Region-of-Interest Optimization for Deep-Learning-Based Breast Cancer Detection in Mammograms

The early detection and diagnosis of breast cancer may increase survival rates reduce overall treatment costs. the is a severe potentially fatal disease that impacts individuals worldwide. Mammography widely utilized imaging technique for surveillance diagnosis. However, images produced with mammography frequently contain noise, poor contrast, other anomalies hinder radiologists from interpreting images. This study develops novel deep-learning using proposed procedure consists two primary steps: region-of-interest (ROI) (1) extraction (2) classification. At beginning procedure, YOLOX model to distinguish tissue background identify ROIs lesions. In second phase, EfficientNet or ConvNeXt applied data benign malignant ROIs. validated large dataset various institutions compared several baseline methods. pF1 index used measure effectiveness technique, which aims establish balance between number false positives negatives, harmonic mean accuracy recall. method outperformed existing methods by an average 8.0%, obtaining superior levels precision sensitivity, area under receiver operating characteristics curve (ROC AUC) precision–recall (PR AUC). addition, ablation research was conducted investigate effects procedure’s numerous components. According findings, another choice could enhance