Breast Cancer Diagnosis in Radiology Images Using Transfer Learning Technique
S. Leena Nesamani, D. Lissy, S Sumathy, S. Nirmala Sugirtha Rajini · 2023
Diseases like cancer which are life-threatening needs accurate diagnosing techniques at a very early stage. This research work is focused on identifying one such technique for identifying breast cancers at a very early stage. This work focuses on employing radiology images of breast mammograms using a transfer learning technique, which is a very powerful deep learning technique that has proven to produce excellent results on computer vision projects. As medical data are very sparse and difficult to obtain transfer learning technique is employed where the knowledge gained through a previous task could be transferred to a new task of a different domain. Breast mammogram images were taken from the publicly available Mini DDSM Dataset which consists of breast mammogram images in the Medio Lateral Oblique (MLO) and bilateral Cranio Caudal (CC) views of both thebreasts. Three experiments were performed on the dataset to demonstrate the effectiveness of the technique by choosing three pre-trained models VGG19, ResNet 50, and Xception and the results were compared. The overall performance of the pre-trained models was found to be elevated due to the deployment of a Logistic Regression (LR) classifier. VGG19 pre-trained model proved to produce the topmost result of 88% accuracy on both the training as well as on the validation datasets. The other two pretrained models exhibited a slightly lower performance of ResNet50 with 80% and Xception with 79% accuracy.