Learning from dermoscopic images in association with clinical metadata for skin lesion segmentation and classification

Caixia Dong, Duwei Dai, Yizhi Zhang, Chunyan Zhang, Zongfang Li, Songhua Xu

Abstract: Highlights•We propose a novel pipeline to accurately and effectively tackle skin lesion segmentation and classification tasks, which could also be applied to address other clinical applications.•A novel Multi-scale Holistic Feature Exploration (MSH) module is proposed to thoroughly exploit perceptual clues latent among multi-scale feature maps as synthesized by the decoder.•A novel Cross-Modality Collaborative Feature Exploration (CMC) module is proposed to collaboratively exploit potential relationships between cross-modal features of dermoscopic images and clinical metadata.•We conducted extensive experiments on several benchmark datasets achieving good performance.