Affiliations: [a]
School of Information and Software Engineering, University of Electronic Science and Technology of China, Chengdu, P.R. China | [b]
Department of Computer Science and Informatics, University of Energy and Natural Resources, Sunyani, Ghana
| [c]
Department of Computer Science, Faculty of Mathematical Sciences, CK Tedam University of Technology and Applied Sciences, Navrongo, Ghana
Correspondence:
[*]
Corresponding author. Yongbin Yu, School of Information and Software Engineering, University of Electronic Science and Technology of China, Chengdu, P.R. China. E-mail: ybyu@uestc.edu.cn.
Abstract: Colorectal cancer is the third most diagnosed malignancy in the world. Polyps (either malignant or benign) are the primary cause of colorectal cancer. However, the diagnosis is susceptive to human error, less effective, and falls below recommended levels in routine clinical procedures. In this paper, a Capsule network enhanced with radon transforms for feature extraction is proposed to improve the feasibility of colorectal cancer recognition. The contribution of this paper lies in the incorporation of the radon transforms in the proposed model to improve the detection of polyps by performing efficient extraction of tomographic features. When trained and tested with the polyp dataset, the proposed model achieved an overall average recognition accuracy of 94.02%, AUC of 97%, and an average precision of 96%. In addition, a posthoc analysis of the results exhibited superior feature extraction capabilities comparable to the state-of-the-art and can contribute to the field of explainable artificial intelligence. The proposed method has a considerable potential to be adopted in clinical trials to eliminate the problems associated with the human diagnosis of colorectal cancer.
