A Breast Cancer Detection Model Based on Modified Convnext v2
Senhao Cheng, Esther Sun, W Qian, Yang Han · 2023
Breast cancer constitutes a prevalent global health challenge, characterized by a substantial incidence and profound consequences. Computer-Aided Diagnosis (CAD) stands as a paramount asset in enhancing mammography-based detection methods. CAD empowers healthcare practitioners in facilitating early breast cancer diagnosis and contributes to the advancement of diagnostic capabilities in the field of oncology. In this paper, we propose Convnext v2 as the backbone model and combines with Generalized-Mean Pooling layer for feature extracting. We use the cross-entropy as the loss function and using AdaBeliefz for better optimization. The evaluation metrics is pF1. The result shows that ConvNextV2 owns the best performance on pF1, which is 0.040, 0.043, 0.031 higher than ResNet50, GoogleNet50, EfficientNet B2 respectively.