Effectiveness of an artificial intelligence-empowered video-game system on stroke patients with dysphagia : a randomized controlled trial

Background Post-stroke dysphagia (PSD) affects 50-70% of stroke survivors, significantly increasing risks of aspiration pneumonia, malnutrition, and reduced quality of life. Traditional dysphagia rehabilitation requires frequent hospital visits, and the repetitive nature of traditional exercises often results in low compliance and engagement, leading to ineffective dysphagia management and limited improvement in swallowing function. Applying game thinking to rehabilitation training can increase participant involvement, making the rehabilitation process enjoyable. Although artificial intelligence (AI) combined with gaming has the potential to increase patient engagement, it has not been systematically explored or implemented in dysphagia rehabilitation. Objectives This aim of this randomized controlled trial (RCT) was to develop and assess the efficacy of a novel artificial intelligence-empowered video game system (AI-VG) in enhancing swallowing function among PSD patients. Additionally, it examined whether AI-VG system offers superior benefits in enhancing daily feeding function and training compliance compared to conventional methods. Methods This study employed a comprehensive approach: (1) A systematic review and network meta-analysis was conducted to evaluate and rank various therapeutic interventions for enhancing swallowing function, and feeding and daily function in PSD patients. (2) An AI-VG system was designed and developed for swallowing function training based on evidence-based nursing research methods. (3) The technology acceptance model was employed as the theoretical foundation, while in-depth interviews facilitated exploration of AI-VG system adoption among healthcare professionals and individuals with dysphagia. (4) A pilot RCT with a 4-week intervention was conducted at a Beijing rehabilitation center to assess the feasibility and acceptance of AI-VG system. (5) A RCT was conducted at the rehabilitation center between October 2023 to July 2024. Participants were allocated to either an AI-VG system intervention or conventional therapy group, completing 30-minute daily sessions (5 days/week) for four weeks. The primary outcome measured swallowing function improvement across three timepoints (baseline [T0], post-intervention [T1], and one-month follow-up [T2]). Secondary outcomes included laryngeal function, dietary intake level, swallowing safety, nutritional status, and swallowing-related quality of life. Treatment adherence, satisfaction and acceptance were assessed, analyzing temporal outcome variations through generalized estimating equation modeling. Results The systematic review and network meta-analysis identified acupuncture as the most effective interventions for enhancing swallowing function, followed by the chin tuck against resistance exercise. The AI-VG system was successfully developed with three game components targeting lip, tongue, and neck exercises. The qualitative research revealed favorable perceptions regarding AI-VG system utility and usability, despite implementation challenges. In the pilot RCT, the AI-VG system group showed greater improvement in adherence, swallowing function, oral-intake function, and nutrition status than the control group after treatment. The comprehensive RCT encompassed 84 participants (experimental group: n=42, mean age=64.98±9.66 years; control group: n=42, mean age=66.43±13.12 years). The AI-VG system intervention yielded significantly enhanced swallowing function compared to conventional therapy, with mean group differences of 4.02 (95% CI=-6.16 to -1.89, P<0.001) at T1 and 4.14 (95% CI=-6.16 to -2.12, P<0.001) at T2. Oral consumption capabilities, nutritional status indicators, and swallowing-related quality of life metrics showed significant improvement (P<0.001 for overall group × time interaction). Treatment adherence was markedly superior in the experimental cohort versus controls. The intervention group demonstrated strong acceptance and satisfaction toward the AI-VG system, while intergroup comparisons revealed no significant differences in laryngeal function or swallowing safety parameters. Conclusion An AI-VG system for swallowing rehabilitation was developed and its effectiveness and acceptability were assessed in individuals with PSD. AI-VG system intervention can be an effective way to strengthen swallowing, feeding, and daily living functions in patients with PSD. These findings provided valuable insights for clinical implementation of technology-enhanced rehabilitation interventions that could improve dysphagia management in stroke survivors. Significance By addressing key shortcomings of conventional swallowing therapies, this study presents the first empirical support for the AI-VG system as a viable rehabilitation tool for PSD. The integration of qualitative and quantitative data yielded important perspectives on its acceptance and efficacy. This study demonstrated the potential of AI technology to improve treatment adherence and patient participation in rehabilitation, while establishing a scalable framework for extending specialized dysphagia care to settings with limited access to speech-language pathologists.