Integrating a Quantitative Approach to Deep Learning for Patient Monitoring

Patient monitoring is of utmost importance in rehabilitation scenarios in order to ensure adequate interventions and therapies, particularly for the elderly. Remote patient monitoring and teleoperation represent widespread techniques for daily assisting people in their homes, with the possibility to collect significant amounts of data and perform real-time analysis. However, the acquisition of periodic and structured information raises a fundamental question about the handling capabilities of traditional systems and emphasizes the need for cutting-edge systems. Deep Learning (DL) has emerged as a groundbreaking tool for processing tons of data and rapidly extracting information, notably improving the potential for disease diagnosis and providing effective treatments. In this work, we propose a quantitative approach for patient monitoring by assessing their walking behaviors. Specifically, we introduce a set of metrics used to assign a quantitative score to each behavior and train a DL model for the ability to estimate the quality of the observed behaviors. Our preliminary investigation indicates that the approach is feasible and generates good performance, which could represent a valuable tool for clinicians and medical operators. However, the definition of metrics must be carefully fine-tuned in order to mitigate the risk of grouping completely different behaviors under the same category.