Abstract
Skin lesion classification remains challenging because of class imbalance, substantial intra-class variation, and subtle differences between diagnostic categories. Paired dermoscopic and clinical images provide complementary visual evidence, but effectively leveraging widely available single-view data when learning a dual-view model remains challenging. We propose DermaNet, a unified flexible-view framework that learns a shared lesion representation from paired dermoscopic and clinical images while accommodating single-view inputs. Separate Efficient-NetV2 backbones extract representations from multiple network depths, which are combined within each view through learned weighted depth aggregation. The resulting view-specific embeddings are fused before classification, while modality dropout exposes the model to paired- and single-view conditions during training. DermaNet also uses a clinically motivated hierarchical prediction strategy, in which a group-level classifier first separates melanocytic from non-melanocytic lesions and group-specific classifiers then perform fine-grained diagnosis within each category. Progressive experiments show that multi-level aggregation, hierarchical prediction, and unified feature-level fusion provide complementary improvements and support both paired and single-view inference. DermaNet achieves 49.7% Macro-F1 on MILK10k and 77.42% Micro-F1 on the official ISIC2019 test set while increasing the parameter count by only 0.4%. An implementation is available at https://github.com/aniltanaktan/DermaNet.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ISIC_018.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=xDP0Va0g8E&nesting=2&sort=date-desc
BibTex
@InProceedings{AktAni_DermaNet_MICCAISAT2026,
author = { Aktan, Anil Tan AND Sen, Mehmet Umut AND Aptoula, Erchan AND Yanikoglu, Berrin},
title = { { DermaNet: Flexible-View Fusion for Dermoscopic–Clinical Skin Lesion Classification } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17271},
month = {pending},
page = {pending}
}