lesion-detection

---

name: lesion-detection description: Classifies skin conditions including Melanoma and Basal Cell Carcinoma using TF.js MobileNetV3 license: MIT compatibility: opencode metadata: audience: developers workflow: clinical-pipeline

What I do

I classify skin conditions using TensorFlow.js with MobileNetV3. I identify patterns consistent with various skin conditions including Melanoma, Basal Cell Carcinoma (BCC), and other dermatoses. I return confidence scores for each classification.

When to use me

Use this when:

• Feature extraction is complete and you need lesion classification
• You need confidence scores for risk assessment
• You're identifying potential areas of concern in skin images

Key Concepts

• MobileNetV3: TF.js model for skin condition classification
• Confidence Score: 0-1 probability for each condition
• Lesion Types: Melanoma, BCC, Actinic Keratosis, etc.
• lesions_detected: State flag after detection complete

Source Files

• services/vision.ts: Lesion detection implementation
• types.ts: AnalysisResult interface with lesions array

Code Patterns

• Load MobileNetV3 model via TF.js
• Run inference on feature vectors
• Return array of detected lesions with confidence and risk levels

Operational Constraints

• MUST use tf.tidy() or explicit dispose() for all tensors
• Confidence scores required for all detections
• High-risk detections trigger elevated scrutiny in downstream