Artificial Intelligence machine learning and conventional physical therapy for upper limb outcome in patients with...

The goal of this study was to compare the effect of different artificial intelligence (AI) machine learning and conventional therapy (CT) on upper limb impairments in patients with stroke.

PubMed, PubMed Central, Google Scholar, MEDLINE, Cochrane Library, Web of Science, Research Gate, and Wiley Online Library were searched. Descriptive statistics about variables were reported to calculate standardized mean differences in outcomes of motor control (the primary outcome), functional independence, upper extremity performance, and muscle tone. The Physiotherapy Evidence Database (PEDro) Scale was used to assess qualitative papers. The primary outcomes of AI and CT have been included in the meta-analyses.

Ten papers with a total of 481 stroke patients were included and upper limb rehabilitation, upper limb functioning, and basic manual dexterity were examined. The heterogeneity test of the whole included measures (I2=45%) was medium. There were significant differences between the included measures (p-value=0.03) with a total SMD of 0.10 [0.01, 0.19]. According to the test for subgroup difference, it was found that there was a highly significant difference between the subgroups of the included measures (p-value=0.01) and the heterogeneity test (I2=59.8%).

AI is a feasible and safe method in post-stroke rehabilitation and improves upper-extremity function compared to CT. Significant AI post-treatment effects on upper-limb impairments have been observed. The findings showed that higher-quality evidence was detected in six assessment scales. However, a lower quality of evidence was detected in other scales. This indicated large or very large and consistent estimates of the treatment effects, and researchers were confident about the results. Therefore, the included observational studies are likely to provide an overestimate of the true effect.