A Diagnosis Model Based on Federated Learning for Lung Cancer Classification

Lung cancer, causing a significant number of fatalities annually, is a prominent contributor to global mortality rates among individuals of both genders, with an estimated five million cases resulting in death each year. The utilization of Computed Tomography (CT) scans contribute valuable data for the detection of lung cancer. The main objectives of this study are to diagnose malignant lung nodules in the provided input lung images and to classify lung nodule based on its level of severity. This study proposes a solution wherein locally computed updates are combined to stimulate the learning of a shared model, while the training data remains dispersed throughout mobile devices. Federated Learning (FL) is an advanced technology that operates in a collaborative and decentralized manner, with the primary objective of preserving privacy. Its purpose is to address the obstacles posed by data silos and the sensitivity of data. Therefore, the FL approach is employed for the classification of medical images in collaborative and real-world circumstances. A new approach is proposed in this paper, where the model is trained on data distributed across multiple institutions, and the training process occurs locally at each institution. The performance of the proposed method in terms of accuracy, specificity and sensitivity is evaluated and compared with that of the existing models such as Googlenet, Alexnet, and SVM. The proposed method performs comparably better in all three performance measures indicating that it works better than other methods.