Internal clinical tool — pathologist workstation

AI-Assisted
Breast Cancer Detection

Four-modality AI detection system for pathologists and radiologists. Mammography, histopathology, dermoscopy, and blood biomarkers — all in one workstation.

Open Workstation Sign In

Four Detection Modalities

Each powered by a specialized AI model with GradCAM visualization

Mammography

DICOM 2D/3D mammogram analysis with BI-RADS classification, GradCAM heatmaps, and microcalcification detection.

BI-RADS 1–6

Histopathology

H&E slide analysis identifies IDC, ILC, DCIS, and benign tissue types using UNI pathology foundation model with Macenko stain normalization.

IDC · ILC · DCIS

Dermoscopy

Skin manifestation detection for Paget's disease, inflammatory breast cancer signs, and ABCDE dermoscopy criteria scoring.

ABCDE scoring

Biomarkers

Breast cancer molecular subtype prediction from CA-153, HER2, BRCA1/2, CEA, CA-125, and receptor status.

Luminal A/B · HER2 · TNBC

Built for clinical workflows

Not just a model — a complete pathologist workstation with case management, sign-off workflows, PDF reports, and a HIPAA audit trail baked in.

  • GradCAM heatmap overlays on all image modalities
  • Pathologist sign-off workflow with audit trail
  • HIPAA-aware — no PHI stored, de-identified case refs only
  • PDF clinical report generation
  • Case management — group mammography + biopsy + labs
  • Molecular subtype classification (St Gallen criteria)
  • BRCA variant risk flagging
  • Microcalcification cluster detection

Technology Stack

Mammography modelVinDr-Mammo DenseNet121
Histopathology modelUNI (ViT-Large) / BreaKHis
Dermoscopy modelEfficientNet-B4 (ISIC 2020)
Biomarker modelXGBoost (St Gallen criteria)
HeatmapsGradCAM + morphological
Stain normalizationMacenko H&E normalization
Inference serverFastAPI + PyTorch 2.3
FrontendNext.js 14 + Tailwind
DatabasePostgreSQL 16 + pgvector
HostingRender (CPU/GPU) + Vercel

HIPAA-Aware Design

This is an internal clinical tool. No PHI is stored — cases use de-identified reference codes only. All detection actions are recorded in an audit log. Uploaded images are processed in memory and not persisted beyond the session unless explicitly saved. All data transfers use TLS 1.3.

AI findings must be reviewed and confirmed by a licensed pathologist or physician before any clinical action. This tool is not FDA-cleared for independent diagnostic use.

Ready to start?

Access the pathologist workstation

Open Workstation