	Huntington's disease
Diffuse cortical + hippocampal	Global	DLB (less MTA than AD)

NIA-AA ATN Biomarker Framework:

A (Amyloid): Amyloid PET or CSF Abeta-42/40 ratio
  A+: Abnormal amyloid -> Alzheimer's continuum
  A-: Normal amyloid -> Non-Alzheimer's pathway

T (Tau): Tau PET or CSF p-tau-181/217
  T+: Abnormal tau -> AD pathologic change
  T-: Normal tau

N (Neurodegeneration): MRI atrophy, FDG-PET, CSF NfL, CSF total-tau
  N+: Evidence of neurodegeneration
  N-: No neurodegeneration

Classification:
  A-T-N-: Normal biomarkers
  A+T-N-: Preclinical Alzheimer's (asymptomatic)
  A+T+N-: Alzheimer's pathologic change
  A+T+N+: Alzheimer's disease
  A-T-N+: Non-AD neurodegeneration (suspected SNAP)
  A-T+N+: Non-AD tauopathy (consider FTLD-tau)

Workflow 3: Epilepsy Focus Localization¶

Attribute	Value
Workflow ID	`epilepsy_focus_localization`
Input	Brain MRI, EEG data, clinical seizure semiology, PET/SPECT if available
Target Latency	< 3 minutes
Models	VISTA-3D (hippocampal volumetry, lesion detection), volumetric comparison
Key Outputs	ILAE seizure classification, epilepsy syndrome identification, MRI-EEG concordance assessment, surgical candidacy evaluation, ASM recommendation by seizure type
Severity Routing	Status epilepticus or drug-resistant epilepsy -> Epilepsy center referral
Cross-Modal Trigger	Drug-resistant epilepsy + age < 25 -> epilepsy gene panel (SCN1A, KCNQ2, CDKL5, TSC1/2, DEPDC5, SLC2A1, PCDH19)
Guideline Alignment	ILAE 2017 Classification, AAN/AES 2018 First Seizure Guidelines

Concordance Matrix for Surgical Evaluation:

Data Source          | Finding               | Lateralization | Region
---------------------|----------------------|----------------|--------
Seizure semiology    | Aura type, motor sx   | L / R / Bilat  | Temporal/Frontal/etc
Interictal EEG       | Spike location         | L / R / Bilat  | Temporal/Frontal/etc
Ictal EEG            | Seizure onset          | L / R / Bilat  | Temporal/Frontal/etc
MRI                  | Structural lesion      | L / R / Bilat  | Temporal/Frontal/etc
PET                  | Hypometabolism         | L / R / Bilat  | Temporal/Frontal/etc
Neuropsych           | Memory lateralization  | L / R           | Temporal/Frontal/etc

Concordance Score: 6/6 = Excellent surgical candidate
                   4-5/6 = Good candidate, consider invasive monitoring
                   < 4/6 = Invasive monitoring required (sEEG/grids)

Workflow 4: Brain Tumor Grading and Molecular Classification¶

Attribute	Value
Workflow ID	`brain_tumor_grading`
Input	Brain MRI (T1, T1+C, T2, FLAIR, DWI, perfusion), molecular data if available
Target Latency	< 3 minutes
Models	VISTA-3D (tumor segmentation -- enhancing, non-enhancing, edema), perfusion analysis
Key Outputs	WHO 2021 integrated diagnosis prediction, tumor volume (enhancing + FLAIR), eloquent cortex proximity, molecular marker predictions (IDH, MGMT, 1p/19q), treatment protocol recommendation, RANO baseline measurements
Severity Routing	Large mass effect or midline shift -> Emergent neurosurgery
Cross-Modal Trigger	Suspected glioma -> molecular panel (IDH1/2, ATRX, TERT, CDKN2A, H3K27M)
Guideline Alignment	WHO 2021 CNS Tumor Classification, NCCN CNS Cancers Guidelines

WHO 2021 Integrated Classification:

Histology	IDH	1p/19q	ATRX	Grade	Integrated Diagnosis	Prognosis
Astrocytic	Mutant	Intact	Lost	2-4	Astrocytoma, IDH-mutant	Better (mOS 5-10y)
Oligodendroglial	Mutant	Codeleted	Retained	2-3	Oligodendroglioma, IDH-mutant, 1p/19q-codeleted	Best (mOS 12-15y)
Astrocytic/GBM features	Wildtype	Intact	Retained	4	Glioblastoma, IDH-wildtype	Worst (mOS 15 mo)
Diffuse midline	Wildtype	--	--	4	Diffuse midline glioma, H3K27M-altered	Poor (mOS 9-11 mo)

Workflow 5: Multiple Sclerosis Monitoring¶

Attribute	Value
Workflow ID	`ms_monitoring`
Input	Brain and spinal cord MRI (current + prior), clinical data, EDSS, labs
Target Latency	< 3 minutes
Models	VISTA-3D (lesion segmentation, lesion comparison, brain volume change)
Key Outputs	New T2 lesion count, enhancing lesion count, total lesion volume change, brain volume change (annualized), NEDA-3 status, DMT escalation recommendation, PML risk (JCV index)
Severity Routing	Multiple new enhancing lesions on DMT -> Treatment failure, urgent MS specialist
Cross-Modal Trigger	Atypical MS features (progressive from onset, bilateral optic neuritis) -> NMOSD/MOGAD antibody panel + AQP4/MOG genetics
Guideline Alignment	AAN DMT Guidelines 2018, McDonald 2017 Criteria

NEDA-3 Assessment:

No Evidence of Disease Activity (NEDA-3):
  1. No relapses in past 12 months
  2. No new or enlarging T2 lesions on MRI
  3. No confirmed disability progression (EDSS stable or improved)

All three criteria must be met = NEDA-3 achieved
Any criterion failed = NEDA-3 not achieved -> Consider DMT escalation

