Cervical Cancer Lesion Grading Based on Clinical Diagnostic Criteria and Multi-View Approach Using Intraoperative Colposcopic Images

Current research on cervical cancer grading often ignores the integration of clinical diagnostic practice, and effectively fusing multiple views of the lesion site is crucial for accurate analysis. In order to solve this problem, this paper proposes a multi-view classification network using acetic acid-stained colposcopy to screen intraoperative images. This model introduces a novel multi-linkage attention mechanism based on sliding Windows to enhance feature fusion across multiple views. Experimental results show that compared with Swin-Transformer, Vision Transformer and MobileNetV2(all without pre-training weights), the accuracy of the proposed model is improved by 2.60%, 9.68% and 14.43%, respectively. These results highlight the superior performance of the model in grading cervical cancer lesions and its potential as a valuable tool for clinical diagnosis.