An Enhanced Neural Network Based on Deep Metric Learning for Skin Lesion Segmentation

Accurate segmentation of skin lesion is a very important step in computer-aided diagnosis (CAD). However, this task is full of challenges due to the significant variations of lesion appearances across different patients. Aiming at solving the inaccuracy of segmentation, we propose a deep metric learning enhanced neural network (DMLEN) to promote the lesion segmentation result from existing method. We make full use of the relationships of pixels between the lesion area and the non-lesion area, and change the problem of pixels classification into the problem of distance measurement of pixels. In addition, an iterative enhancement scheme is proposed that gradually revises the lesion segmentation and results in a better result. Experimental results on two datasets show that the DMLEN decreases MAE 3% at least than previous methods.