Artificial Intelligence in Medical Imaging: From Task-Specific Models to Clinical Evaluation

Artificial intelligence (AI) is increasingly used in medical imaging, with research spanning task-specific classification, detection, and segmentation, representation learning, image reconstruction, generative models, foundation models, and vision–language models (VLMs). This review synthesizes 74 core publications and organizes them into four groups: task-specific image analysis; representation learning and image reconstruction; generative and generalist models; and robustness, generalizability, and clinical evaluation. Across these research directions, reported performance is shaped by data sources, patient-level separation, acquisition conditions, label quality, model adaptation, and validation design. Generative and multimodal systems also raise questions about factual consistency and clinical use. Among the 45 primary empirical studies in the selected core set, multicentre evaluation and temporal/prospective evaluation were each assigned as the primary evaluation profile to one study. No prospective interventional or randomized clinical evaluation was identified within this core set. The review summarizes methodological developments and evaluation challenges and discusses priorities for reliable medical imaging AI, including external validation and evaluation in intended clinical settings.