摘要
Precise segmentation for skin cancer lesions at different stages is conducive to early detection and further treatment. We propose a weakly supervised semantic segmentation algorithm (CNN-SRR) for dermoscopy images through CNN responding superpixel regions, given that the substantial cost of obtaining perfect pixel annotation for these tasks. CNN-SRR combines a modified classifier based on deep learning and unsupervised superpixel algorithm. The former leverages abundant image-level labeled data to tune parameters to focalize on lesion regions. The extraction of lesion region responses consists of two stages, training a modified CNN classifier and back-propagate peak values of the classifier top layer. Afterward, a test image is over-segmented to a set of primitive superpixels that are merged into a few regions as proposals, several of which are activated as the segmented mask by lesion region responses via non-maximal suppression. Quantified experiments on ISBI2017 and PH2 datasets prove that the proposed algorithm can effectively discriminate lesion regions and the segmentation results even achieve competitive accuracy to the supervised segmentation approaches. We evaluate the proposed CNN-SRR algorithm on ISBI2017 and achieve that the Jaccard coefficient and Accuracy of segmentation task are improved by 12.4% and 3.3% compared with the unsupervised superpixel segmentation algorithm.
学科与关键词
- Semi-Supervised Skin Lesion Segmentation With Coupling CNN and Transformer FeaturesAlahmadi, Mohammad D. D.*; Alghamdi, WajdiIEEE Access, 2022, 10: 122560-122569.被引 9
- A hybrid CNN architecture for skin lesion classification using deep learningJasil, S. P. Godlin*; Ulagamuthalvi, V.SOFT COMPUTING, 2023.被引 8
- Skin Cancer Classification With Deep Learning: A Systematic ReviewWu, Yinhao; Chen, Bin; Zeng, An; Pan, Dan; Wang, Ruixuan; Zhao, Shen*Frontiers in Oncology, 2022, 12: 893972.被引 32
