FMDDU-Net: An Effective and Efficient Deep Learning Method for Automatic Colorectal Polyp Detection

Colorectal polyps are one of the key indicators in the early detection of colorectal cancer. Deep learning-based automatic detection has emerged as a promising approach to enhance accuracy and efficiency in polyp identification. In this study, we propose the Factorized Multiscale Double Decoder UNet (FMDDU-Net), a novel deep learning method designed to automatically detect colorectal polyps. FMDDU-Net integrates convolution factorization, multi-scale, attention, and residual connection mechanisms to maximize the representation of spatial and textural features. Experiments were conducted using the CVC-ClinicDB public dataset. The evaluation results show that FMDDU-Net achieves higher accuracy than previous methods, with an average specificity of 99.61%, sensitivity of 94.29 %, accuracy of 99.13 %, F1-Score of 94.90%, and Jaccard Index (IoU) of 90.53%. Additionally, the number of FMDDUNet parameters is relatively small, at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{4 5 9, 0 9 1}$</tex>. With its superior performance in accuracy and efficiency, FMDDU-Net demonstrates great potential for application in clinical decision support systems, particularly in assisting doctors in diagnosing colorectal polyps more quickly and accurately. This research lays the groundwork for the further development of lightweight and robust deep learning technologies in the medical field.