Multi-level GAN based enhanced CT scans for liver cancer diagnosis. (March 2023)

• Record Type: Journal Article
• Authors: Khan, Rayyan Azam; Luo, Yigang; Wu, Fang-Xiang
• Is Part Of: Biomedical signal processing and control. Volume 81(2023)
• Publication Date: 2023-03
• Subjects: Liver cancer -- Image enhancement -- Multi-level generative adversarial network -- CT scan -- Computer-aided diagnosis; 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.2022.104450
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Liver cancer diagnosis requires preprocessing of images with preserved structural details. In this study, a multi-level generative adversarial network (GAN) is proposed to enhance computed tomographic (CT) images. The generated enhanced images are used to perform computer-aided diagnosis between malignant and normal liver. Three publicly available datasets, Ircadb, Sliver07, and LiTS, are used to investigate the performance of the proposed method with qualitative and quantitative analysis, namely performance metrics and computer-aided diagnosis. The mean structure similarity index of 0.45 and peak signal-to-noise ratio of 16.20 dB is achieved for the metric analysis. The AlexNet is adopted to perform binary classification with the testing accuracy of 90.37% and 85.90% for enhanced and non-enhanced images, respectively, which demonstrates the effectiveness of the proposed multi-level GAN in producing enhanced biomedical images with preserved structural details and favorable reduction in artifacts. Moreover, the consistently better performance among three datasets confirms the merits of the proposed multi-level GAN for computer-aided diagnosis. Graphical abstract: Highlights: We utilize multi-level GAN to enhance low-contrast CT scanned images. We evaluate multi-level GAN on three datasets with qualitative and quantitative analysis. We reproduce several state-of-the-art algorithms for fair comparative analysis.