Diagnomass Derma-Vision AI · v2.5
SYSTEM ONLINE · MODEL READY
AUC 0.9941
99.41%
Validation AUC
96.59%
F1-Score Accuracy
97.80%
Clinical Sensitivity
B7×2
EfficientNet Ensemble
Diagnostic Input UPLOAD · METADATA
Dermoscopic Image
⬡ DRAG · DROP · OR CLICK JPG / PNG — MAX 10MB Lesion
Age
Biological Sex
Anatomical Site
Analysis Output CLASSIFICATION · CONFIDENCE
AWAITING SCAN INPUT
SYSTEM READY
FUSING DUAL-STREAM
NETWORKS
—
—
// CONFIDENCE: —
Clinical Intelligence ARCH · METRICS
Dual-Stream EfficientNet-B7

Neural Pipeline

RGB pixels fused with offline CTE Maps (Color, Texture, Edge) — natural visual patterns combined with mathematical gradients for superior lesion differentiation.

📷
INPUT
›
🧠 RGB
EXPERT B7
🔬 CTE
EXPERT B7
›
⚡ ENSEMBLE
FUSION
›
📊 TABULAR
MLP
›
🏥 OUTPUT
Balanced Anatomical Test Set

Model Performance

Validation AUC0.9941
F1-Score (Accuracy)0.9659
Clinical Recall (Sensitivity)0.9780
Spatial Feature Localization ANATOMICAL SITE PRIORS
Tabular MLP Site Weighting

Anatomical Intelligence

Malignancy probabilities shift based on anatomical origin. The Tabular MLP weighs visual inference against geographical prior probabilities encoded during training.

HEAD/NECK High UV exposure — distinct keratinocyte texture signatures and solar damage markers.
TORSO Largest surface area — nodular and superficial spreading variants most common.
EXTREMITY Arms & Legs — distinct vascular patterns; notable acral lentiginous risk.
ACRAL Palms & Soles — unique ridge patterns, higher acral lentiginous melanoma incidence.
Scan History SESSION LOG
Recent sessions stored locally in this browser.
// NO SCANS RECORDED IN THIS SESSION