An Adaptive Trans-ResUnet++Based Segmentation and Hybrid CNN-Aided Classification for Detecting Breast Cancer from Mammogram Images
Potta Mounika, B. Narayanan, Kavitha Rani Balmuri · Cybernetics & Systems · 2025
The most prevalent cancers among females in worldwide as breast cancer increasing the mortality rate. For analyzing the accurate breast cancer, radiologists assist by several advanced models for earlier diagnosis with exact treatment. Mammogram scans are widely regarded as the highest precision imaging framework for diagnosing breast cancer. The key challenges in mammogram images reduce the sensitivity in women having dense breast tissue. In addition to this, numerous deep learning techniques are being assisted to address the issues, but the main concern is increasing the false positives and false negatives during the screening process. Therefore, an intellectual framework for cancer detection utilizing deep learning is developed. At first, the breast mammogram images are sourced from established benchmark sources. These mammogram images are preceded by the segmentation phase. For implementing the segmentation process, an Adaptive Trans-ResUnet++ (AT-RUnet++) mechanism is developed. Here, the parameters from ResUnet++ are optimized to enhance the segmentation performance using the Improved Random Variable-based Ali Baba and the Forty Thieves algorithm (IRV-AFT). For classifying the breast cancer, the Hybrid Variants of Convolutional Neural Network (Hv-CNN) approach is developed. This mechanism is formed by combining Graph Convolutional Networks (GCNN) and Residual Attention Network (RAN). The effectiveness of the recommended technique is contrasted with other classification mechanisms and approaches. The numerical outcomes reveal that the implemented model ensures 92.41% accuracy and 98.61% specificity in 3-fold to prove its effectiveness. This reliable performance helps the developed model suggest earlier breast cancer diagnosis.