Multi-Label classification of multi-modality skin lesion via hyper-connected convolutional neural network

作者：

Highlights：

• A new multi-modality multi-label skin lesion classification method based on hyper-connected convolutional neural network.

• A hyper-branch enables fusion of multi-modality image features in various forms.

• A hyper-connected module helps to iteratively propagate multi-modality image features across multiple correlated image feature scales.

• A multi-scale attention block ensures the network can prioritize the semantically more important regions.

• Our method can achieve consistent classification results in datasets with imbalanced label distributions.

摘要

•A new multi-modality multi-label skin lesion classification method based on hyper-connected convolutional neural network.•A hyper-branch enables fusion of multi-modality image features in various forms.•A hyper-connected module helps to iteratively propagate multi-modality image features across multiple correlated image feature scales.•A multi-scale attention block ensures the network can prioritize the semantically more important regions.•Our method can achieve consistent classification results in datasets with imbalanced label distributions.

论文关键词：Classification,Melanoma,Convolutional neural networks (cnns)

论文评审过程：Received 13 November 2019, Revised 20 May 2020, Accepted 12 June 2020, Available online 18 June 2020, Version of Record 22 June 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107502