AS-UNet: An Efficient Attention based Skin Lesion Segmentation Model

Skin Cancer is the most widely found disease that may leads to poor prophetic responses as well as survival rates. Melanoma is one of the types of skin cancer that achieves the cure rate higher by surgical treatment when it is detected earlier. The early diagnosis of this type of cancer depends on the manual segmentation of alleged lesions but manual segmentation may leads to misclassification. In order to overcome these issues, there is a need of automatic image segmentation technique. In this paper, a modified UNet algorithm termed as AS-Unet is proposed to hasten the segmentation process therefore it provides the precise outcomes in skin lesion segmentation. In the experimental process, ISIC 2017 datasets is used to compare the performance of the proposed model with the recently existing models and is shown the improved performance. In the International Skin Imaging Collaboration dataset, AS-UNet obtained a 92.6% of Dice coefficient 96.4 % of IoU and 96.8% accuracy than other models.