Early Stage Detection of Parkinson's Disease using AI

Abstract: Parkinson’s disease (PD) is the second most common neurodegenerative disease. The diagnosis of PD is inaccurate a quarter of the time and diagnosed at a later stage. Early symptoms of PD are similar to many other diseases or of healthy aging. If PD is detected in its early stage, there are several treatment options available. This research is aimed at early detection of PD, with a noninvasive, easy to administer and inexpensive test. In this study, advanced machine learning algorithms were created and then trained with existing voice data from people with and without PD. Using voice samples, two all-encompassing statistical measures are derived called Pitch Period Entropy (PPE) and Spread1. Three different kind of algorithms are developed from scratch. The first algorithm uses Logistic Regression, the second uses an Artificial Neural Network (ANN) and a third one uses Support Vector Machine. All the models output F2 scores of .93 and more. Further research was conducted by differentiating PPE and Spread1 values of people with PD for less than 4 years and people with PD for greater than 4 years. Both groups were found to be indistinguishable suggesting that PPE and Spread1 could be used for early-stage detection. This detection can also be administered remotely making it suitable for anybody around the world. This can be used as part of regular health checkups for the susceptible populations of PD as this is both affordable and non-invasive. This has the potential to revolutionize the diagnosis and treatment of PD and increase the quality of life of PD patients.