Breast cancer develops when a malignant tumor forms as a result of some breast cells growing abnormally quickly. It can be avoided and, if detected early enough, can significantly increase the prognosis and chances of survival by enabling patients to get timely clinical therapy.
Using the Database for Screening Mammography (DDSM) dataset, we developed an accurate method in this study for classifying mammograms or breast cancer images into cancerous and normal images. Two types of deep learning and transfer learning models were used, a pretrained resnet50 and an untrained resnet50 and the test results were compared using a confusion matrix.
We analyzed the accuracy value for pretrained model and compared it with a model we trained from scratch only to prove pretrained model(with an accuracy of 43.08300395256917 ) are not very accurate or ready to use as per differently defined use case situations , compared to the user trained ResNet50 model showed a test accuracy of 80.23715415019763.