phat 15(2): e1

Research Article

Detecting affective states in virtual rehabilitation

Download163 downloads
  • @ARTICLE{10.4108/icst.pervasivehealth.2015.259250,
        author={Jes\^{u}s Rivas and Felipe Orihuela-Espina and L. Sucar and Lorena Palafox and Jorge Hern\^{a}ndez-Franco and Nadia Bianchi-Berthouze},
        title={Detecting affective states in virtual rehabilitation},
        journal={EAI Endorsed Transactions on Pervasive Health and Technology},
        volume={1},
        number={2},
        publisher={IEEE},
        journal_a={PHAT},
        year={2015},
        month={8},
        keywords={affective computing, virtual rehabilitation, stroke, hand, motor recovery},
        doi={10.4108/icst.pervasivehealth.2015.259250}
    }
    
  • Jesús Rivas
    Felipe Orihuela-Espina
    L. Sucar
    Lorena Palafox
    Jorge Hernández-Franco
    Nadia Bianchi-Berthouze
    Year: 2015
    Detecting affective states in virtual rehabilitation
    PHAT
    EAI
    DOI: 10.4108/icst.pervasivehealth.2015.259250
Jesús Rivas1,*, Felipe Orihuela-Espina1, L. Sucar1, Lorena Palafox2, Jorge Hernández-Franco2, Nadia Bianchi-Berthouze3
  • 1: Instituto Nacional de Astrofísica, Óptica y Electrónica (INAOE)
  • 2: Instituto Nacional de Neurología y Neurocirugía (INNN)
  • 3: University College of London (UCL)
*Contact email: jrivas@ccc.inaoep.mx

Abstract

Virtual rehabilitation supports motor training following stroke by means of tailored virtual environments. To optimize therapy outcome, virtual rehabilitation systems automatically adapt to the different patients’ changing needs. Adaptation decisions should ideally be guided by both the observable performance and the hidden mind state of the user. We hypothesize that some affective aspects can be inferred from observable metrics. Here we present preliminary results of a classification exercise to decide on 4 states; tiredness, tension, pain and satisfaction. Descriptors of 3D hand movement and finger pressure were collected from 2 post-stroke participants while they practice on a virtual rehabilitation platform. Linear Support Vector Machine models were learnt to unfold a predictive relation between observation and the affective states considered. Initial results are promising (ROC Area under the curve (mean+/-std): 0.713 +/- 0.137). Confirmation of these opens the door to incorporate surrogates of mind state into the algorithm deciding on therapy adaptation.