Diagnostic AI tools in musculoskeletal imaging-Present and future

Artificial intelligence (AI) is widely promoted as the next step in musculoskeletal (MSK) imaging, yet its translation into routine clinical care, including osteoarthritis (OA), has been selective and challenging. This Perspective summarizes where diagnostic MSK AI tools stand today and argues that the central question is not whether models perform well in benchmarks, but whether they create measurable clinical value. Based on a keynote presentation at IWOAI 2026, we relate diagnostic AI tools in MSK imaging to the theme "Linking OA Tissues" by asking whether current AI tools can connect tissue-level imaging findings with whole-joint disease, longitudinal change, and clinical context. Drawing on the regulatory landscape and peer-reviewed evidence, we show that AI tools have been selectively translated into clinical routine: radiographic morphometry and fracture detection dominate among commercially available products because they are narrow, standardized tasks with transparent, easily verifiable outputs and an obvious clinical need. By contrast, MRI, the modality on which advanced OA assessment largely depends, remains a bottleneck. OA is a particularly challenging target because it is a multi-tissue, whole-joint, longitudinal, and context-dependent disease for which binary, single-task outputs are too crude. Furthermore, we propose a deliberately non-quantitative framework in which useful AI is the product of model performance, relevant endpoints, workflow fit, and human-AI interaction; failures of translation typically reflect deficits in the latter three rather than in model performance alone. Realizing clinical value in MSK and OA imaging will require validation against patient-relevant endpoints, attention to how readers interact with AI, and reference standards that go beyond curated benchmarks.