Artificial intelligence tools for engagement prediction in neuromotor disorder patients during rehabilitation.
Simone Costantini, Anna Falivene, Mattia Chiappini and 7 others
PMID 39702317WHAT IT FOUND
AI models predicted patient engagement during robot-assisted gait rehab with high accuracy.
Combining heart rate and sweat signals outperformed single signals. Data augmentation further improved prediction. This supports real-time monitoring to adapt therapy intensity, but clinical utility is not yet proven.
Key findings
01The best models achieved macro-averaged F1 scores of 95.6% for self-perceived engagement and 95.4% for therapist-perceived engagement.
02Using a bimodal dataset combining heart rate variability and electrodermal activity features outperformed unimodal datasets for both engagement types.
03Data augmentation with 3-minute windows significantly improved model performance compared to the baseline bimodal dataset.
STILL TO COME
How it was doneWhat they foundWhat it means for PTsWhat it means for OTsWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The study is a technical validation of AI models, not a clinical trial. It does not prove that using these predictions improves patient outcomes. The dataset was small (46 subjects) and highly unbalanced, with very few 'Underchallenged' sessions. Engagement labels were subjective, relying on questionnaires and expert review, which may not capture dynamic changes during a session. Physiological data contained motion artifacts, which were manually cleaned; this may not be feasible in real-world noisy clinical settings. The high age variance (children and adults) may affect the generalizability of physiological features.
Declared interests
Funded by the Italian Ministry of University and Research and the Italian Ministry of Health. No commercial conflicts of interest declared.
The easy way to misread this
Do not interpret the high accuracy scores (95.6% and 95.4%) as evidence that this AI tool is ready for clinical use. These scores reflect the model's ability to classify engagement in a controlled, cleaned dataset, not its ability to improve patient care or function in real time.