Title: Automatic detection of rectal cancer lesion images based on the improved MR-U-Net deep learning network

Authors: Fang Wu; Yihua Gu; Haifei Zhang

Addresses: School of Yongyou Digital and Intelligence, Nantong Institute of Technology, Nantong, Jiangsu, China ' School of Yongyou Digital and Intelligence, Nantong Institute of Technology, Nantong, Jiangsu, China ' School of Information Engineering, Nantong Institute of Technology, Nantong, Jiangsu, China

Abstract: Traditional image processing methods are prone to segmentation and overlap when processing medical images of rectal cancer, which makes it difficult for algorithms to distinguish surrounding tissues, organs, and lesions accurately. In addition, the small size of the cancerous area may also lead to errors and inaccuracies in the segmentation results. This study proposed an enhanced mechanical residual U-shaped network (MR-U-Net) model for automatically detecting rectal cancer lesions in computed tomography (CT) images to address these issues. The model achieves high segmentation accuracy by combining a deep supervision mechanism with attention guidance. Experimental results showed a significant improvement in the dice coefficient and other indices compared to the baseline U-shaped network (U-Net) model. After comparing the corresponding index values, it can be seen that the designed system has achieved accurate segmentation of lesions in general. At the same time, we also hope to achieve better results by continuously improving research directions in the future.

Keywords: deep learning; medical image segmentation; MR-U-Net; attention mechanisms; In-depth monitoring mechanism.

DOI: 10.1504/IJSNET.2025.146124

International Journal of Sensor Networks, 2025 Vol.48 No.1, pp.54 - 64

Received: 20 Nov 2024
Accepted: 11 Dec 2024
Published online: 07 May 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article