Systematic review of AI/ML applications in multi-domain robotic rehabilitation: trends, gaps, and future directions.
Giovanna Nicora, Samuele Pe, Gabriele Santangelo and 6 others
PMID 40205472WHAT IT FOUND
Most AI robotic rehabilitation tools are tested on healthy volunteers, not patients.
Performance drops significantly when systems move from lab settings to real-time clinical use. Clinicians should expect these tools to work less reliably on their actual patients than published accuracy figures suggest.
Key findings
0172% of the reviewed studies used healthy individuals for training and testing, while only 55 involved actual patients.
02All 8 studies that compared offline versus online performance reported a significant decrease in accuracy when deployed in real-time settings.
03Children are severely underrepresented, appearing in only 3% of studies, making it difficult to predict how these adult-trained algorithms will perform in pediatric therapy.
STILL TO COME
How it was doneWhat they foundWhat it means for PTsWhat it means for OTsWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The review only included papers published in English. Many included studies had very small patient cohorts (median of 9 patients), limiting the generalizability of their results. The review could not assess the actual clinical outcomes of patients because it focused on the technical performance of the AI/ML systems rather than patient recovery metrics. Only 201 papers were included from a rapidly evolving field, and the search was limited to two databases.
Declared interests
The paper is supported by Non-U.S. Government funding. No specific conflicts of interest involving commercial robotics companies were declared in the provided text.
The easy way to misread this
Do not interpret the high accuracy rates reported in these studies as evidence that the technology is clinically effective for patients. Most systems were validated on healthy volunteers, and performance consistently drops when applied to impaired populations or used in real-time clinical settings.