Automated breast cancer diagnosis via intensity-guided contour segmentation and ResNet-based classification using local intensity pattern tensors.

In the world, breast cancer is one of the leading causes of mortality among women who have been diagnosed with cancer. A timely finding and accurate diagnosis are essential to the development of effective therapy and increased survival rates. This study proposes a novel approach for the segmentation and classification of breast cancer lesions using the Digital Database for Screening Mammography (DDSM). The methodology begins with the preprocessing of mammogram images to enhance contrast and reduce noise. A segmentation technique, Intensity-Guided Adaptive Contour (IGAC), is employed to accurately extract regions of interest (ROIs) representing potential lesions. The IGAC method adapts based on intensity variations to capture the complex boundaries of lesions, minimizing interference from surrounding tissues. For feature extraction, a Local Intensity-Pattern Tensor (LIPT) is utilized to derive shape, texture, and intensity-based metrics from the segmented regions. These features provide a comprehensive representation of the lesions, which are indicative of benign or malignant patterns. A deep learning model based on the Residual Neural Network (ResNet) architecture is then used for classification, categorizing mammogram images into benign, malignant, or normal. The ResNet model leverages its depth and skip connections to ensure robust learning and high classification accuracy. Performance evaluation focuses on accuracy, sensitivity, and specificity, demonstrating the effectiveness of this integrated approach for breast cancer detection and classification. The performance of this work is demonstrated by the following metrics: an accuracy of 99.97%, a sensitivity of 99.34%, a specificity of 99.13%, a precision of 98.39%, and an F1 score of 99.01%.