Domain Adaptation in Medical Imaging: Evaluating the Effectiveness of Transfer Learning

Deep learning (DL) shows great promise in medical imaging, yet its widespread application across various medical fields encounters obstacles due to distinct data distribution variations in each domain. This research delves into the effectiveness of transfer learning, specifically within the domain adaptation framework for medical imaging, addressing the challenges posed by varying data distributions across different medical domains. This paper used two modified models, MobileNet and EfficientNet, to classify medical datasets. We studied transfer learning with metadata using two medical datasets: the MRI samples dataset and the chest X-ray samples dataset. We compared the achievement of our approach to the most advanced method. In the two models, EfficientNet B2 and MobileNet V2, total categorization accuracy was for brain tumors, 97.48 and 95.09%, and lung diseases, 97.77 and 96.67%. We created a model that could be trained on devices with low computational power, making it ideal for deployment in smaller IoT devices.