Abstract

Parkinson’s disease (PD) is a progressive disorder of slow progress of the nervous system produced by the absence of levels of dopamine, which can incite unrestrained instinctive movements of the body and psychological affections. Physicians usually clinically diagnose the disease according to their abilities and experience. But, due to the subjectivity of the diagnosis, erroneous diagnoses and treatments can occur. For the development of a practical,  low-cost, and general diagnosis system of the symptoms to support Parkinson’s disease patients, it is necessary to define the methods to be accomplished first, the sources of data information, and the algorithms that allow performing the analysis of the symptoms of the disease and also medication adherence monitoring.  Artificial intelligence techniques in medicine, specifically in the advanced diagnosis of Parkinson’s disease, has been demonstrated to be very effective and efficient.  Particularly, for the diagnosis and classification of Parkinson’s disease patients, the Unified Parkinson’s Disease Rating Scale (UPDRS) is used, which requires the patient to perform a series of tests among which the biomarkers of speech, tremor, and walking analysis are considered, after which the physician diagnoses whether the patient suffers from Parkinson’s disease or not according to the score obtained.
This work proposes an information management system for analyzing symptoms and support of Parkinson's patients using smartphones and their automatic classification according to the data collected by the smartphone’s motion controller sensors under the information processing pipeline abstraction paradigm. The system starts from the presupposition that Parkinson’s disease patients have different abnormalities when they do not follow the required medication or the Disease Rating Scale level is changed. To develop a practical, low-cost, and general diagnosis system of the symptoms to support PD patients, it is necessary to implement an IoT health monitoring system that uses smartphones for data collection. The smartphone collects the data passively while the patient is using it. After that, the data preprocessor helps extract the information these  Parkinson-related biomarkers contain. A deep reinforcement learning algorithm with a selective attention mechanism using multi-modality sensor data classification with selective attention and long short-term memory is proposed for classification and medication adherence monitoring to develop a person-centered protocol capable of autonomous performance. It is implemented using a hybrid of Active Learning and LSTM algorithms in the KNIME platform. Finally, in perspective, is planned use of the obtained data to create models of Parkinson´s disease that include organisms and artificial systems of different organizational levels to support diagnostics and therapy. With this work, we demonstrate the viability of passive medication adherence monitoring for Parkinson’s disease patients, the dynamic treatment regimens, and the development of an autonomous person-centered protocol for better patient medication. Safety and ethical considerations were taken into account when designing the system.