AI-powered human action recognition of functional activities in stroke: a systematic review

Abstract Artificial intelligence (AI)-based human activity recognition (HAR) is a promising approach for detecting functional activity in individuals with stroke, potentially enabling objective and ecologically valid assessment, monitoring and treatment. We aimed to synthesize current evidence on AI-based HAR for functional activity recognition after stroke, focusing on sensing technologies, target activities, settings, and model architectures. PubMed, Embase, WOS, and Scopus were searched and seventeen studies involving 224 individuals with stroke met the inclusion criteria. Inertial measurement units were the predominant sensing modality (12/17, 70.6%), with several studies demonstrating the feasibility of recognizing relevant functional activities using a single device. Single-sensor approaches were also used in home-based and simulated-home settings, suggesting potential for scalable and unobtrusive monitoring outside clinical environments; however, most studies were conducted in controlled clinical settings (11/17, 64.7%), with limited real-world validation to date. Convolutional neural networks were the most frequently used deep learning architecture, while Random Forest was the most common traditional machine learning method. Multimodal sensing may enrich activity characterization, although greater sensing complexity may reduce feasibility for prolonged home monitoring. Methodological and reporting limitations, including concerns regarding confounding and reproducibility, should be considered when interpreting the current evidence. Overall, single-sensor AI-based HAR appears promising for low-burden, continuous, and ecologically valid monitoring. Future research should prioritize longitudinal validation in everyday environments, transparent reporting of code and data availability, and inclusion of acute and severely impaired stroke populations to facilitate clinical translation toward objective monitoring, personalized rehabilitation, and remote care.

Authors

Publication Details

Journal
Artificial Intelligence Review
Published
2026-11-15
DOI
https://doi.org/10.1007/s10462-026-11753-w
Primary Topic
Context-Aware Activity Recognition Systems
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

AI-powered human action recognition of functional activities in stroke: a systematic review

Valentina Pezzopane, Beatrice De Lazzari, Sonia Bonnı̀, Elisa Masi et al.
Artificial Intelligence Review
Context-Aware Activity Recognition Systems
article

AI-powered human action recognition of functional activities in stroke: a systematic review

Valentina Pezzopane, Beatrice De Lazzari, Sonia Bonnı̀, Elisa Masi, Alex Martino Cinnera, Calogero Malfitano, Irene Ciancarelli, Giacomo Koch, Francesco Negrini, Martina D’Arienzo, Pietro Deledda, Giovanni Morone
article en

Abstract

Abstract Artificial intelligence (AI)-based human activity recognition (HAR) is a promising approach for detecting functional activity in individuals with stroke, potentially enabling objective and ecologically valid assessment, monitoring and treatment. We aimed to synthesize current evidence on AI-based HAR for functional activity recognition after stroke, focusing on sensing technologies, target activities, settings, and model architectures. PubMed, Embase, WOS, and Scopus were searched and seventeen studies involving 224 individuals with stroke met the inclusion criteria. Inertial measurement units were the predominant sensing modality (12/17, 70.6%), with several studies demonstrating the feasibility of recognizing relevant functional activities using a single device. Single-sensor approaches were also used in home-based and simulated-home settings, suggesting potential for scalable and unobtrusive monitoring outside clinical environments; however, most studies were conducted in controlled clinical settings (11/17, 64.7%), with limited real-world validation to date. Convolutional neural networks were the most frequently used deep learning architecture, while Random Forest was the most common traditional machine learning method. Multimodal sensing may enrich activity characterization, although greater sensing complexity may reduce feasibility for prolonged home monitoring. Methodological and reporting limitations, including concerns regarding confounding and reproducibility, should be considered when interpreting the current evidence. Overall, single-sensor AI-based HAR appears promising for low-burden, continuous, and ecologically valid monitoring. Future research should prioritize longitudinal validation in everyday environments, transparent reporting of code and data availability, and inclusion of acute and severely impaired stroke populations to facilitate clinical translation toward objective monitoring, personalized rehabilitation, and remote care.

Artificial Intelligence Review
Context-Aware Activity Recognition Systems
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon