Ghosh, Tapotosh; Sheikhi, Farnaz; Guo, Junlin; Singh, Yashbir; Younis, Khaled; Kuanar, Shiba; Faghani, Shahriar; Farina, Eduardo Moreno Judice de Mattos; Huo, Yuankai; Maleki, Farhad (2026).听.听Electronics (Switzerland), 15(6), 1245.听
Foundation models鈥攍arge, pre-trained artificial intelligence systems that can be adapted to many different tasks鈥攁re increasingly transforming medical image analysis. While earlier work focused mostly on 2D images, there is growing interest in applying these models to聽volumetric (3D) medical images, such as聽CT, MRI, and PET scans, which capture the body in three dimensions and provide richer clinical information. This review summarizes recent progress in building such models, including their use of聽3D neural network architectures聽(AI systems designed to process 3D data) and different training approaches.
The paper highlights how these models are being used for important medical tasks, such as聽classification聽(identifying diseases),聽segmentation聽(outlining organs or tumors),聽image registration聽(aligning different scans), improving image quality, and even聽visual question answering聽(answering questions about medical images). At the same time, it discusses key challenges, including the high computational resources required, the limited availability of large and diverse 3D medical datasets, and difficulties in adapting models across different clinical settings.
Overall, the review provides a clear overview of the field and outlines future directions for developing more scalable and reliable AI tools that can work effectively with 3D medical images in real-world healthcare.
Figure 1.聽Paper selection process. 376 articles were initially found from different repositories. After removing duplicates and imposing strict criteria, 60 articles were selected for review.