<p>Cancer is regarded as one such top mortality cases that occurs with the abnormal growth of cells in human body. It becomes a great challenge causing huge threat to the people’s safety and their wellbeing’s throughout the world. For analyzing an image, the study of biopsy (histopathology) is carried most often that aids in the detection of breast cancer. The main intention of this work is to present an improved model for different phases of Computer Aided Diagnosis (CAD) system which plays a significant and challenging role on minimizing the gap between the existing research. An average weighted-based Gaussian filtering approach is employed for the purpose of pre-processing which is followed by Cauchy distribution-based segmentation of pre-processed image. From the segmented image, the features are extracted with the use of an optimal feature extraction model termed conjugate sword bee optimization-based feature extraction to attain optimal features. In addition, an improved classification design to detect breast cancer is employed as self-attention mechanism coupled with Long Short Term Memory classifier approach that is SAC-LSTM to diagnose and classify the kind of breast cancer like malignant or benign. In practice, the application of presented optimal feature extraction and SAC-LSTM classifier model for CAD system thus reduces the number of incorrect detections and diagnoses thereby increasing the classifier accuracy. The BreakHis dataset consisting of histopathological image is considered as an input for the performance estimation. Hence, this framework might serve as an effective opinion strategy for the pathologists thus aiding in the diagnosis of disease at an early stage.</p>