DMT Escalation Ladder:
  Platform therapies: Interferons, glatiramer acetate, teriflunomide, DMF
  Moderate efficacy: Fingolimod, cladribine
  High efficacy: Natalizumab (JCV risk), ocrelizumab, ofatumumab, alemtuzumab
  Emerging: BTK inhibitors (tolebrutinib, fenebrutinib -- trials ongoing)

Escalation triggers:
  - >= 1 relapse on current DMT
  - >= 2 new T2 or >= 1 enhancing lesion
  - Confirmed EDSS worsening (>= 1.0 if EDSS <= 5.5, >= 0.5 if EDSS > 5.5)
  - Suboptimal brain volume loss (> 0.4%/year)

Workflow 6: Parkinson's Disease Assessment¶

Attribute	Value
Workflow ID	`parkinsons_assessment`
Input	Clinical motor exam, DaT-SPECT, brain MRI, cognitive testing
Target Latency	< 2 minutes
Key Outputs	MDS clinical diagnostic criteria assessment, UPDRS Part III motor score, Hoehn & Yahr stage, tremor-dominant vs PIGD classification, DaT-SPECT interpretation, medication optimization (LEDD calculation), DBS candidacy assessment, red flags for atypical parkinsonism
Severity Routing	Red flags for MSA/PSP/CBD -> Movement disorder specialist referral
Cross-Modal Trigger	Early-onset PD (<50) or Jewish ancestry -> PD gene panel (LRRK2, GBA, SNCA, PARK2/Parkin, PINK1, DJ-1)
Guideline Alignment	MDS Clinical Diagnostic Criteria, AAN PD Treatment Guidelines

Atypical Parkinsonism Red Flags:

Red Flag	Suggests
Poor levodopa response	MSA, PSP, CBD, vascular parkinsonism
Early falls (within 1 year)	PSP
Vertical supranuclear gaze palsy	PSP
Cerebellar ataxia	MSA-C
Severe autonomic failure	MSA
Asymmetric cortical signs (apraxia, alien limb)	CBD
Rapid cognitive decline	DLB, CBD, CJD
Wheelchair-bound within 5 years	PSP, MSA
Midbrain atrophy ("hummingbird sign")	PSP
Pontine "hot cross bun" sign	MSA-C
Putaminal rim sign (T2*)	MSA-P

Workflow 7: Headache Classification and Management¶

Attribute	Value
Workflow ID	`headache_classification`
Input	Headache characteristics, associated symptoms, exam findings, imaging if indicated
Target Latency	< 30 seconds
Key Outputs	ICHD-3 classification, red flag assessment (SNOOP), imaging recommendation, acute treatment plan, preventive treatment candidacy, disability score (HIT-6, MIDAS), CGRP therapy eligibility
Severity Routing	Red flags (thunderclap, new neurological deficit, papilledema) -> Emergent imaging
Guideline Alignment	ICHD-3, AAN Migraine Prevention Guidelines, AHS Position Statements

SNOOP Red Flag Mnemonic:

S - Systemic symptoms/signs (fever, weight loss, cancer, HIV, pregnancy)
N - Neurologic symptoms/signs (confusion, focal deficit, papilledema, seizure)
O - Onset: Sudden/thunderclap (< 1 minute to peak -> SAH until proven otherwise)
O - Older: New headache onset after age 50 (temporal arteritis, mass, subdural)
P - Previous headache history: First or worst headache, change in character
    + Pattern change, Positional component, Precipitated by Valsalva

Any SNOOP flag present -> Imaging and further workup required before primary headache diagnosis

Workflow 8: Neuromuscular Disease Evaluation¶

Attribute	Value
Workflow ID	`neuromuscular_evaluation`
Input	Clinical presentation, EMG/NCS data, serological data, genetic data
Target Latency	< 2 minutes
Key Outputs	EMG/NCS pattern classification (axonal vs demyelinating, motor vs sensory, proximal vs distal), differential diagnosis ranked by probability, antibody testing recommendations, genetic testing recommendations, treatment plan, functional assessment
Severity Routing	Rapidly progressive weakness (GBS) or respiratory compromise -> ICU admission
Cross-Modal Trigger	Hereditary neuropathy pattern -> CMT gene panel (PMP22, MFN2, GJB1, MPZ); Suspected SMA -> SMN1 testing; Suspected muscular dystrophy -> dystrophy panel (DMD, DMPK, CNBP)
Guideline Alignment	AAN Practice Parameters for GBS, CIDP, MG, ALS

EMG/NCS Pattern Recognition:

Pattern	NCS Motor	NCS Sensory	EMG	Diagnosis
Diffuse demyelinating motor + sensory	Slow CV, prolonged F-waves, temporal dispersion	Slow CV, low amplitude	Reduced recruitment	CIDP, GBS (AIDP)
Length-dependent axonal sensorimotor	Low amplitude distally	Low amplitude distally	Fibrillations, large MUAPs	Diabetic/metabolic neuropathy
Motor-predominant axonal	Low amplitude, normal CV	Normal	Fibrillations, fasciculations, large MUAPs, reduced recruitment	ALS, MMN
NMJ decrement	Decrement >10% at 3Hz RNS	Normal	Unstable MUAPs, variable firing	Myasthenia gravis, LEMS
Myopathic	Normal	Normal	Short, small, polyphasic MUAPs, early recruitment	Inflammatory myopathy, dystrophy
Uniform demyelinating motor + sensory	Uniformly slow CV	Uniformly slow	Reduced recruitment	CMT1A (hereditary)

8.1 Neurogenetics Triggers¶

The Neurology Intelligence Agent implements cross-modal triggers that automatically query the shared genomic_evidence collection (3.5M variants) when clinical or imaging findings suggest a genetic etiology:

