YOLO-CPC: a breast tumor detection and identification algorithm based on improved YOLOv7

• Published in: Signal, image and video processing Vol. 19; no. 3; p. 260
• Main Authors: Du, Yanan, Liu, Wanmin, Wang, Yixuan, Li, Rutian, Xie, Li
• Format: Journal Article
• Language: English
• Published: London Springer London 01.03.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Architecture; Breast cancer; Clinical medicine; Computer Imaging; Computer Science; Datasets; Deep learning; Identification; Image Processing and Computer Vision; Mammography; Medical diagnosis; Medical personnel; Medical screening; Modules; Multimedia Information Systems; Neural networks; Object recognition; Original Paper; Pattern Recognition and Graphics; Signal,Image and Speech Processing; Target detection; Tumors; Ultrasonic imaging; Vision; Deep learning; Machine vision; Breast tumor; YOLOv7; Object detection
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-024-03811-z

Intelligent detection of breast tumors is crucial for enhancing the accuracy and efficiency of diagnosis. However, current target detection algorithms face challenges such as missed and false detections, limiting their utility in clinical settings. To address this issue, this study introduces a novel breast tumor target detection and in-stance segmentation recognition algorithm called YOLO-CPC, which is built upon enhancements to the YOLOv7 algorithm. By incorporating the CBAM attention mechanism module to mitigate background interference, utilizing PConv in the ELAN module to reduce redundant computations and memory usage, and replacing the 1 × 1 convolutional layer in PA-FPN with coordinate convolution to enhance positional aware-ness and mask robustness, the algorithm improves the extraction of tumor features. Additionally, the enhanced CA-FPN structure integrates geometric and semantic in-formation through residual linkage, further enhancing tumor detection capabilities. Experimental results show that the enhanced algorithm achieves detection accuracy, recall, and average precision rates of 97.01%, 97.98%, and 90.78%, respectively, representing improvements of 4.31, 6.41, and 1.55 percentage points over the YOLOv7 base-line. Moreover, compared to Faster R-CNN, YOLOv3, YOLOv5, and YOLOv6 algorithms, the average precision is enhanced by 10.42, 12.32, 2.28, and 7.76 percentage points, respectively. The enhanced model reduces instances of missed and false detections, improves target localization precision, and provides a theoretical foundation for intelligent breast tumor detection and practical clinical implementation.