Precision Detection of Gastric Cancer Using MCSCN

Gastric cancer continues to impose a substantial global health burden, making timely identification crucial for improving clinical prognosis. Although gastroscopy is the primary diagnostic tool, accurate identification of early-stage lesions requires substantial clinical expertise and remains prone to misdiagnosis due to subtle visual features. This paper proposes the Multi-Scale Channel Soft-Thresholding Cascade Network (MCSCNet) for automated early gastric cancer detection from gastroscopic images. MCSCNet extends Cascade R-CNN by incorporating a Swin Transformer backbone, MSCA-based channel enhancement, and BiwFPN-driven bidirectional feature fusion. Comprehensive ablation experiments demonstrate the effectiveness of each component, and comparative results show that MCSCNet outperforms benchmark methods in precision, F1 score, and accuracy. Its generalizability is further validated on both proprietary and public datasets. These results indicate that MCSCNet has potential to improve diagnostic accuracy and reduce missed diagnoses in clinical practice.