Medical Image Classification

Medical image classification studies how to assign diagnoses or severity grades from images such as X-rays, MRI, CT, ultrasound, fundus photographs, and pathology slides. Common approaches include convolutional neural networks, vision transformers, 2D and 3D models, and foundation models adapted through transfer learning or self-supervised pretraining; some combine images with clinical records, text, or multiple views. Work also explores interpretability, uncertainty calibration, fairness, and efficient learning when labels are scarce or data cannot be pooled across hospitals. A central challenge is that strong results on internal benchmarks may not transfer across sites, devices, populations, or disease presentations, so patient-level splits and external validation remain essential.