Go to MIDL 2026 Conference homepageMultiPersistence Topological Fusion with Vision Transformers for Skin Cancer Detection
Keywords: Skin lesion classification, dermoscopy, medical image analysis, cubical multiparameter persistence, topological data analysis, vision transformer, supervised contrastive learning, topology-aware fusion
TL;DR: TopoCon-MP fuses multipersistence topological features with Vision Transformers, improving skin lesion classification and cross-dataset transfer over strong CNN/ViT baselines while providing interpretable lesion structure cues.
Abstract: Skin cancer is a common and potentially fatal disease where early detection is crucial, especially for melanoma. Current deep learning systems classify skin lesions well, but they primarily rely on appearance cues and may miss deeper structural patterns in lesions. We present TopoCon-MP, a method that extracts multiparameter topological signatures from dermoscopic images to capture multiscale lesion structure, and fuses these signatures with Vision Transformers using a supervised contrastive objective. Across three public datasets, TopoCon-MP improves in-distribution performance over strong pretrained CNN and ViT baselines, and in cross-dataset transfer, it maintains competitive performance. Ablations show that both multiparameter topology and contrastive fusion contribute to these gains. The resulting topological channels also provide an interpretable view of lesion organization that aligns with clinically meaningful structures. Overall, TopoCon-MP demonstrates that multipersistence-based topology can serve as a complementary modality for more robust skin cancer detection.
Primary Subject Area: Application: Dermatology
Secondary Subject Area: Geometric Deep Learning
Registration Requirement: Yes
Reproducibility: https://github.com/sayoni-c98/MIDL2026-TopoConMP
Visa & Travel: Yes
Read CFP & Author Instructions: Yes
Originality Policy: Yes
Single-blind & Not Under Review Elsewhere: Yes
LLM Policy: Yes
Submission Number: 170
Loading