Diagnosis of Parkinson’s Disease through Spiral Drawing Classification via VGG19 and AnoGAN
Abstract – Millions of people worldwide suffer from Parkinson’s disease, a neurodegenerative disease that comes with symptoms such as tremors and difficulty in bodily function. Around 60,000 new cases arise annually, and yet ironically, no standardized way of diagnosing Parkinson’s exists to this day. To enable a more effective process of medication, therapy, and treatment, early diagnosis of PD through a standardized measure is crucial. Previous attempts of achieving this include the analysis of MRI data, speech signals, and the drawing of spirals. Our objective in the study was to enhance the spiral drawing test by combining a CNN model (VGG19) and an anomaly detection system, AnoGAN. The initial trial with the VGG19 model produced an accuracy score of 94%, higher than that of an existing study which produced an accuracy score of 88%. To further refine this algorithm, we incorporated AnoGAN, which allowed for more effective detection of anomaly data and enabled the process of cross-checking and feedback, increasing the accuracy further. This algorithm produced an anomaly score for each of the spiral drawings in the data set; a larger score indicated a higher likelihood for the data to belong to a PD patient. Though we have not yet established an explicit standard for the minimum anomaly score one must have to be diagnosed of PD, this will be possible once more data is accumulated and trained in the algorithm. Overall, the high accuracy and the standardized nature of this design evince its possible application in hospitals in the future.