Abnormalities Detection in Chest CT Scan Images using Federated and Deep Learning

Lung cancer, which accounts for an estimated 2.20 million reported cases and leading to 1.80 million fatalities in the year 2020, represents a significant global cause of mortality. Early detection during Computed Tomography (CT) chest scans can improve survival rate of patients. To address the growing use of CT scan, a robust automated model is required to eliminate tedious manual procedures and reduce variability in diagnosis. Moreover, there are challenges associated with medical research such as data privacy, collaboration, regulatory compliance, and data diversity. In this study, we propose two machine learning (ML) methods and subsequently conduct a comparative analysis. In the first approach, we presented a deep learning technique to achieve high image classification accuracy by using multi-layer CNN and two pre-trained models, namely ResNet50 and DenseNet201. In the second approach, we presented a Federated Learning (FL) technique, which entails the utilization of all three models for classifying abnormalities in chest CT scans. Our results illustrate that the FL-based approach can assist in the diagnosis of anomalies in chest CT scan, while ensuring the privacy of patient data. By exploring the delicate balance between privacy and accuracy in medical image classification, our study will contribute to the advancing realm of research.