Abstract Stroke-associated hand impairment is considered the major challenge in neurorehabilitation, which often causes loss of motor control and functional independence. While traditional therapeutic approaches have positive outcomes, they are labor-intensive and costly, and they lack the capacity for continuous performance monitoring. New technologies in wearable sensors and artificial intelligence (AI) have made it possible to produce smart gloves that can record detailed hand kinematics and, thereby, personalized therapeutic interventions. This is a systematic literature review of 101 peer-reviewed articles published between 2011 and 2025 that will address AI-driven smart glove systems incorporating flex sensors, inertial measurement units (IMUs), and force-sensitive resistors (FSRs) to identify gestures and gauge the rehabilitation condition. The review uses the PRISMA 2020 methodology to compare sensing modalities, architectures of algorithms, and clinical validation results. Results show that 72% of the studies used multimodal sensor fusion (flex + IMU). Recognition accuracy generally ranged between 82% and 99% across the included studies. The models based on deep learning, like convolutional neural networks (CNNs) and long short-term memory (LSTM) networks, demonstrate superior recognition performance, although they consume more computational resources; the classical models, like support vector machines (SVMs) and k-nearest neighbors (KNNs), are more computationally efficient and less resource-intensive. The remaining issues include sensor drift, insufficient diversity of the dataset, and large-scale clinical validation. The review emphasizes the necessity for future studies to concentrate on edge-optimized AI models, distinctive learning algorithms, and standardized clinical datasets, thus facilitating scalable, real-time therapy.
Mohammed et al. (Sat,) studied this question.