Improved Breast Cancer Classification Approach Using Hybrid Deep Learning Strategies for Tumor Segmentation

Breast malignancy is one of the mainly prominent tumors that affect individuals today, especially women. This disease has certain morphological features, so it is heterogeneous. Thus, a practically usable classification technique is developed to detect tumors, and the developed technique is scientifically valid, medically acceptable, and often repeatable. Detecting this malignancy in its initial phases will considerably reduce the harm formed by this tumor. The traditional breast cancer identification process is problematic and produces a high rate of fatalities and morbidity. It takes a long time, at least months or weeks. The effect of this tumor is reduced because of its immediate identification. So, utilizing a deep structured approach to diagnose breast cancer allows rapid tumor identification. In this task, an effective breast cancer classification approach is implemented with the utilization of deep structured mechanisms. At first, the required images are gathered from multiple benchmark datasets. Further, the collected images are forwarded to the segmentation phase, where the Hybrid Convolution-based Trans-MobileUNet +  + (HCTMUNet + +) is adopted. The segmented images after subjected to the classification stage. In this phase, the Hybrid Adaptive and Attentive Network (HAAN) is utilized, which is the integration of ShuffleNet and MobileNet with Attention Mechanism. The parameter presented in this network is optimally tuned by the Entrenched Fitness-based Gannet Optimization Algorithm (EFGOA) for effective results. From the result analysis, the accuracy and Net Present Value (NPV) of the offered model is 95.72% and 95.68%. In the end, the classified findings are obtained, and these results are contrasted with divergent optimization algorithms and classifiers to showcase the developed model’s effectiveness.