Trigger Condition	Gene Panel Queried	Clinical Rationale
Early-onset AD (<65) or family history	PSEN1, PSEN2, APP, APOE	1-5% of AD is autosomal dominant; APOE e4 major risk factor
FTD features (behavioral/language)	MAPT, GRN, C9orf72	30-50% of FTD is genetic; C9orf72 is most common cause
Early-onset PD (<50) or family history	LRRK2, GBA, SNCA, PARK2, PINK1, DJ-1	GBA variants in 5-15% of PD; affects prognosis and therapy
Drug-resistant epilepsy, age < 25	SCN1A, KCNQ2, CDKL5, TSC1/2, DEPDC5, SLC2A1, PCDH19	Genetic diagnosis changes ASM selection (e.g., avoid Na+ channel blockers in SCN1A)
Unexplained neuropathy (young onset)	PMP22, MFN2, GJB1, MPZ, TTR	CMT affects 1:2,500; TTR amyloid is treatable
Suspected SMA/muscular dystrophy	SMN1, DMD, DMPK, CNBP	Gene therapies available (nusinersen, onasemnogene)
Huntington's features (chorea + cognitive)	HTT (CAG repeat)	Definitive diagnosis; predictive testing for family members
Cryptogenic stroke, age < 50	NOTCH3 (CADASIL), COL4A1/2, MTHFR, Factor V Leiden	Monogenic small vessel disease, thrombophilia
Unexplained leukoencephalopathy	CSF1R, EIF2B1-5, LMNB1, AARS2, DARS2	Adult-onset leukodystrophies are underdiagnosed
Atypical MS / NMOSD features	AQP4, MOG (serological), HLA-DRB1	NMOSD requires different treatment than MS

8.2 Neuroimaging -> Genomics -> Therapeutics Pipeline¶

Neuroimaging Finding (MRI, CT, PET, DaT-SPECT)
    |
    v
[Cross-Modal Trigger] -- Clinical criteria met?
    |                         |
    YES                       NO
    |                         |
    v                         v
[Query genomic_evidence]   [Standard RAG response]
(3.5M variant vectors)
    |
    v
[Variant Annotation]
ClinVar pathogenicity, AlphaMissense score, ACMG classification
    |
    v
[Precision Therapeutics Selection]
Gene-specific therapy matching:
  SCN1A+ -> Avoid carbamazepine, phenytoin; use stiripentol + VPA + CLB
  GBA+ PD -> Consider GBA-targeted therapies (venglustat, ambroxol trials)
  SMN1 del -> Nusinersen, onasemnogene, risdiplam
  TTR+ amyloid neuropathy -> Tafamidis, patisiran, vutrisiran
    |
    v
[HCLS AI Factory Stage 3: Drug Discovery]
Novel compound generation for undruggable targets
    |
    v
[Clinical Output]
FHIR R4 DiagnosticReport with genomic enrichment

8.3 Integration with Other Agents¶

Integration	Direction	Data Flow
Imaging Agent	Bidirectional	Shares brain MRI/CT workflows; receives DICOM routing for neuro studies
Cardiology Agent	Bidirectional	Stroke-cardiology interface (AF detection -> stroke prevention); cardio-embolic stroke workup
Precision Biomarker Agent	Inbound	Receives NfL, tau, amyloid biomarker reference ranges
Precision Oncology Agent	Bidirectional	Neuro-oncology molecular data; CNS metastasis management
Genomics Pipeline	Read-only	Queries `genomic_evidence` for neurological gene variants
Drug Discovery Pipeline	Outbound	Sends confirmed genetic targets for compound screening

9. NIM Integration Strategy¶

9.1 Shared NIM Services¶

NIM	Port	Neurology Application
VISTA-3D	8530	Brain segmentation (hippocampus, ventricles, cortex, white matter, tumors, lesions), volumetrics
MAISI	8531	Synthetic brain MRI generation for training, rare pathology simulation
VILA-M3	8532	Brain image interpretation, EEG pattern recognition, radiology report assistance
Llama-3 8B	8520	Clinical reasoning fallback when Claude API unavailable

9.2 VISTA-3D Neurology Applications¶

Hippocampal volumetry: Bilateral hippocampal segmentation for dementia evaluation; age/sex-percentile comparison
White matter lesion quantification: FLAIR lesion segmentation for MS monitoring, small vessel disease grading
Brain tumor segmentation: Enhancing tumor, non-enhancing tumor, peritumoral edema volumes for RANO criteria
Brain parenchymal fraction: Global and regional atrophy quantification for neurodegeneration tracking
Ventricular volume: Hydrocephalus assessment, NPH evaluation
Cortical thickness mapping: Regional cortical atrophy patterns for differential diagnosis

10. Knowledge Graph Design¶

10.1 Graph Structure¶

Dictionary	Entries	Content
Neurological Conditions	~45	ICD-10, diagnostic criteria, imaging patterns, scales, genetic associations
Clinical Scales	~30	Input variables, scoring, interpretation, clinical use
Drug Classes	~35	Mechanism, dosing, interactions, specific neuro indications/contraindications
Neuroimaging Patterns	~25	Modality, sequence, pattern description, differential diagnosis
Neurological Genes	~60	Gene symbol, inheritance, phenotype, genetic testing indications, treatment implications
EEG/EMG Patterns	~20	Pattern name, morphology, clinical correlation, urgency

10.2 Example Knowledge Graph Entries¶

Condition: Dravet Syndrome

{
    "name": "Dravet Syndrome",
    "icd10": "G40.409",
    "aliases": ["SMEI", "severe myoclonic epilepsy of infancy"],
    "prevalence": "1:15,000 - 1:40,000",
    "inheritance": "De novo (90%), autosomal dominant",
    "age_of_onset": "4-8 months, first febrile seizure",
    "diagnostic_criteria": {
        "clinical": "Prolonged febrile seizures before age 1, followed by afebrile seizures",
        "eeg": "Normal initially, then generalized spike-wave, photosensitivity",
        "mri": "Normal initially, later hippocampal sclerosis possible",
        "genetic": "SCN1A pathogenic variant (>80%)"
    },
    "genes": ["SCN1A"],
    "treatment": {
        "effective": ["Valproic acid", "Clobazam", "Stiripentol", "Cannabidiol (Epidiolex)", "Fenfluramine (Fintepla)"],
        "contraindicated": ["Carbamazepine", "Oxcarbazepine", "Phenytoin", "Lamotrigine", "Vigabatrin"],
        "contraindication_reason": "Sodium channel blockers worsen seizures in SCN1A loss-of-function"
    },
    "prognosis": "Drug-resistant epilepsy, intellectual disability, increased SUDEP risk",
    "cross_modal_trigger": True,
    "critical_pharmacogenomic": True
}

