Web interface for a test battery for status assessments in advanced Parkinson

Abstract: The purpose of the thesis work is to design and construct a web-application that calculates and presents summaries for test battery data (Parkinson-related questions and on-screen motor tests namely different tapping tests and spiral drawings). The dataset includes about seventy patients at nine clinics in Sweden. Statistical methods, dimension reduction (unsupervised learning) techniques and knowledge elicitation methods have been applied about how data should be summarized and presented. This work presents a model which helps in assessing status of patients with advanced Parkinson. An overall score is calculated to represent the overall state of the patient in a test period. This score is composed of six dimensions based on summarized data. Expert Knowledge has been introduced to choose which questions to combine to form these dimensions. Within each dimension, calculated features were combined by Principal Component Analysis (PCA). To illustrate the patient severity, each dimension becomes a corner in a hexagon. Cumulative distribution function is used to normalise the values, so the scale direction is from worst or 0 (middle of the hexagon) to best or 1 (corner of the hexagon). A clinical rating score, Unified Parkinson Disease Rating Scale (UPDRS) is used to optimise weighting of the six test battery dimensions. The overall score is calculated using Neural Networks, Multiple Linear Regression (MLR) and the hexagon area approach. The MLR method was chosen since it gives good R-squared results and is easily interpreted. A user evaluation for the Overall Feedback System has been performed and the results showed to be satisfactory. Drawn spirals were displayed in a separate web application, and they were clinically rated by a clinician. Graphical presentations are included and different prototype versions have been iterated with users and user requirements specifications is produced.