Authors: Gavali Pournima Dattatraya, Prof. C. A. Kolse, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Medical imaging allows clinicians to examine internal structures and identify disease, and Artificial Intelligence, particularly machine learning and deep learning, is increasingly used to analyse X-rays, CT, MRI, ultrasound and retinal photographs. This review summarizes common AI techniques, major applications, benefits, limitations and responsible-use considerations in medical image analysis. Convolutional neural networks for image classification and transfer learning, encoder decoder networks such as U-Net and its self-configuring and three-dimensional variants for organ and lesion segmentation, and detection of abnormalities for case prioritization are examined. Newer approaches including vision transformers, hybrid transformer–CNN segmenters and foundation models trained on large image collections, which can be adapted to many tasks or prompted to segment arbitrary structures, are discussed. Evidence from systematic comparisons with clinicians and from studies of generalization across hospitals is reviewed. A five-layer architecture for an AI-assisted imaging workflow is proposed, with research gaps, objectives, a comparison of model families and an analysis of challenges including dataset shift, annotation cost, bias, explainability, privacy and regulation. The review concludes that AI should support clinicians within validated workflows rather than replace their judgement.
DOI: https://doi.org/10.5281/zenodo.24220579
