Skip to content

Clinical Imaging Engine

live   Domain: clinical   Type: engine

Live DICOM analysis (VISTA-3D segmentation, VILA-M3 image/report reasoning, FHIR R4 export) with cross-modality reasoning against genomics. MAISI synthetic imaging is research/augmentation/QA only, never diagnostic. Decision support. (Honesty: the live-verified modality is chest X-ray analysis via DenseNet-121 on real DICOM pixels; the VISTA-3D / VILA-M3 CT path is a planned/placeholder modality, not yet live.)

A narrated, captioned explainer. Decision support for a qualified clinician.

Clinical Imaging Engine — what it takes in, what it computes, what it returns

Illustrative. Decision support for a qualified clinician — never autonomous diagnosis or prescribing.

In plain terms

The Clinical Imaging Engine reads medical images and reasons about them alongside the patient's genomics. It turns a scan — delivered as standard DICOM — into structured findings a clinician can act on, and it can cross-reference those findings against what the genome says. Image and gene together often say more than either alone.

Why it matters

A radiology read and a genomic finding usually live in separate systems and separate heads. Bringing them into one reasoning step — a calcium score next to a lipid genotype, a tumor's imaging next to its fusion — is where cross-modal precision medicine actually happens.

For a patient: their scan and their genetics read together, so nothing that matters falls through the gap between two systems.

How it works

Inside the Clinical Imaging Engine — ingest DICOM, analyze pixels, reason cross-modally, export FHIR

DICOM in, structured findings out — reasoned against genomics. Illustrative.

  1. Ingest — a standard DICOM imaging study comes in (referenced through the patient context).
  2. Analyze — the live, verified modality today is chest X-ray analysis via a DenseNet-121 model on real DICOM pixels.
  3. Reason — that live read is joined cross-modally with the patient's genomics (the deeper image-and-report reasoning, VILA-M3 on CT, is on the planned path — see Honest limits).
  4. Export — findings are written out as FHIR R4 so they slot into clinical systems.

What goes in, what comes out

  • In: a query and the patient context (which references the DICOM study).
  • Out: a structured image read.

Where it fits

Where the Clinical Imaging Engine sits — feeding oncology, cardiology, and cross-modal reasoning

Imaging joins the tumor board, the calcium-score-to-statin path, and genomic reasoning. Illustrative.

Its reads feed the Precision Oncology Engine (imaging joins the tumor board) and the Cardiology Engine (a coronary-calcium input — a CT-derived value, on the planned CT path, not produced by the live chest-X-ray model), and they reason cross-modally with genomics throughout.

Honest limits

  • What's live vs. planned. The chest X-ray / DenseNet-121 path is the live-verified modality. The VISTA-3D segmentation and VILA-M3 CT path is planned / placeholder — not yet live.
  • Synthetic imaging is not diagnostic. MAISI synthetic image generation is for research / augmentation / QA only — never a diagnostic source.
  • Decision support, not diagnosis. The read supports a qualified clinician; it does not diagnose on its own.

Interface

  • Endpoint: localhost:8525 · Invoke path: /
  • Serving: native · GPU: no · Cost class: —

Inputs

Name Shape Semantic Notes
query scalar —
patient_context map patient_context

Outputs

Name Shape Semantic Notes
answer map image_read

Tags: engine · imaging

↩ Back to the Engines index · the Capability Maturity Matrix · the Capability Brief.

Note

Status and interface are generated from the capability registry (lib/hcls_common/capabilities.json) — the site cannot claim ahead of the code. All clinical output is decision support for a qualified clinician, never autonomous diagnosis.