Research on Mass Detection in Mammographic Images Based on an Improved YOLO Network Model

Breast cancer is one of the most common malignant tumors among women worldwide, and early diagnosis is crucial for improving patient survival rates. This study proposes an improved YOLOv8-based mass detection model for mammographic images, referred to as Mass Detection YOLOv8 (MD-YOLOv8). To address the limitations of existing detection models—such as insufficient feature extraction and a high miss rate for small targets—this work integrates the Convolutional Block Attention Module (CBAM) and the Content-Aware ReAssembly of Features (CARAFE) into the original YOLOv8 architecture to enhance its feature representation and small mass detection capabilities. Specifically, CBAM is introduced before the Upsample and CBS layers in the Neck of the network to adaptively recalibrate the feature maps along channel and spatial dimensions, thereby improving the model’s sensitivity to fine-grained details and enhancing its focus on mass regions in mammographic images. In addition, the original upsampling module of YOLOv8 is replaced with the CARAFE module, which strengthens multi-scale feature fusion and optimizes feature reconstruction through a content-aware upsampling strategy, ultimately improving the model’s ability to detect masses. The proposed model is evaluated on the public CBIS-DDSM mammographic image dataset, achieving a mean Average Precision ([email protected]) of 0.491, which represents a 3.2% improvement over the baseline YOLOv8 model.