FIRST PRINCIPAL DATA DRIVEN MODELS IN NEUROLOGICAL DISEASES WITH WEARABLE TECHNOLOGY: PARKINSON DISEASE AND POST STROKE

FIRST PRINCIPAL DATA DRIVEN MODEL FOR THE ASSESSMENT OF MOTOR DYSFUNCTION IN NEUROLOGICAL DISEASES WITH WEARABLE TECHNOLOGY PHD Candidate: Abdul Haleem Butt Supervisor: Prof. Filippo Cavallo Tutor: Angelo Maria Sabatini Co-tutor: Carlo Maremmani Ambient Assistive Living Lab The BioRobotics Institute Scoula Superiore Sant’Anna My name is Abdul Haleem Butt S/o Prof. Hafiz Muhammad Yasin Butt (Late). I am from Pakistan and basically Computer Engineer. I obtained my bachelor’s in computer engineering from University of Engineering and Technology Taxila Pakistan in 2008. After that i spend one year in Telecom company (China Mobile Pakistan Islamabad) as an Engineer. In obtained Master of Science in Computer Engineering (Artificial Intelligence) from Dalarna University Sweden in 2012. I spend two years as a Lecturer in Computer Science Department in The University of Lahore, Lahore Pakistan. Where i taught fundamental courses of computer science and engaged with the students with their final year projects as a supervisor. My main area of specialization is in Artificial Intelligence (Data Mining, Fuzzy Logic, Neural Network). My MSC thesis was under the title: “Speech Assessment for the classification of Hypokinetic Dysthria in Parkinson Disease”. http://www.diva-portal.org/smash/record.jsf?pid=diva2%3A524605&dswid=-1842 I enrolled as Doctoral candidate in Bio Robotics Institute “SERVICE ROBOTICS AND AMBIENT ASSISTED LIVING LABORATORY” Scuola Superiore Sant’anna in PHD cycle 2015/2016. I perused my PHD journey under the supervisor of Prof. Filippo Cavallo, Tutor: Prof. Angelo Maria Sabatini and Co-tutor: Dr. Carlo Maremmani. Main research topic is “First Principal Data Driven Models for the assessment of motor dysfunction in Neurological Disease with Wearable Technology “ In the first year of my PHD my focus was literature review for the novel machine learning techniques for the health care application most specifically in Parkinson Disease to assess the motor dysfunction in PD Patients. On the same time statically, analysis on the available Database to understand the statically and clinical significance of the recorded data from wearable sensors (SensHandV1 and SensFootV2). In the second year of PHD my focus was developed novel machine learning methods for the prediction of the PD, since the clinical scale such MDS-UPDRSIII does not tell anything about the disease progression. On the same time investigate other cheap technological solutions available for the assessment of the motor dysfunction in PD. In the reaming time till now i worked with post stroke patients in order to assess the motor performance of the post-stroke patients to provide the feedback for rehabilitation of the patients during daily living activities in home environment with inertial sensors device(SensHandv1). We have following collaborations:  Department of Translational Research of New Technologies in Medicine and Surgery, University of Pisa, Italy.  Departmental Section of Severe Acquired Brain Injuries - University of Pisa, Italy.  TIM WHITE (Wellbeing and Health Innovative Technologies) Joint Open Lab, Pisa, Italy.  Institute of Clinical Physiology, National Research Council, Pisa, Italy  Area Health Authority District 1, Neurology Operative Unit, Carrara, Italy  Laboratorio de análisis del movimiento Rhb.Funcional Institut Guttmann FUNDACIÓ INSTITUT GUTTMANN Camí de can Ruti s/n, 08916 Badalona  Eurecat The Technology Centre of Catalonia ABSTRACT This study focuses on the development of an objective, automated method to extract clinically useful information from the motor symptoms in the context of Neurological Diseases such as Parkinson Disease (PD) and for the Post-Stroke Rehabilitation. The motor symptoms of PD vary with the disease progression and occur in varying frequency and duration. In clinical studies, symptoms are assessed using the unified Parkinson’s disease rating scale (UPDRS). On one hand, the subjective rating using UPDRS relies on clinical expertise. On the other hand, it requires the physical presence of patients in clinics which implies high logistical costs. Moreover, for a patient with motor fluctuations, a single clinical exam is not sufficient to evaluate the spectrum of motor impairment that a patient may experience over the course of a typical day. There is need a toll to monitor patients in home environment to monitor disease progression. Stroke leads to losses of various body functions depending on the affected area of the brain and leave significant impacts to the victim’s daily life. Post-stroke rehabilitation plays an important role in improving the life quality of stroke survivors. They receive feedback to maintain a correct posture as well as other feedback from therapists during rehabilitation. Objective and automated methods can allow the patients to get feedback in home environment during the daily living activities. The aim of the study focused one side to investigate the clinical significance of the extracted features from wearable inertial sensors (SensHandv1 and SensFootV2) and optical sensor (Leap Motion Controller). On the other side investigated the novel machine learning techniques for the correct objective assessment of the disease focusing on the binary classification problems in PD and Post-Stroke. In the first part of the study comprehensive framework was developed to analyze the extracted biomechanical features performance for the classification of the two group of patients belongs to different level of Parkinson disease Slight-Mild (SM) and Moderate-Severe (MS). This study purpose novel schemes to improve the overall classification accuracy using robust feature selection methods to obtain a best subset of the features and with state of the art machine learning algorithms. To contributing to better diagnosis and monitoring, better characterization might also lead to a better understanding of the disease processes underlying neurodegenerative conditions. The features were selected with their best classification performance. The results strongly encourage to adopt the developed scheme to improve the overall sensitivity and specificity of the classification. The maximum obtained classification accuracy between SM and MS was 83.10%, with 0.889 area under the curve with Neural Network developed classifier. In second part of the study classification between healthy control and PD subjects was performed with novel feature selection scheme using semi-supervised learning technique Adaptive Neuro Fuzzy C-Mean. The aim of this study was fourfold: (a) first to evaluate the significance of the extracted biomechanical parameters from each exercise with ANIS model and, b) evaluate the potential of selected features from upper and lower limbs in the mean of knowledge representation by ANFIS and, c) third to make evidence that the fusion of most significant features from upper and lower limbs improve the overall classification accuracy and, d) the ANFIS model has ability to incremental update of the data, which could use in an incremental situation for disease prediction. The results endorsed the potential of the ANFIS model to be implemented as an intelligent system for prediction. Third study was for the classification of the PD and healthy subjects were based on the extracted parameters from Leap Motion Controller. One of our main goal in this study was to investigate metrices that could be extracted from a MDS-UPDRSIII tasks, and to confirm its merit when combined with the other well-established metrics included in our analysis. During our treatment of the data with machine learning techniques, we opted against the complexity of re-introducing feature selection inside the classification training and assessed the performance of the individual feature subsets. Multiple feature selection methods applied to find out the best suitable subset of the features in order to improve the overall classification accuracy. The obtained results support the fact that most of the set of selected features contributed significantly to classify the PD and healthy subjects. Results revealed that the system did not return clinically meaningful data for measuring postural tremor in PwPD. In addition, it showed limited potential to measure the forearm pronation/supination. In contrast, for finger tapping and hand opening/closing, the derived parameters showed statistical and clinical significance. Our finding showed that LMC has a limited angle of view to assess the motor dysfunction in PwPD. Certain gestures could not be recorded properly depending on the placement of the Leap Motion controller. In general, this study revealed the functional limitation of LMC-provided SDK. There is need improvements in SDK algorithm accuracy and usability are required. In the fourth study on Post-stroke patients aims the classification between purposeful movements and non-purposeful movements in daily living activities. This study focuses on stroke survivors with upper-limp impairment and propose an event detection approach aimed at monitoring arm movements through inertial sensors (SensHandv1). Besides walking (non-purposeful movement), nine activities (purposeful movement): resting; eating; pouring water; drinking; teeth brushing; folding towel; grasp toothbrush; grasp bottle; grasp towel. The preliminary results showed the highest classification between purposeful and non-purposeful movements with both healthy and post-stroke subjects. The highest obtained classification accuracy was 80.0% with Naive Bayes classifier with 0.90 Receiver operating characteristics curve (ROC). There is need to investigate other features and machine learning methods to classify different purposeful activities and their performance in order to assess the improvement in the rehabilitation process. List of Publications 1. Butt, Abdul Haleem, et al. "Biomechanical parameter assessment for classification of Parkinson’s disease on clinical scale." International Journal of Distributed Sensor Networks13.5 (2017): 1550147717707417. 2. Butt, Abdul Haleem, et al. "Leap motion evaluation for assessment of upper limb motor skills in Parkinson's disease." 2017 International Conference on Rehabilitation Robotics (ICORR). IEEE, 2017. 3. Butt, Abdul Haleem, et al. "Wearable Sensors for Gesture Analysis in Smart Healthcare Applications." Human Monitoring, Smart Health and Assisted Living: Techniques and Technologies 9 (2017): 79. 4. Butt, Abdul Haleem, et al “A Design of Adaptive Neuro-Fuzzy Inference System (ANFIS) using Fuzzy C-Means Clustering for the Diagnosis of Parkinson Disease based on Biomechanical Parameters” (Submitted in Journal Computers in Biology and Medicine) 5. Butt, Abdul Haleem, et al “A DESIGN OF HYBRID INTELLIGENCE SYSTEM FOR THE DIAGNOSIS OF PARKINSON DISEASE BASED ON BIOMECHANICAL PARAMETERS” (To be submit soon in Journal: Medical Engineering & Physics) 6. Butt, Abdul Haleem, et al “OBJECTIVE AND AUTOMATIC CLASSIFICATION OF PARKINSON DISEASE WITH LEAP MOTION CONTROLLER” (Submitted in Journal: BioMedical Engineering OnLine) 7. Butt, Abdul Haleem, et al “ASSESSMENT OF PURPOSEFUL MOVEMENTS FOR POST STROKE PATIENTS IN DAILY LIVING ACTIVITIES WITH INERTIAL SENSORS DEVICES” (To be submit soon in Sensors — Open Access Journal)