Clinical Scale: NIHSS (National Institutes of Health Stroke Scale)

{
    "name": "NIHSS",
    "full_name": "National Institutes of Health Stroke Scale",
    "clinical_use": "Quantify stroke severity, guide treatment decisions, predict outcome",
    "scoring": {
        "range": "0-42",
        "categories": {
            "0": "No stroke symptoms",
            "1-4": "Minor stroke",
            "5-15": "Moderate stroke",
            "16-20": "Moderate to severe stroke",
            "21-42": "Severe stroke"
        }
    },
    "domains_assessed": [
        "1a. Level of consciousness",
        "1b. LOC questions (month, age)",
        "1c. LOC commands (open/close eyes, grip/release)",
        "2. Best gaze (horizontal eye movement)",
        "3. Visual fields",
        "4. Facial palsy",
        "5. Motor arm (L and R)",
        "6. Motor leg (L and R)",
        "7. Limb ataxia",
        "8. Sensory",
        "9. Best language",
        "10. Dysarthria",
        "11. Extinction/inattention"
    ],
    "clinical_decision_thresholds": {
        "tpa_consideration": ">= 4 (measurable deficit)",
        "thrombectomy_consideration": ">= 6 with LVO",
        "minor_stroke_no_intervention": "0-3 (unless disabling deficit)"
    },
    "guideline_source": "AHA/ASA 2019 Acute Ischemic Stroke Guidelines"
}

11. Query Expansion and Retrieval Strategy¶

11.1 Neurology-Specific Query Expansion Maps¶

Sixteen domain-specific expansion maps:

Map	Keywords -> Terms	Example
Stroke	25 -> 200	"brain attack" -> stroke, CVA, cerebrovascular accident, ischemic stroke, hemorrhagic stroke, TIA, thrombectomy, tPA
Dementia	25 -> 180	"memory loss" -> dementia, Alzheimer's, cognitive decline, MCI, neurodegeneration, amyloid, tau
Epilepsy	20 -> 160	"fits" -> seizure, epilepsy, convulsion, ictal, interictal, epileptiform, paroxysmal, ASM
MS	20 -> 140	"relapse" -> multiple sclerosis, demyelination, lesion, RRMS, SPMS, DMT, NEDA
Parkinson's	20 -> 140	"tremor" -> Parkinson's disease, parkinsonism, dopaminergic, bradykinesia, rigidity, levodopa
Brain Tumor	15 -> 100	"brain cancer" -> glioma, glioblastoma, meningioma, IDH, MGMT, neuro-oncology, WHO grade
Headache	15 -> 100	"bad headache" -> migraine, cluster, tension, CGRP, triptan, aura, photophobia
Neuromuscular	15 -> 90	"weak muscles" -> neuropathy, myopathy, myasthenia, ALS, GBS, CIDP, motor neuron
EEG	15 -> 100	"brain waves" -> EEG, electroencephalogram, epileptiform, spike, seizure monitoring, VEEG
Neuroimaging	15 -> 100	"brain scan" -> MRI, CT, PET, FLAIR, DWI, brain volumetrics, white matter
Neurogenetics	15 -> 100	"genetic brain disease" -> neurogenetic, channelopathy, trinucleotide repeat, gene panel
Movement	10 -> 70	"shaking" -> tremor, dystonia, chorea, ataxia, tics, myoclonus, DBS
Sleep Neurology	10 -> 60	"sleep problems" -> insomnia, narcolepsy, RBD, sleep apnea, restless legs
Neuroimmunology	10 -> 80	"brain inflammation" -> encephalitis, autoimmune, NMDA, LGI1, CASPR2, MOG, AQP4
Neurorehab	10 -> 60	"brain recovery" -> neuroplasticity, rehabilitation, functional recovery, constraint therapy
CSF Analysis	10 -> 70	"spinal tap" -> lumbar puncture, CSF, oligoclonal bands, protein, glucose, cytology

Total: ~250 keywords -> ~1,750 expanded terms

11.2 Comparative Analysis Detection¶

Cardiology-specific comparisons:

Comparison	Clinical Relevance
"Lecanemab vs Donanemab"	Anti-amyloid therapy selection in early AD
"Carbamazepine vs Lamotrigine"	First-line ASM for focal epilepsy
"Natalizumab vs Ocrelizumab"	High-efficacy DMT selection in MS
"Levodopa vs Dopamine Agonist"	Initial PD therapy
"tPA vs Thrombectomy"	Acute stroke intervention strategy
"TMZ vs CCNU"	Chemotherapy for recurrent glioma
"DBS STN vs GPi"	DBS target selection in PD
"Erenumab vs Fremanezumab"	CGRP mAb selection for migraine prevention

12. API and UI Design¶

12.1 FastAPI Endpoints (Port 8528)¶

Method	Path	Purpose
GET	`/health`	Service health, collection stats, NIM status
GET	`/collections`	List all collections with vector counts
POST	`/query`	Full RAG query with evidence synthesis
POST	`/search`	Evidence-only search (no LLM)
POST	`/api/ask`	Chat-style question answering
POST	`/find-related`	Cross-collection entity linking
GET	`/workflows`	List available clinical workflows
POST	`/workflow/{name}/run`	Execute a clinical workflow
GET	`/demo-cases`	List pre-loaded demo cases
POST	`/demo-cases/{id}/run`	Run a demo case
POST	`/scale/calculate`	Calculate validated neurological scale
POST	`/diagnostic/algorithm`	Run diagnostic algorithm (e.g., dementia differential)
POST	`/protocol/recommend`	Imaging protocol recommendation
POST	`/reports/generate`	Generate report (markdown, JSON, PDF, FHIR)
GET	`/knowledge/stats`	Knowledge graph statistics
GET	`/metrics`	Prometheus-compatible metrics

