Prediction of IDC Breast Cancer by the Application of Transfer Learning with an Ensemble Method

Cancer mainly is the result of uncontrollable changes in genes. Hence breast cancer is a result of the unstoppable growth of breast cells. A common type of breast cancer is Invasive Ductal Carcinoma (IDC) also popular as Infiltrating Carcinoma. Detection of IDC is necessary and precise identification is expected. If the diagnosis is accurate then the patient will be able to get the necessary treatment needed. For the detection of IDC positive or negative, our proposed methodology employs transfer learning by using ResNet-50, Inception V2, and EfficientNet deep learning models with an ensemble method. The dataset used here consisted of 162 whole-mount slide images of Breast Cancer specimens. From that, 277,524 patches of size 50 × 50 were extracted (198,738 IDC negative and 78,786 IDC positive). This dataset contains a huge number of histopathology images and is passed through the models for training, testing, and validation. In the training section, 10-fold cross-validation was used. This method gives better performance than any single method as it uses an ensemble method with all three mentioned models. The accuracy of separate models is compared with the ensemble methods result. Having made the ensemble approach, we got an accuracy of 94.047% which is higher than the models performing separately. Moreover, the other performance parameters also showed improvement in the ensemble approach. With this satisfactory result, this approach can be of great use in detecting IDC breast cancer.