Dynamically aggregating MLPs and CNNs for skin lesion segmentation with geometry regularization

• Published in: Computer methods and programs in biomedicine 2023-08, Vol.238, p.107601-107601, Article 107601
• Main Authors: Qin, Chuanbo, Zheng, Bin, Zeng, Junying, Chen, Zhuyuan, Zhai, Yikui, Genovese, Angelo, Piuri, Vincenzo, Scotti, Fabio
• Format: Article
• Language: English
• Subjects: Feature interaction; Geometric information; Multi-layer perceptions; Skin lesion segmentation
• ISSN: 0169-2607; 1872-7565
• DOI: 10.1016/j.cmpb.2023.107601

•A novel encoder combines CNNs branch and MLPs branch in parallel, which is conducive to the extraction of rich local features while simultaneously building global-spatial-dependencies and global-channel-dependencies. Furthermore, a new feature-interaction module is applied to refine the features of MLPs branch and CNNs branch by allowing dynamic exchange of spatial and channel information.•The introduction of auxiliary loss on SDM predictions is conducted to learn the position and shape information of the skin lesions, further highlighting the boundaries of segmentation predictions.•Comprehensive experiments show that CFF-Net achieves state-of-the-art segmentation performance in four public skin lesion datasets (namely ISIC 2016, ISIC 2017, ISIC 2018, PH2), while having comparable parameters. Additionally, we adopt comparison experiments on three public datasets to demonstrate the applicability of CFF-Net for other segmentation tasks. Melanoma is a highly malignant skin tumor. Accurate segmentation of skin lesions from dermoscopy images is pivotal for computer-aided diagnosis of melanoma. However, blurred lesion boundaries, variable lesion shapes, and other interference factors pose a challenge in this regard. This work proposes a novel framework called CFF-Net (Cross Feature Fusion Network) for supervised skin lesion segmentation. The encoder of the network includes dual branches, where the CNNs branch aims to extract rich local features while MLPs branch is used to establish both the global-spatial-dependencies and global-channel-dependencies for precise delineation of skin lesions. Besides, a feature-interaction module between two branches is designed for strengthening the feature representation by allowing dynamic exchange of spatial and channel information, so as to retain more spatial details and inhibit irrelevant noise. Moreover, an auxiliary prediction task is introduced to learn the global geometric information, highlighting the boundary of the skin lesion. Comprehensive experiments using four publicly available skin lesion datasets (i.e., ISIC 2018, ISIC 2017, ISIC 2016, and PH2) indicated that CFF-Net outperformed the state-of-the-art models. In particular, CFF-Net greatly increased the average Jaccard Index score from 79.71% to 81.86% in ISIC 2018, from 78.03% to 80.21% in ISIC 2017, from 82.58% to 85.38% in ISIC 2016, and from 84.18% to 89.71% in PH2 compared with U-Net. Ablation studies demonstrated the effectiveness of each proposed compon