I-UNeXt: A Skin Lesion Segmentation Network Based on Inception and UNeXt

Abstract

Segmentation of skin lesions from dermoscopic images is very important for clinical diagnosis and treatment planning. In order to segment skin lesions quickly and effectively, the segmentation network I-UNeXt was proposed in this paper. I-UNeXt is to add the Inception module to UNeXt. Compared with UNeXt's original ordinary convolution module, the Inception module added enhances the feature extraction capability of UNeXt by using different convolution kernels to extract information of different scales. At the same time, dilated convolution is introduced into the original Inception module, which reduces the amount of computation in the module while maintaining the original receptive field of convolution. We used the ISIC2017 dataset to train and test the segmentation performance of I-UNeXt. The experimental results show that F1-score, IOU and DICE are 81.95%, 71.10% and 82.46%, respectively. The overall performance of the network is better than that of other most advanced networks. Experiments show that the I-UNeXt network proposed in this paper can effectively segment the skin lesions and provide help for the diagnosis of modern skin diseases.