Demystifying Medical AI with Prototypical Reasoning.
"Odis was developed to bridge the gap between high-performance neural networks and the expert intuition of diagnostic dermatologists."
Beyond the Black Box
Standard Deep Learning models in dermatology are often accurate but utterly uninterpretable. A doctor cannot see *why* a model labeled a lesion as malignant. This lack of transparency is a barrier to diagnostic trust.
Explainable Reasoning
The system justifies every diagnosis by aligning visual features with a curated reference database, mirrors specialist intuition.
Research Grounded
Based on the "This Looks Like That" (ProtoPNet) architecture published by Duke University scholars.
Legacy AI
Black Box Uncertainty
Odis
Transparent Justification
Training Data
We leverage the HAM10000 dataset, containing over 10,000 dermatoscopic images, ensuring our AI learns from a globally recognized gold-standard reference corpus.
The "ProtoPNet"
Our model uses 105 distinct "Neural Prototypes" (15 per each of the 7 classes) that represent common disease features.
Showing 15 prototypes per class
Human-In-The-Loop
The diagnostic specialist remains the final authority. Our system is designed for Decision Support, providing the "Why" so the doctor can make the final "What".
The Processing Pipeline
Input
Raw Sample
EfficientNet-B4
Feature Extraction
ProtoPNet
Prototypical Alignment
Clinical Inference
Evidence Synthesis
Experience Explainability Today.
Create an account to begin using our ProtoPNet-based diagnostic workstation.
Initialize Portal Access