Breast abnormalities classification using pre-processing and deep transfer learning techniques

Heart disease is no longer the second leading cause of mortality for women; instead, breast cancer has replaced it. Mammography examinations must accurately determine the precise type and subtype of breast cancer to effectively diagnose earlystage breast cancer, thereby enhancing the probability of the patient's survival. However, given the range of breast kinds and subtypes, as well as the complexity of their microenvironment, it continues to be a severe issue. To solve these issues, we provide a two-stage classification technique. The pre-processing step, which comprised multiple noise reduction filters, was where it all began. Then, using transfer learning approaches, we suggested a CNN architecture with feature extractors based on pre-trained models from Resnet50 V2 and Xception. As a consequence, we proved that the suggested method worked effectively for breast Mass instances by developing such a model and training it using our pre-processed dataset. Using the CBIS-DDSM we reached the highest accuracy of 99.99% for classifying breast Nasses and Calcs.