CoroYOLO: a novel colorectal cancer detection method based on the Mamba framework

Colorectal cancer (CRC) is one of the most common malignant tumors worldwide, and early detection is crucial for improving cure rates. In recent years, object detection methods based on convolutional neural networks (CNNs) and transformers have made significant progress in medical image analysis. However, CNNs have limitations in capturing global contextual information, and while transformers can handle long-range dependencies, their high computational complexity limits their efficiency in practical applications. To address these issues, this paper proposes a novel object detection model—CoroYOLO. CoroYOLO builds upon the YOLOv10 architecture by incorporating the concept of State Space Model (SSM) and introduces the TSMamblock module, which dynamically models the input data, reduces redundant computations, and improves both computational efficiency and detection accuracy. Additionally, CoroYOLO integrates the Efficient Multi-Scale Attention (EMA) mechanism, which adaptively strengthens focus on critical regions, enhancing the model’s robustness in complex medical images. Experimental results show that after training on the SUN Polyp and PICCOLO datasets, CoroYOLO outperforms existing mainstream methods on the Etis-Larib dataset, achieving state-of-the-art performance and demonstrating the model’s effectiveness for early colorectal cancer detection.