AI-assisted detection
Imaging models designed for mammography and breast imaging workflows surface suspicious findings and reduce missed signals.
Oncology diagnostic support
AI-assisted breast cancer imaging, with the clinician in command
Lotus Health is LOTUS CHAIN Hub's clinical product for oncology: AI-assisted breast cancer detection and diagnostic support workflows that run on the AlPa engine's local inference runtime — every finding reviewed, overridable, and audited before it reaches a report.
Product overview
Lotus Health combines imaging-focused AI with structured clinical reasoning to help identify suspicious patterns, guide review, and improve workflow efficiency in breast cancer pathways. Models surface candidate findings; clinicians confirm, modify, or override them — and every decision is written to an append-only audit trail. The long-term opportunity lies in pairing technical accuracy with deployable, trusted healthcare experiences.
Clinical decision-support capabilities
Clinical decision support
Structured outputs — findings, confidence, comparisons, and follow-up suggestions — help professionals review and act with more context, not black-box scores.
DICOMweb & FHIR integration
Connects to hospital PACS via DICOMweb (STOW-RS / QIDO-RS / WADO-RS) and exchanges structured results through FHIR-based observation and diagnostic report resources.
Clinician override interface
An un-bypassable review step: clinicians explicitly confirm, modify, or override every AI finding before anything enters the clinical record.
On-device processing
Inference executes locally through the AlPa engine (WebGPU/WASM with Dawn on native builds) — imaging data never leaves the trust boundary.
Append-only audit graph
Every model version, override, and approval is immutably recorded — a forensic trail built for clinical governance and regulatory review.
Interactive workflow demo
Walk the clinical pathway
Step through a typical AI-assisted breast cancer imaging workflow to see where the model assists — and where the clinician always has the final say.
- 01System
Ingest & de-identify
A mammography study arrives via DICOMweb. Header metadata is scrubbed against institutional privacy templates and burned-in annotations are masked before anything is processed.
- 02System
Local AI screening
The AlPa runtime executes the detection model on-device (WebGPU on the clinician's workstation, or Dawn-backed native on the edge server). Candidate findings are returned with confidence and heatmaps.
- 03Clinician
Clinician review
The radiologist reviews each candidate finding against the imaging evidence, accepting, annotating, or rejecting the model's suggestion. Nothing moves forward without this step.
The clinician is always the final decision-maker — the model never signs a report.
- 04Clinician
Override & sign-off
The clinician's final determination — including any override of the AI position — is captured verbatim with attribution and reasoning.
- 05System
Structured report & audit
A FHIR diagnostic report is drafted with the agreed findings, watermarked with pipeline version, model manifest IDs, and the reviewing clinician's approval — then appended to the audit graph.
Powered by AlPa
On-device processing, measured
Because Lotus Health runs on the AlPa engine, inference happens where the images already are. Representative figures from local deployments:
- 0
- images leaving the trust boundary during inference
- ~90%+
- cloud compute cost reduction vs server-side GPU
- 100%
- of findings pass through clinician review
- 100%
- of decisions written to the append-only audit graph
Real-world deployment
Built for clinical reality
Lotus Health is designed around the operational realities of breast imaging practice — integration with existing systems, measurable review burden reduction, and governance from day one.
Screening workflow integration
Deployment alongside existing mammography reading workflows via DICOMweb connectivity — no rip-and-replace of the hospital's PACS or RIS.
Radiologist-in-the-loop validation
Continuous agreement tracking between clinician decisions and model suggestions supports drift detection and evidence generation for clinical governance.
Governed data handling
Fail-closed de-identification gates ensure privacy anomalies halt processing rather than risking exposure — aligned with HIPAA expectations.