<![CDATA[As physical therapy practice moves toward the increasingly large and complex multimodal patient data stream (imaging, gait kinetics, wearable-sensor streams, and patient-reported outcomes), relying on unaided clinical judgment is insufficient for pattern recognition. Twenty-five papers were retrieved between 2018 and 2025 for artificial intelligence (AI) clinical decision support systems (CDSS) related to diagnostic accuracy and personalized treatment planning in physical therapy.Twenty-five papers were identified between 2018 and 2025 for AI clinical decision support systems (CDSS) for diagnostic accuracy and personalized treatment planning in physical therapy. The deep-learning models for knee osteoarthritis grading, low back pain classification, and sarcopenia-related gait screening are analyzed, as well as the large language model-based clinical reasoning models, multi-sensor rehabilitation-monitoring platforms, and myoelectric control systems for upper-limb recovery. The reported diagnostic accuracy of imaging-based models ranges from 86.2% to 92.5% and the evidence from the network meta-analysis suggests that the improvement of pain and ROM outcomes by AI-assisted rehabilitation is greater than conventional rehabilitation. The main barriers to the adoption are clinician trust, burden of integration to the workflow, and data-privacy concerns; while the facilitators are explainability and demonstrated diagnostic benefit. Ethical, legal, and regulatory issues related to the use of AI-CDSS in rehabilitation are also explored in the synthesis. The results confirm the hybrid model (clinician in the loop) of integrating AI-CDSS within the task of physical therapist judgment.]]>
Copyrights © 2025