12.2 Streamlit UI (Port 8529) -- 10 Tabs¶

Tab	Purpose
Evidence Explorer	Multi-collection RAG Q&A with citations, pre-filled neuro queries, Plotly donut chart
Workflow Runner	8 clinical workflows with pre-loaded demo cases, AI-annotated brain images, pipeline animation
Neuroimaging Gallery	Brain MRI/CT/PET showcase with AI annotations, 3D volume viewer, before/after toggle
Clinical Scales	Interactive NIHSS, GCS, MoCA, UPDRS, EDSS, mRS, HIT-6, ALSFRS-R calculators
Diagnostic Algorithms	Step-through diagnostic pathways for dementia DDx, seizure classification, neuropathy workup
Device & AI Ecosystem	120+ FDA-cleared neuro AI devices, searchable by modality and condition
Protocol Advisor	Patient-specific neuroimaging protocol recommendations
Reports & Export	Markdown, JSON, NVIDIA-branded PDF, FHIR R4 DiagnosticReport
Patient 360	Cross-modal neuro dashboard: imaging + EEG + genomics + scales with Plotly network graph
Guidelines & Trials	AAN/EAN/ILAE guideline browser, landmark trial summaries

12.3 Demo Cases¶

ID	Title	Workflow	Key Features
DEMO-001	Acute Stroke: LVO with Thrombectomy Decision	acute_stroke_triage	NIHSS 14, ASPECTS 8, LVO M1, DAWN criteria
DEMO-002	Memory Clinic: Early-Onset Alzheimer's	dementia_evaluation	MoCA 18, hippocampal atrophy, amyloid PET+, PSEN1 trigger
DEMO-003	Drug-Resistant Epilepsy: Surgical Evaluation	epilepsy_focus_localization	Concordance matrix, hippocampal sclerosis, SCN1A trigger
DEMO-004	New Brain Mass: Suspected Glioblastoma	brain_tumor_grading	Ring-enhancing mass, tumor volumes, IDH/MGMT prediction
DEMO-005	Young Woman with Optic Neuritis: MS vs NMOSD	ms_monitoring	McDonald criteria, OCB+, NMOSD differential

13. Clinical Decision Support Engines¶

13.1 Validated Neurological Scales¶

Scale	Use Case	Range	Interpretation
NIHSS	Stroke severity	0-42	0=no symptoms, 1-4=minor, 5-15=moderate, 16-20=mod-severe, 21+=severe
GCS	Consciousness level	3-15	3-8=severe (coma), 9-12=moderate, 13-15=mild
MoCA	Cognitive screening	0-30	>=26 normal, <26 cognitive impairment, <18 probable dementia
UPDRS Part III	PD motor severity	0-132	Higher = more severe motor impairment
EDSS	MS disability	0-10	0=normal, 4=walking limited, 6=unilateral assistance, 7=wheelchair, 10=death
mRS	Global disability (stroke)	0-6	0=no symptoms, 2=slight disability, 4=mod-severe, 6=dead
HIT-6	Headache impact	36-78	<=49=little/no impact, 50-55=some, 56-59=substantial, >=60=severe
ALSFRS-R	ALS function	0-48	48=normal, decline ~1 pt/month typical, faster=worse prognosis
CDR	Dementia staging	0-3	0=normal, 0.5=MCI, 1=mild, 2=moderate, 3=severe dementia
ASPECTS	CT stroke scoring	0-10	10=normal, >=6=good candidate for intervention, <6=large infarct

14. Reporting and Interoperability¶

14.1 Export Formats¶

Format	Use Case	Standards
Markdown	Clinical narrative, consultation notes	--
JSON	Programmatic consumption, dashboards	--
PDF	NVIDIA-themed clinical documentation	ReportLab
FHIR R4	EHR integration, interoperability	SNOMED CT, LOINC, DICOM

14.2 FHIR R4 Neurological Coding¶

Element	Code System	Example Codes
Findings	SNOMED CT	230690007 (Stroke), 26929004 (AD), 84757009 (Epilepsy), 24700007 (MS), 49049000 (PD)
Observations	LOINC	72172-0 (NIHSS), 9269-2 (GCS), 72133-2 (MoCA), LP6464-2 (MRI Brain)
Procedures	SNOMED CT	429064006 (Thrombectomy), 445185007 (Brain MRI), 54550000 (EEG)
Medications	RxNorm	Lecanemab, levetiracetam, ocrelizumab, levodopa/carbidopa, sumatriptan

15. Product Requirements Document¶

15.1 Product Vision¶

Vision Statement: Enable any neurologist, anywhere, to access integrated neurological intelligence combining neuroimaging AI, electrophysiology correlation, genomic analysis, validated clinical scales, and evidence synthesis -- on a single $3,999 device.

