Effects of Artificial Intelligence Rehabilitation on Motor ability and Daily living ability of Hemiplegic Patients with Stroke—Meta-Analysis of Randomized Controlled Trials (Preprint)

BACKGROUND A large number of hemiplegic stroke patients worldwide require rehabilitation. Artificial intelligence (AI) has the potential to conserve human resources and offers broad application prospects. With advancements in medicine and technology, AI has begun integrating into rehabilitation, providing personalized rehabilitation plans. However, the effects of AI rehabilitation on the motor and daily living abilities of hemiplegic stroke patients remain unclear. OBJECTIVE Evaluate the effects of AI rehabilitation on motor and daily living abilities in hemiplegic stroke patients. METHODS The Cochrane Library, Web of Science, PubMed, Embase, CINAHL, CNKI, VIP, and Wanfang databases were systematically searched for randomized controlled trials (RCTs) on AI rehabilitation for hemiplegic patients with stroke. The search timeframe was from the construction of the database to January 1, 2025. The literature was screened according to the nerfing criteria, relevant information was extracted, and Meta-analysis was performed using RevMan5.3 software. RESULTS 16 studies involving 565 stroke patients with hemiplegia were included. Meta-analysis showed that, compared with conventional rehabilitation, AI rehabilitation was more effective in improving motor ability [MD=3.35, 95%CI (1.39, 5.32), P<0.001], balance [MD=7.26, 95%CI (6.37, 8.14), P<0.001], muscle strength and grip strength [SMD=0.65, 95%CI (0.25, 1.04), P=0.001], and the ability to perform activities of daily living [SMD=1.71, 95%CI (0.73, 2.69), P<0.001]. However, improvements in limb function [MD=0.11, 95%CI (-0.06, 0.28), P=0.210], muscle tone [MD=-0.28, 95%CI (-0.57, 0.02), P=0.060], function in activities of daily living [MD=-0.04, 95%CI (-0.49, 0.41), P=0.860], and hand dexterity [MD=9.31, 95%CI (-7.48, 26.09), P=0.280] were not statistically significant. Subgroup analyses of motor ability revealed no statistical difference between rehabilitation machines and conventional rehabilitation in improving motor ability [MD=1.80, 95%CI (-1.37, 4.97), P=0.270]. In contrast, virtual reality [MD=5.07, 95%CI (4.23, 5.91), P<0.001], brain-computer interface [MD=6.99, 95%CI (3.06, 10.92), P<0.001], and telerehabilitation [MD=0.96, 95%CI (0.23, 1.68), P=0.010] all significantly improved motor performance. Additionally, interventions with a total frequency of ≥20 [MD=4.29, 95%CI (2.21, 6.36), P<0.001] and a duration of ≥6 weeks [MD=3.73, 95%CI (1.22, 6.24), P=0.004] were more effective. A total intervention duration of ≥10 hours [MD=5.71, 95%CI (3.02, 8.40), P<0.001] also had a better effect on motor ability improvement. Subgroup analyses of daily living ability showed that interventions with a total duration of >10 hours [SMD=3.18, 95%CI (1.44, 4.93), P<0.001] were more effective in improving daily living ability. CONCLUSIONS AI rehabilitation can improve the motor ability of stroke patients with hemiplegia. Using virtual reality, brain-computer interface, and telerehabilitation is recommended, ,with a total frequency of ≥20 interventions, ≥6 weeks of interventions, and a total duration of ≥10 hours. AI rehabilitation can also improve the ability to perform daily activities, with a recommended total intervention duration of >10 hours. Additionally, AI rehabilitation can enhance balance function, muscle strength, and grip strength. However, it does not significantly improve limb function. The total intervention duration is recommended to be >10 hours. More high-quality studies are needed to validate these findings further. CLINICALTRIAL PROSPERO CRD42025636225；https://tinyurl.com/2uc3eac2.