Large Kenel Multi-Scale Contrast Former for CIN Grades Prediction

Cervical cancer is the second most common cancer in women worldwide. Colposcopy is an integral part of cervical intraepithelial neoplasia (CIN) and cervical cancer screening, but it has a high misdiagnosis rate [1]. In this paper, a multi-scale feature contrast transformer based on colposcopic images was constructed to classify different CIN grades. We propose a multi-scale comparator based on pixel position information, which extracts and compares features of two different scales, and learns features of different scales to distinguish images more accurately. Experimental results show that the classification accuracy of the proposed model framework reaches 96.61%.