Evaluation Metrics
AUROC evaluates sensitivity across all decision thresholds, avoiding threshold-tuning bias. Quadratic weighted kappa accounts for ordinal distances between diabetic retinopathy severity stages (0–4).
RESEARCH INTEGRITY & SCIENTIFIC RIGOREvidence and limits.
Evidence and limits.
In clinical context.
Review reported results, evaluation context, and the limits of each model.
MODEL EVALUATION REGISTRY
Reported benchmark metrics.
Team-reported metrics from the September 2026 model overview. See each model card for source limitations and evaluation scope.
| Model | Family | Evaluation Metric | Reported Performance | Clinical Status |
|---|---|---|---|---|
| Wrist Fracture Detection | Prism | Overall diagnosis AUROC | 0.974 | PREVIEW |
| ChestVision | Prism | Mean AUROC across 14 conditions | 0.805 | PREVIEW |
| SightDx | Lumen | Quadratic weighted kappa | 0.8965 | PILOT |
| BoneScan X | Prism | Listed AUROC · attribution pending | 0.929 | PREVIEW |
| TB Audio Screening | Aperture | Team-reported accuracy | ~84% | RESEARCH |
AUROC and kappa measure different tasks. These results are not directly comparable.
Sachdeva Netralaya Validation
SightDx underwent multi-reader clinical validation at Sachdeva Netralaya, Jammu, evaluating agreement against veteran retinal specialists on optical fundus photography.
Stanford MURA Disclosures
BoneScan X was trained utilizing Stanford's MURA musculoskeletal dataset, matching its published 0.929 DenseNet baseline. Vessel Vision distinguishes open-source baselines from proprietary models.
Original Stanford MURA Research Paper ↗SYSTEM SAFETY PROTOCOL
Scope. Uncertainty. Review.
Three research priorities for model development and future healthcare deployment.
Explicit scope definitions
Each model has strict input eligibility criteria. Out-of-distribution scans trigger automatic rejection and physician handoff.
Calibrated confidence
Our models output calibrated Bayesian uncertainty metrics alongside findings, highlighting ambiguous pathology for senior review.
Concept heatmaps
Pixel-level attention heatmaps attribute diagnostic predictions to concrete anatomical abnormalities (fracture lines, hemorrhages).