Improved PAA algorithm for breast mass detection in mammograms

• Published in: Computer methods and programs in biomedicine 2024-06, Vol.251, p.108211-108211, Article 108211
• Main Authors: Liu, Weixiang, Zeng, Pengcheng, Jiang, Jiale, Chen, Jingyang, Chen, Linghao, Hu, Chuting, Jian, Wenjing, Diao, Xianfen, Wang, Xianming
• Format: Article
• Language: English
• Subjects: Algorithms; Breast - diagnostic imaging; Breast cancer; Breast Neoplasms - diagnostic imaging; Databases, Factual; Deep Learning; Early Detection of Cancer - methods; False Positive Reactions; Female; Humans; Mammography - methods; Mass detection; Probability; Probability anchor assignment algorithm; Radiographic Image Interpretation, Computer-Assisted - methods
• ISSN: 0169-2607; 1872-7565
• DOI: 10.1016/j.cmpb.2024.108211

•In the research related to breast mammography nodule detection, we are the first to have made targeted improvements and enhancements to the backbone network of PAA, the FPN fusion module, and the dense detection head, specifically based on the characteristics of breast mammography.•We selected the EfficientNet-B3 network as the PAA backbone network to improve the feature extraction capabilities while minimizing the number of network parameters required.•Addition of attention mechanism to FPN, aiming to enhance the representation power of the network by integrating multiple features while selectively attending to important information and ignoring irrelevant information.•To enhance the spatial alignment capabilities and improve the interaction between classification and localization tasks, we introduced an efficient task alignment module (TAM) in the task alignment prediction head.•A comparative experiment was conducted on the improved PAA network, and the results demonstrated a higher accuracy and sensitivity in mass identification, validating the effectiveness of the proposed improvement. And compared to other research, our experimental results reveal that we manage to uphold a high true positive rates (TPR) and at the same time, keep the false positives per image (FPPI) relatively lower. These results signify the efficacy of our strategy in handling class imbalances, particularly in single lesion detection situations. Mammography screening is instrumental in the early detection and diagnosis of breast cancer by identifying masses in mammograms. With the rapid development of deep learning, numerous deep learning-based object detection algorithms have been explored for mass detection studies. However, these methods often yield a high false positive rate per image (FPPI) while achieving a high true positive rate (TPR). To maintain a higher TPR while also ensuring lower FPPI, we improved the Probability Anchor Assignment (PAA) algorithm to enhance the detection capability for mammographic characteristics with our previous work. We considered three dimensions: the backbone network, feature fusion module, and dense detection heads. The final experiment showed the effectiveness of the proposed method, and the TPR/FPPI values of the final improved PAA algorithm were 0.96/0.56 on the INbreast datasets. Compared to other methods, our method stands distinguished with its effectiveness in addressing the imbalance between positive and negative classes in cases of si