Unveiling Breast Tumor Characteristics: A ResNet152V2 and Mask R-CNN Based Approach for Type and Size Recognition in Mammograms

As one of the most prevalent and lethal diseases afflicting women today, breast cancer detection remains a pivotal area of focus. Although mammogram images, exploited in Computer-Aided Design (CAD) systems, provide an early detection avenue, their reliability for accurate recognition of tumor density types and sizes, particularly in type C and D breasts, is questionable. To address this challenge, a novel approach for tumor identification, categorization, and size estimation in various breast types is put forth in this study. In the proposed model, features are extracted from a mammographic image analysis dataset using a pre-trained Convolutional Neural Network (CNN) architecture for left-right comparison, followed by the deployment of ResNetl52V2 for distinguishing between the four mammogram types (A, B, C, and D). Subsequently, normal, and abnormal breasts are differentiated within the mammogram images. The final step employs a Mask Region-Based Convolutional Neural Network (Mask R-CNN) to discern malignant from benign tumors and to estimate tumor size. The experimental outcomes demonstrate an impressive 100% overall accuracy in type comparison using ResNetl52V2, thereby substantiating its viability as a model for mammogram type detection and classification. This study thus provides a compelling argument for the application of ResNetl52V2 in the context of breast cancer detection and diagnosis.