CDA-Net: A contrastive deep adversarial model for prostate cancer segmentation in MRI images. (May 2023)

• Record Type: Journal Article
• Authors: Li, Zhixun; Fang, Jiancheng; Qiu, Ruiyun; Gong, Huiling; Zhang, Wei; Li, Linghao; Jiang, Jian
• Is Part Of: Biomedical signal processing and control. Volume 83(2023)
• Publication Date: 2023-05
• Subjects: Contrastive learning -- Generative adversarial network -- Prostate cancer segmentation -- Magnetic resonance imaging; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2023.104622
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Prostate cancer is becoming one of the deadliest cancer in men, and the early diagnosis and detection of cancer can be effective in improving patient survival. Computer-aided diagnosis (CAD) has evolved rapidly in recent years, and more and more computer technologies are now being used for prostate cancer detection. Apparently, automatic and accurate segmentation is the most important step in it. However, the existing prostate cancer segmentation methods still have problems with low accuracy and efficiency, and these deficiencies prevent them from providing adequate results at the pixel level. In this paper, a prostate cancer segmentation network (CDA-Net) is proposed from the perspectives of the global cancer region localization and the local cancer edge recognition. Specifically, we propose a parallel dilated U-Net (dila-UNet) to extract deep features for more accurate localization, as well as design a connectivity mechanism of a generative adversarial network (GAN) and a contrastive learning module for finer edge recognition. Compared with some classic and state-of-the-art (SOTA) segmentation methods, the results show the segmentation performance of the proposed network is superiorly increased by ∼ 1.7%, ∼ 3.8% and ∼ 1.7% on IoU, PA and Dice, respectively, and the 95%hausdorff distance is decreased by ∼ 1.82, in MRI images. Highlights: A novel deep network is proposed for prostate cancer segmentation in MRI images. A dilated convolutions module is designed to improve the cancer localization ability. A connectivity mechanism is introduced to improve the recognition accuracy on edges. The proposed network achieves excellent segmentation results.