Motion sensor network based on low power wireless technologies for the assessment of neurological diseases

In the last years, the spread in the market of miniaturized magneto-inertial sensors, thet are so small, low cost and power efficient to be attachable to any system (human body included), has expanded their potential in a myriad of applications. Since their introduction on the market, bioengineers have exploited their potential for the motion analysis in healthy subjects, but also in patients for the assessment of pathologies characterized by motor disfunctions, such as neurological diseases. One of the main challenges related to the use of those sensors in biomedial applications is to develop an user friendly, low power, low cost and high throughput M-IMU systems which allow monitoring patients with motor impairments and acquiring reliable data to support clinical decisions. In fact, despite the large number of commerical devices based on magneto-inertial sensors available on the market, their use for clinical assessment in unstructured environments (e.g. at-home) is still limited. On one side most of the portable system coming from the consumer electronics market (smartbands or smartwatches) allow for pervasive, low-power communication with smart devices (thus usable at home), but do not permit a high throughput and multi-node information, but are quite expensive, usable only by trained staff and often have limited battery life capabilities (since are not based on low energy wireless communication) or use non-standard and widespread communication technology, making them suitable only for use in structured or semi-structured environments (i.e. research laboratories, research hospitals). To overcome these limitations, the overall aim of this PhD project is to design, test and validate a M-IMU based wireless sensor network, for motor assessment of neurological patients in unstructrured environments. To do this, our first goal is to select and test a suitable wireless technology, with reduced power consumption and wide compatibility with consumer electronics for out-of-the-lab and at-home applications. We found Bluetooth Low Energy (BLE) to be an excellent candidate for our requirements and we performed for the first time an extensive and systematic analysis of BLE, in order to demonstrate whether it is a suitable candidate for wireless communication in high throughput and low energy sensor network applications. Defining a standard methodology to test wireless seonsor networks, we tested BLE performance in a sensor network with a star topology, showing the possibility to use this communication technology in high datarate applications (~170 kbps), thus enabling the streaming of 5 nodes with 9-axes M-IMU data (16 bits of resolution per axis) at more than 200 Hz. The second goal of our work id to assess the reliability of M-IMU sonsors for motion analysis. In fact, despite their pervasive use in biomedical applications, the assessment of their accuracy and reliability for motion tracking has been only partially addressed. Our main contribution in this field is focused on the design of a standard methodology to tune and optimize filter parameters in M-IMU sensor fusion algorithms, able to measure the error with respect of a ground truth (represented by the optical motion tracking system), in order to evaluate the goodness of these filters. The results show how filters' performance may be different, depending on an orientation error of about 2° and lower than 5°, respectively in static and dynamic conditions. Finally, our third goal is to use a reliable technology in terms of both wireless communication and sensor technology, to objectively assess motor conditions of patients with neurological deficits. Among several possible pathologiew with relevant movement disorders, we focus on Parkinson's disease. In fact, PD may hyghly benefit from the use of magneto-inertial sensors for the diagnosis and assessment of its motor symptoms. As a matter of fact, motor symptoms in PD are not only hyghly relevant, but they are also very representative of the evolution of the pathology. For this reason they are strictly related to diagnosis, assessment and monitoring; indeed, a very important part of the widely used PD assessment clinical scale (i.e. the UPDRS scale), administred by physician, is focused on the evaluation of motor disorders. The clinical analysis, which is occasionally administred by the doctor, presents some limits: PD has fluctuations during different days, but also in the same day, whhich do not guarantee the efficacy of the medical assessment in the hospital. Hence, we decide to continuously monitor PD patients using M-IMUs: in detail, we use these typology of sensors to evaluate PD's cardinal motor symptoms in different tasks of its clinical scale. As result, our main contribution in this topic is to exploit a M-IMU sensor network to discriminate fluctuations in subjects with Parkinson's disease (i.e. ON and OFF status), and to detect statistically significant differences between PD and healthy subjects with a few number of sensor and performing easy tasks. After a deep analysis of some of the UPDRS tasks administred by the physician to the patients, we evidenced statistically significant results to assess PD. The first analysis regards patients during the execution of arm prono-supination task and evaluating the total power as a kinematic feature on only one sensor positioned on the index, thumb or wrist. These sensor locations and kinematic index can detect statistically significant differences between PD and healthy subjects (index, thumb, wrist: p <0.0001). Moreover, we also found that using one M-IMU only on the trunk while performing a sit to stand task and evaluating trunk acceleration during trunk flexion it is possible to differentiate OFF and ON with respect to healthy subjects (p <0.05).