Purpose Unilateral neglect affects up to 80% of right hemisphere stroke survivors and poses a significant barrier to rehabilitation. It is a strong predictor of poor prognosis and leads to prolonged hospital stays, yet no established treatment currently exists. Methods We propose an innovative approach, NeuroNavAR, a novel treatment method that utilizes neural network models to train the human nervous system under the guidance of augmented reality (AR). This method focuses on an AR-enhanced smart training program specifically for post-stroke spatial neglect. While the current implementation runs on a GPU server for technical validation, the application is designed for future deployment on Android and iOS devices. It employs image processing algorithms to identify real-world objects, such as clocks and chairs, segment their contours, and generate virtual bee patterns. These virtual bees navigate along the contours of the objects, and patients are encouraged to follow the bee patterns, receiving rewards upon completing a cycle. If patients fail to complete the cycle, virtual birds are introduced into their field of view to guide them back to the bee patterns, thereby enhancing engagement. Results This study reports only on the technical feasibility of the system. No clinical, behavioral, or usability data were collected. On a test dataset comprising 537 indoor images from the ADE20K dataset (accessed June 2024), the segmentation achieved a mean Intersection over Union (IoU) of 76.70% (SD = 8.2%) and an accuracy of 88.51% (SD = 4.7%). Conclusion The NeuroNavAR program represents a preclinical engineering feasibility study that provides a more intuitive and user-friendly training experience compared to traditional rehabilitation methods. Its innovative use of augmented reality and real-time feedback is designed to support, but has not yet demonstrated, improved rehabilitation outcomes for patients with unilateral spatial neglect. Clinical validation is planned for 2026.
Yanzhen et al. (2026) studied this question.