Affective Computing in Serious Games for Physical Rehabilitation: Scoping Review.

Only 9 studies met the criteria, and the authors say that is too few to draw firm conclusions about how affective computing is used in rehabilitation games. They describe the trends as preliminary. The included studies were tiny, typically 1 to 11 participants, and recruited from single centres. None reported random or consecutive recruitment, so 8 of 9 were judged at high risk of bias for how patients were chosen. In 8 of 9 studies, emotions were labelled by experts watching video recordings. It is therefore unclear whether the software is measuring emotion or simply reproducing the observers' own judgements. Most of the work happened in controlled research laboratories, not real clinics, so it does not reflect the movement, noise and patient variability of a normal therapy session. The search covered only English-language, peer-reviewed papers and excluded grey literature, so relevant work published elsewhere may have been missed. Most studies only detected emotions. They did not test whether responding to that information changes anything for the patient.

The authors declared that no financial support was received for this work and that they have no conflicts of interest.

Do not read this as evidence that emotion-sensing rehabilitation games work or are ready to use with patients. Only 2 of the 9 studies changed anything in the game in response to the player's state, none of them tested whether doing so improves rehabilitation, and the review itself says the evidence for clinical efficacy remains inconclusive.