A Dual Approach for Optimizing Breast Cancer Mammogram Classification using ResNet50, SMOTE, CNN and LSTM

Abstract

Breast Cancer is one of the major health concern in the world where this disease can kill large number of women globally. This health concern is caused by the growth of abnormal tissue in the breast. The early detection is very crucial to prevent it from growing into a last stage, to detect this in early stages we need models that are able to detect the disease efficiently. Deep learning algorithms are shown effective results in predicting the diseases, the imbalanced datasets in medical diagnosis helps in developing classification models that gives accurate results. Some of the existing models were providing some moderate accuracy results and to improve that accuracy more we are proposing a stacked model that uses balanced and imbalanced data.To handle this imbalanced data we are using a technique called (SMOTE) Synthetic Minority Over-sampling Technique, along with that we are using VGG16 preprocessing for input standardization and ResNet50 for feature exraction,and here in our proposed stacked model we are using CNN with multiple dense layers, although this CNN is best known for optimizing spatial features, it lacks in processing the temporal features that changes time to time so, to overcome this drawback here we are introducing the CNN along with LSTM here both the CNN and LSTM helps in optimizing the temporal features and spatial features that helps the trained model to predict the results correctly. This stacked method approach significantly improved the prediction results giving the accuracy of99.66% along with the other metrics like precision, recall, and F1score are giving 99.65%, 99.66%, 99.65% respectively.