Adaptive route
The case state determines whether CLARIS needs clinical context, imaging execution, treatment context, or evidence retrieval.
CLARIS
HCC case investigation
Clinical insight that shows its work.
CLARIS brings patient state, imaging context, evidence retrieval, and validation into one transparent workstation for liver cancer research and clinician review.
61
backend tests
10
golden cases
100%
supported claims
12/12
adversarial pass
Live Case
HCC_015
Patient case builtcomplete
Imaging measurementsavailable
AFP contextobserved
Previous comparisonlimited
Imaging context, not guesswork.
The interface distinguishes observed measurements, unavailable scans, and review flags.
Liver
CT
Lung modelHow it works
A case investigation workflow, not a black-box answer.
CLARIS shows the path from patient state to assessment. The user can inspect what was used, what was skipped, and why the system stopped.
Patient case
Normalize patient, imaging, treatment, follow-up, and missing-data state.
Investigation
Route the case through clinical, imaging, longitudinal, and evidence capabilities.
Traceable synthesis
Generate a clinician-review assessment with claim-level provenance.
Agentic runtime
The agent is the investigation loop, not a decorative chatbot.
CLARIS uses an explicit state/action/observation loop so the UI can show why each module ran and when the investigation had enough evidence to stop.
1
Goal
2
State
3
Tool
4
Observation
5
Decision
State memory
Every observation becomes structured state before another decision is made.
Tool trace
Each tool call is visible as part of the investigation path.
Source boundary
Claims can point to patient data, measurements, evidence, or validation findings.
Clinical modules
Specialized tools, coordinated through an inspectable runtime.
The medical logic stays deterministic where it should. Routing, state, provenance, and investigation sequencing are what become agentic.
Imaging Investigation
Tracks scan availability, segmentation artifacts, measurement quality, and review flags.
Clinical Context
Preserves AFP, treatment labels, classifications, and unavailable liver-function fields.
Evidence Review
Retrieves only from the controlled corpus and keeps source IDs visible.
Evidence provenance
Every important statement needs a source.
The current system uses a controlled evidence corpus and deterministic trace checks. It does not pretend a missing lab, scan, or rule output exists.
Claim Trace
Current lesion measurement is available
ImagingExecutionResult
observedTACE treatment context exists
PatientCase.treatment_history
observedLongitudinal response cannot be established
TreatmentResponseContext
limitedFull liver-function assessment is incomplete
ClinicalContext.missing
missingCLARISReview Required
Patient
HCC_015
TACE context
AFP observed
Assessment
Current lesion measured
Prior comparison limited
Evidence linked
Activity
Clinical
Imaging
Evidence
Workstation preview
Built for review, not blind trust.
The dashboard exposes case state, investigation activity, missing data, review flags, evidence basis, and final synthesis in one dense clinical workspace.
Open DashboardDemo patients
Choose a case and inspect what CLARIS actually has.
The demo includes backend-generated patient state, evidence, decision traces, and real local CT previews where downloaded HCC-TACE-Seg imaging is available.
About DatasetHCC_015
Measurements available
Open imaging workspace
HCC_014
Measurements available
Open imaging workspace
HCC_009
Measurements require review
Open imaging workspace
HCC_029
Imaging measurements insufficient
Open imaging workspace
HCC_018
Imaging measurements insufficient
Open imaging workspace
HCC_043
Imaging measurements insufficient
Open imaging workspace
Research prototype
Measured on traceability, missing-data honesty, and trajectory behavior.
The current metrics are software robustness metrics from controlled fixtures. They are not clinical validation.
61
backend tests
10
golden cases
100%
supported claims
12/12
adversarial pass
Clinician-review only