Neuro-Symbolic Skin Cancer Classifier

Problem: Conventional deep learning convolutional neural networks in dermatological imaging frequently overfit to visual artifacts, suffer from high false-positive rates, and cannot explain their diagnostic reasoning in terms recognized by clinical dermatologists. Solution: This system implements a neuro-symbolic hybrid architecture that combines deep convolutional neural networks for feature extraction with explicit symbolic logic rules grounded in clinical ABCDE (Asymmetry, Border, Color, Diameter, Evolution) criteria. By constraining neural representations with symbolic clinical rules, the model achieves transparent, auditable classifications that align with medical standards. Key Engineering Highlights: • Dual-stream pipeline: CNN feature representation paired with symbolic rule validation. • Dermoscopic lesion boundary extraction and morphological symmetry scoring. • Generation of structured diagnostic reports explaining rule matches and confidence bounds. • Significantly reduced hallucination and artifact vulnerability compared to unconstrained deep models.