Edge Information-Augmented Auxiliary Diagnosis Method for Cervical Cancer in Medical Decision-Making Systems

Abstract

Cervical cancer is a highly prevalent gynecological malignancy, and its secondary reproductive complications are one of the main reasons for the decline in China's fertility rate. Early screening for Cervical intraepithelial neoplasia (CIN) is a key component of cervical cancer diagnosis and treatment. However, existing models for simultaneously addressing nucleolus identification and lesion prediction suffer from insufficient utilization of multiscale features and inefficient feature sharing between tasks. To address this, this study proposes an edge-enhanced intelligent cervical cancer screening (EICCS) method, which achieves feature reuse and improves efficiency through joint optimization of nucleolus segmentation and lesion classification. First, a nucleolus segmentation model based on edge feature enhancement is designed. This model enhances cell edge perception through edge attention and optimizes feature fusion through adaptive weighting, improving segmentation accuracy in complex scenarios. Second, a classification network that integrates segmentation features is constructed. This model uses a convolutional attention mechanism to dynamically optimize multiscale, fine-grained features to prioritize the identification of key pathological areas. Results demonstrate that EICCS outperforms existing methods in both segmentation and classification tasks.