Intelligent BCDecNet Framework: Adaptive Segmentation with Optimized Multiscale Dilated Deep Residual Attention Network for Mammogram Breast Cancer Detection
J. Sivamurugan, Sureshkumar Govindarajan · International Journal of Image and Graphics · 2025
Breast cancer is a dreadful disorder which causes increasing fatality rates, especially in women. Early diagnosis as well as proper treatment of breast cancer can control the mortality rate, which tends to identify both malignant and benign cases. Mammography is used to recognize breast cancer at the initial phase to reduce the death rate. From the existing research, it is noted that the detection becomes more complicated to achieve accurate results. Since most of the time, errors attained in the detection process are based on the noise pixel as a false positive. Radiologists use the deep learning technique to make an accurate diagnosis and enhance the prediction outcome. Hence, this paper plans to implement an effective breast cancer diagnosis framework for securing the lives of patients by providing the appropriate treatment at an earlier stage. Initially, the mammogram images are gathered from benchmark online datasets. Here, the images are utilized in the primary stage of processing to remove the noisy regions among the images. In the phase of pre-processing, the breast cancer segmentation phase takes place with the Adaptive Swin Transformer-based Attention Unet (AST-AUnet), where the parameters in the segmentation technique are tuned with the Randomized Condition of Driving Training and Dingo Optimization (RCDDO), which is the combination of Driving Training-Based Optimization (DTBO) and Dingo Optimization Algorithm (DOA). This helps to enhance the segmentation rate. Further, the breast cancer diagnosis is performed with an Optimized Multiscale Dilated Deep Residual Attention Network (OMD-DRAN) to accurately classify the breast cancer. Here, the parameter tuning is done using the same hybrid RCDDO algorithm for achieving a better performance rate. Experimental outcomes of the recommended approach for breast cancer diagnosis attain better performance by analyzing the diverse segmentation and classification conventional models. The empirical findings of the recommended model achieve 96.50%, 93.19%, and 98.25% regarding accuracy, precision and MCC, respectively, to enhance breast cancer detection appropriately. Moreover, accurate performance improvement in healthcare settings could detect the possibility of breast cancer promptly.