Deep Learning for Breast Cancer Detection from Mammograms Images

Mammograms has been widely used for early detection of breast cancer; however, radiologists are prone to error which could lead to unnecessary tests or missed treatment window. With the advancement of state-of-the-art CNN models, transfer learning has become a popular technique for CAD development. In this study, a total of six state-of-the-art models are used, namely VGG16, Inception V3, ResNet50, EfficientNet B3, EfficientNet V2M and ConvNext were used to classify mammograms from two public datasets, INbreast and CBIS-DDSM. Prior to training and validation, the datasets were preprocessed with artefact removal, median filter, CLAHE and ROI extraction. The best performance was achieved by the ConvNext model trained on INbreast dataset, with an accuracy, sensitivity, specificity, precision, f1-score and AUC of 0.81, 0.8, 0.82, 0.67, 0.73 and 0.88, respectively. The results indicate that the model has the potential to improve the accuracy of breast cancer detection and reduce the risk of missed diagnoses.