15.2 User Stories¶

Epic 1: Evidence-Based Clinical Queries¶

ID	User Story	Priority	Acceptance Criteria
US-001	As a neurologist, I want to ask clinical questions and receive evidence-grounded answers with citations.	P0	Query returns answer with >=3 citations from >=2 collections; <30 sec
US-002	As a neurologist, I want comparative analysis ("Lecanemab vs Donanemab") with structured tables.	P0	Auto-detected; side-by-side evidence; structured comparison
US-003	As a fellow, I want pre-filled example queries for common neuro scenarios.	P1	>=4 clickable examples; each returns relevant results
US-004	As a researcher, I want to filter by neurological domain, imaging modality, and year.	P1	Sidebar filters applied; results reflect filters

Epic 2: Clinical Workflows¶

ID	User Story	Priority	Acceptance Criteria
US-005	As a stroke neurologist, I want acute stroke triage with NIHSS, ASPECTS, LVO detection, and thrombectomy eligibility.	P0	Correct NIHSS scoring; ASPECTS calculation; DAWN/DEFUSE-3 criteria applied
US-006	As a memory clinic physician, I want dementia differential diagnosis with ATN classification.	P0	Atrophy pattern -> differential; ATN staging; genetic triggers for early-onset
US-007	As an epileptologist, I want seizure classification and concordance assessment for surgical evaluation.	P0	ILAE classification; concordance matrix; ASM recommendations by type
US-008	As a neuro-oncologist, I want WHO 2021 integrated tumor classification prediction from imaging.	P0	Tumor segmentation volumes; molecular prediction; treatment protocol
US-009	As an MS specialist, I want NEDA-3 assessment with DMT escalation recommendations.	P1	Lesion comparison; brain atrophy rate; NEDA-3 status; escalation ladder

Epic 3: Clinical Scales¶

ID	User Story	Priority	Acceptance Criteria
US-010	As a neurologist, I want interactive NIHSS, GCS, and MoCA calculators with interpretation.	P0	All items scored; total with severity category; guideline recommendation
US-011	As an MS specialist, I want EDSS calculation with progression tracking over time.	P1	EDSS 0-10; functional system scores; progression confirmation criteria
US-012	As a movement disorder specialist, I want UPDRS scoring with LEDD calculation.	P1	UPDRS Part III; H&Y staging; LEDD from medication list

ID	User Story	Priority	Acceptance Criteria
US-013	As a neurogenetics specialist, I want automatic genomic triggers from neuroimaging findings.	P0	Qualifying imaging pattern triggers gene panel query; genomic hits displayed
US-014	As a neurologist, I want Patient 360 combining imaging, EEG, genomics, and scales.	P1	Interactive network graph; cross-modal connections; drill-down to evidence

Epic 5: Reporting and Export¶

ID	User Story	Priority	Acceptance Criteria
US-015	As a neurologist, I want PDF clinical reports with neurological findings, scales, and evidence.	P0	NVIDIA-branded PDF; all sections populated; download button
US-016	As a health IT engineer, I want FHIR R4 DiagnosticReport with neuro SNOMED/LOINC codes.	P1	Valid FHIR R4; neuro-specific coding; passes validator

Epic 6: Demo and Presentation¶

ID	User Story	Priority	Acceptance Criteria
US-017	As a demo presenter, I want 5 pre-loaded demo cases covering the breadth of neurology.	P0	5 cases executable; <30 seconds each; realistic output
US-018	As a new user, I want a sidebar guided tour for the 10-tab interface.	P1	Expandable tour; numbered steps; dismiss button

15.3 Non-Functional Requirements¶

Requirement	Target
RAG query latency	< 30 seconds end-to-end
Scale calculator latency	< 5 seconds
Workflow execution (mock)	< 10 seconds
Memory footprint	< 16 GB (agent only)
Seed data completeness	2,270+ records across 13 collections
Unit test coverage	> 80%
FHIR R4 compliance	Passes HL7 FHIR Validator

15.4 Prioritization Matrix¶

Phase	Features	Timeline
Phase 1 (MVP)	RAG engine (13 collections), Evidence Explorer, 3 workflows (stroke, dementia, epilepsy), 4 scales (NIHSS, GCS, MoCA, mRS), 3 demo cases, PDF export	5-7 weeks
Phase 2 (Complete)	All 8 workflows, all 10 scales, diagnostic algorithms, FHIR R4, neuroimaging gallery, Patient 360, all 5 demo cases	5-7 weeks
Phase 3 (Polish)	Guided tour, pipeline animation, cross-modal triggers, network graph, Guidelines & Trials tab, EEG gallery	2-3 weeks

16. Data Acquisition Strategy¶

16.1 Automated Ingest Pipelines¶

Source	Collection(s)	Method	Update Cadence
PubMed (NCBI E-utilities)	neuro_literature	MeSH-filtered neuroscience retrieval	Weekly
ClinicalTrials.gov (V2 API)	neuro_trials	Neurological condition filter	Weekly
AAN Practice Guidelines	neuro_guidelines	Manual curation + embedding	Per guideline update

16.2 PubMed Search Strategy¶

MeSH Terms:
  "Nervous System Diseases"[MeSH] OR "Neurodegenerative Diseases"[MeSH] OR
  "Epilepsy"[MeSH] OR "Stroke"[MeSH] OR "Multiple Sclerosis"[MeSH] OR
  "Brain Neoplasms"[MeSH] OR "Parkinson Disease"[MeSH] OR "Alzheimer Disease"[MeSH]

AND ("Artificial Intelligence"[MeSH] OR "Machine Learning"[MeSH] OR
     "Deep Learning"[MeSH] OR "Neural Networks"[MeSH] OR
     "Neuroimaging"[MeSH])

Filters: Published 2018-2026, English, Humans
Expected: 3,500-6,000 abstracts

17. Validation and Testing Strategy¶

17.1 Unit Tests¶

Test Category	Target Count
Collection schemas	39 (13 x 3)
Clinical scales	50 (10 scales x 5 cases)
Workflow logic	40 (8 x 5)
RAG engine	20
Knowledge graph	15
API endpoints	30
FHIR R4 export	15
Cross-modal triggers	15
Diagnostic algorithms	20
Total	~245

17.2 Clinical Validation¶

Validation Type	Method	Target
Scale accuracy	Compare against published validation data	< 1% deviation
Diagnostic algorithms	Board-certified neurologist review	95%+ guideline concordance
Seizure classification	Compare against ILAE criteria	100% match for standard inputs
FHIR R4 compliance	HL7 FHIR Validator	Zero validation errors

18. Regulatory Considerations¶

18.1 FDA CDS Exemption (21st Century Cures Act)¶

Criterion	Assessment
Not intended to acquire, process, or analyze a medical image or signal	Met -- RAG and scale calculation; does not process raw images/EEG
Intended for displaying, analyzing, or printing medical information	Met
Intended for use by healthcare professional	Met -- Designed for neurologists
Healthcare professional does not primarily rely on the software	Met -- Provides recommendations for review

Assessment: Core RAG and scale calculator functions likely qualify for CDS exemption. NIM-based image analysis workflows would require separate regulatory consideration.

