mCA-Net: modified comprehensive attention convolutional neural network for skin lesion segmentation

Skin is the first line of defense of the human body. Because the skin is exposed to the outside and suffersvarious aggressions , Skin cancer is the most common cancer. Accurate skin lesions image segmentation is essential for skin disease diagnosis and treatment planning. In order toimprove the segmentation results of the recently proposed comprehensive attention convolutional neural network(CA-Net) for skin lesions image segmentation, In this work, we propose a modified medical image segmentation network-modified comprehensive attention convolutional neural network (mCA-Net) to further improve segmentation performance. In particular, we create a new multi-scale channel attention module-MS-CA, which can display more accurate and relevant feature channels on multiple scales. The experiments shows that our work greatly improve the average segmentation Dice score, accuracy, mean ASSD and mloU and enhance the stability of the segmentation model. Through comprehensive experiments on the ISIC 2018 skin lesion datasets, it is found that our proposed mCA-Netnetwork compared with CA-Net,improve the average segmentation Dice score from 92.08% to 93.56%, the average accuracy score of skin lesions from 92.68% to 93.32% and the mloU from 85.32% increased to 87.89%. The segmentation results have been significantly optimized.