A Federated Learning-based Model for the Detection of Lung Cancer from CT Scan Images

Abstract

The lung is an important organ of the human body. This lung can be affected by different types of diseases. Lung cancer is one of them, and it is one of the most lethal cancers. Early and faster detection of this disease can reduce its spread in the human body. In this study, a privacy-preserving, federated learning-based approach has been proposed to detect lung cancer from CT scan images. For that first, a lung CT scan dataset has been collected, which contains four different classes: adenocarcinoma, large-cell carcinoma, normal, and squamous-cell carcinoma images. Secondly, various preprocessing techniques have been applied to the images. Then, in the third step, different Transfer Learning (TL)-based models, which are: MobileNet, MobileNetV2, ResNet50V2, VGG16, and InceptionV3, have been implemented to find the optimal model. Among them, MobileNetV2 achieved the highest accuracy of 92.27%. In the next and last step, a Federated Learning (FL)-based model has been developed using the MobileNetV2 model. The proposed Federated Learning-based model outperformed the conventional model in terms of performance. It achieved an accuracy, precision, recall, and f-1 score of 93.92 %, 93.50 %, 93.50%, and 93.25%, respectively. Nevertheless, in the Federated Learning-based approach, not only the performance of the model increased, but also the users did not need to share their data. So, this proposed method can also ensure the privacy of the data shared by hospitals or clinics.

Topics & Concepts