Abstract: To address the challenge of balancing segmentation accuracy and complexity in existing segmentation methods, a novel lightweight colorectal polyp image segmentation network, MCANet(Mamba and convolutional attention network), is proposed. The core of this network lies in a multiscale convolutional attention module cascading spatial and channel attention by fusing multiscale features in order to reduce the gap between shallow and deep features. In addition, a parallel Mamba module is introduced to improve the computational efficiency by utilizing parallelized computation. Experimental results on three public datasets show that the proposed method outperforms other state-of-the-art methods in terms of effectiveness and generalization, enabling it to pinpoint the abnormal portions in the colorectum and provide clinicians with critical decision support, which reduces the risk of polyp cancer.

Key words: lightweight network, polyp segmentation, deep learning, Mamba, multi-scale attention module