18.2 Disclaimers¶

This tool is for clinical decision support only and does not replace professional medical judgment. All findings require verification by a qualified neurologist. Scale calculations are estimates and must be confirmed by clinical examination. Not FDA-cleared for autonomous clinical decision-making. For research and educational purposes only.

19. DGX Compute Progression¶

Phase	Hardware	Price	Scope
Phase 1 -- Proof Build	DGX Spark	$3,999	All 8 workflows (mock/cloud NIM), 13 collections, 10 clinical scales
Phase 2 -- Departmental	1-2x DGX B200	$500K-$1M	Full NIM stack, live MRI/CT/EEG processing, PACS integration
Phase 3 -- Multi-Site	4-8x DGX B200	$2M-$4M	NVIDIA FLARE federated learning across neurology centers
Phase 4 -- AI Factory	DGX SuperPOD	$7M-$60M+	National brain health surveillance, real-time ICU neuro monitoring

20. Implementation Roadmap¶

Phase 1: Foundation (Weeks 1-7)¶

Week	Deliverable
1-2	Repository scaffolding, Pydantic models, settings, Docker Compose, 13 collection schemas
3-4	Seed data curation (2,270 records), ingest pipelines (PubMed, ClinicalTrials.gov)
5-7	RAG engine, Evidence Explorer tab, 3 priority workflows (stroke, dementia, epilepsy), 4 clinical scales

Phase 2: Clinical Intelligence (Weeks 8-14)¶

Week	Deliverable
8-9	Remaining 5 workflows (brain tumor, MS, Parkinson's, headache, neuromuscular)
10-11	Remaining 6 clinical scales, diagnostic algorithm engine
12-14	Cross-modal triggers, FHIR R4 export, PDF reports

Phase 3: UI and Polish (Weeks 15-18)¶

Week	Deliverable
15-16	10-tab Streamlit UI, neuroimaging gallery, Patient 360 network graph
17-18	Guided tour, demo cases finalized, documentation, integration testing

21. Risk Analysis¶

Risk	Probability	Impact	Mitigation
Clinical scale implementation errors	Medium	High	Validate against published reference implementations
Diagnostic algorithm complexity (dementia DDx)	Medium	Medium	Start with most common differentials; expand iteratively
Insufficient EEG/EMG seed data	Medium	Medium	Focus on patterns with highest clinical impact
NIM brain segmentation accuracy for volumetrics	Low	Medium	Age/sex-normalized percentiles with published norms
Guideline updates during development	Low	Low	Modular guideline collection
Memory pressure with 13 collections + other agents	Low	Medium	BGE-small (384-dim) is compact
Neurogenetics variant interpretation complexity	Medium	Medium	Leverage ClinVar pathogenicity + AlphaMissense scores from shared collection

22. Competitive Landscape¶

22.1 Positioning¶

                    Multi-Condition Coverage
                           ^
                           |
                    [Neurology Agent]
                           |
         Cloud-only <------+------> On-device
                           |
                    [Viz.ai, Aidoc]
                           |
                           v
                    Single-Condition

No existing product combines: 1. Multi-condition neurological coverage (stroke + dementia + epilepsy + tumors + MS + PD + headache + NMD) 2. Cross-modal integration (imaging + EEG + genomics) 3. RAG-grounded evidence synthesis with citations 4. Validated neurological scales (NIHSS, GCS, MoCA, UPDRS, EDSS) 5. Diagnostic algorithms 6. On-device deployment ($3,999) 7. Open-source (Apache 2.0)

22.2 Defensibility¶

Advantage	Defensibility
Multi-collection RAG architecture	High -- 11 agents prove the pattern
Cross-modal genomic triggers	High -- unique to HCLS AI Factory
On-device inference	High -- DGX Spark + NIM is NVIDIA-exclusive
Breadth of neurological coverage	High -- no competitor covers 8 subspecialties
Open source	Medium -- community adoption
Clinical validation	Medium -- requires domain expertise

23. Discussion¶

23.1 Why Neurology Is a High-Impact Agent¶

Workforce crisis: With a projected shortfall of 11,000 neurologists by 2031, AI-assisted decision support is not optional -- it is a clinical necessity. A community neurologist covering stroke, epilepsy, MS, dementia, and headache cannot maintain subspecialty-level expertise across all domains.
Diagnostic complexity: Many neurological conditions require multi-modal data synthesis (imaging + EEG + genomics + clinical scales) that exceeds the cognitive capacity of a single clinician reviewing siloed data. AI excels at this pattern.
Genomic transformation: Over 1,000 neurological genes with clinical actionability. SCN1A changes epilepsy treatment. GBA changes Parkinson's prognosis. APOE/PSEN1 changes Alzheimer's management. A genomic-integrated intelligence agent enables precision neurology.
Therapeutic pipeline explosion: Anti-amyloid antibodies, gene therapies, antisense oligonucleotides, and BTK inhibitors are transforming previously untreatable conditions. Selecting the right therapy for the right patient at the right time requires evidence synthesis beyond PubMed searches.
Existing foundation: The HCLS AI Factory already has stroke triage (DEMO-001 in Imaging Agent), brain MRI MS lesion tracking, hippocampal volumetry via VISTA-3D, and neurological genes in the genomic_evidence collection.
Differentiation: No competitor addresses neurology's full breadth. Viz.ai does stroke. Persyst does EEG. QMENTA does imaging. None integrate all three with genomics, scales, and literature.

