A Preliminary Study on Assessing Patient Behavioral Quality Using Deep Learning

Monitoring behavioral quality in patients represents an essential task for making diagnoses, providing effective and prompt interventions, and designing appropriate countermeasures. This implies the need to perform recurring analyses of patient movements in order to capture possible declines and/or symptoms of diseases during the evaluation period. Therefore, collecting a sufficient number of video samples is pivotal. Deep Learning models offer great flexibility and capability to handle large amounts of data. In this work, we investigate the possibility of assessing behavioral quality in patients through an advanced system that extracts information via a camera and skeleton tracking software, and categorizes behaviors using a Deep Neural Network model. Patient behaviors were labeled by a human experimenter, who manually assigned a score to each video. Preliminary results indicate that the model performs well in terms of the evaluation of behavioral quality. Moreover, the outcomes highlight that sensitivity to manual labeling plays a non-negligible role in this context.