23.2 Limitations¶

Mock mode inference: Phase 1 uses synthetic results; GPU deployment required for clinical validation.
EEG signal processing: Raw EEG analysis requires specialized preprocessing not covered by standard NIM services. Initial version will work with pre-interpreted EEG findings.
Scale automation: Some scales (NIHSS, UPDRS) require bedside examination and cannot be fully automated from data alone.
Rare disease coverage: With >7,000 rare neurological conditions, complete coverage is infeasible. Focus is on the 50 most common/impactful conditions.

23.3 Future Directions¶

Real-time ICU neuromonitoring: Continuous EEG seizure detection with automated NIHSS trending
Brain-computer interface integration: Neurofeedback and BCI data for rehabilitation
Longitudinal brain health platform: Population-level brain aging trajectories
Digital biomarkers: Smartphone-based gait analysis, speech markers for cognitive decline
Federated learning across epilepsy centers: NVIDIA FLARE for multi-site seizure detection models

24. Conclusion¶

The Neurology Intelligence Agent addresses a critical unmet need at the intersection of the global neurological disease burden, the neurologist workforce shortage, and the fragmentation of neurological data. By leveraging the HCLS AI Factory's proven multi-collection RAG architecture -- adapted with 13 neurology-specific collections, 8 clinical workflows, 10 validated scales, and cross-modal neuroimaging-genomics triggers -- the agent will deliver subspecialty-level neurological intelligence to any neurologist, anywhere.

The breadth of coverage (stroke, dementia, epilepsy, brain tumors, MS, Parkinson's, headache, neuromuscular disease) is unmatched by any existing commercial product, all of which address only one or two subspecialties. Combined with genomic integration that enables precision neurology (SCN1A-guided epilepsy treatment, GBA-stratified Parkinson's management, PSEN1/2 Alzheimer's diagnosis), the agent represents a new paradigm in neurological clinical decision support.

Deploying on a single NVIDIA DGX Spark ($3,999) ensures accessibility for community neurologists, rural health systems, and resource-limited institutions globally -- precisely the settings where the neurologist shortage is most acute and subspecialty expertise least available.

25. References¶

GBD 2021 Nervous System Disorders Collaborators. Global, regional, and national burden of disorders affecting the nervous system, 1990-2021. Lancet Neurol. 2024;23:344-381.
World Health Organization. Intersectoral Global Action Plan on Epilepsy and Other Neurological Disorders 2022-2031. WHO, 2023.
Feigin VL, et al. The global burden of neurological disorders. Lancet Neurol. 2024;23:2-3.
Dorsey ER, et al. Global, regional, and national burden of Parkinson's disease, 1990-2016. Lancet Neurol. 2018;17:939-953.
2023 Alzheimer's Disease Facts and Figures. Alzheimers Dement. 2023;19:1598-1695.
GBD 2016 Stroke Collaborators. Global, regional, and national burden of stroke, 1990-2016. Lancet Neurol. 2019;18:439-458.
Powers WJ, et al. Guidelines for the Early Management of Patients with Acute Ischemic Stroke: 2019 Update. Stroke. 2019;50:e344-e418.
Nogueira RG, et al. Thrombectomy 6 to 24 Hours after Stroke with a Mismatch between Deficit and Infarct (DAWN). N Engl J Med. 2018;378:11-21.
Albers GW, et al. Thrombectomy for Stroke at 6 to 16 Hours with Selection by Perfusion Imaging (DEFUSE-3). N Engl J Med. 2018;378:708-718.
Jack CR, et al. NIA-AA Research Framework: Toward a biological definition of Alzheimer's disease. Alzheimers Dement. 2018;14:535-562.
van Dyck CH, et al. Lecanemab in Early Alzheimer's Disease (CLARITY AD). N Engl J Med. 2023;388:9-21.
Sims JR, et al. Donanemab in Early Symptomatic Alzheimer Disease (TRAILBLAZER-ALZ 2). JAMA. 2023;330:512-527.
Fisher RS, et al. ILAE Official Report: A practical clinical definition of epilepsy. Epilepsia. 2014;55:475-482.
Scheffer IE, et al. ILAE Classification of the Epilepsies: Position paper. Epilepsia. 2017;58:512-521.
Thompson AJ, et al. Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. Lancet Neurol. 2018;17:162-173.
Rae-Grant A, et al. Practice guideline recommendations summary: Disease-modifying therapies for adults with multiple sclerosis (AAN). Neurology. 2018;90:777-788.
Postuma RB, et al. MDS clinical diagnostic criteria for Parkinson's disease. Mov Disord. 2015;30:1591-1601.
Headache Classification Committee of the IHS. The International Classification of Headache Disorders, 3rd edition (ICHD-3). Cephalalgia. 2018;38:1-211.
Louis DN, et al. The 2021 WHO Classification of Tumors of the Central Nervous System. Neuro Oncol. 2021;23:1231-1251.
Hemphill JC, et al. Guidelines for the Management of Spontaneous Intracerebral Hemorrhage. Stroke. 2015;46:2032-2060.
FDA. Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices. FDA Database, 2025.
American Academy of Neurology. 2023 Neurology Workforce Study. AAN, 2023.
Wirrell EC, et al. Dravet syndrome: Diagnosis, treatment, and long-term management. Epilepsia. 2022;63:S131-S147.
Hogan DB, et al. Nusinersen and the treatment of spinal muscular atrophy. CMAJ. 2018;190:E1-E3.
Grossman M, et al. Frontotemporal lobar degeneration. Nat Rev Dis Primers. 2023;9:40.
Filippi M, et al. Assessment of lesions on magnetic resonance imaging in multiple sclerosis. Brain. 2019;142:1858-1875.

HCLS AI Factory -- Neurology Intelligence Agent Research Paper and PRD Apache 2.0 License | March 2026 Author: Adam Jones