Parkinson Disease Detection using Deep CNN Model

Abstract

Prompt diagnosis of PD is important in order to provide patients with appropriate treatment and information on prognosis. However, an accurate early diagnosis can be challenging because the movement symptoms can overlap with other conditions. Doctors make the diagnosis of PD based on clinical evaluation, interpreting information gained predominantly through history-taking and examination of the patient. Sometimes brain imaging may be requested to help support the clinical diagnosis, but there are currently no tests that are wholly sensitive or specific for Parkinson’s. The rate of misdiagnosis of PD is approximately 10–25%, and the average time required to achieve 90% accuracy is 2.9 years. Autopsy is still the gold standard for the confirmation of the disease. Therefore, this project designed an advanced convolution neural network model to predict Parkinson disease from both image and voice data. In general, existing ML algorithms such as SVM, and Random Forest will not filter data multiple times so its prediction accuracy is less hence CNN is used in this project, which filter data multiple times using neuron values so its prediction accuracy can be better. This project uses WAVE and SINE images of normal and Parkinson disease patients for imaging data and UCI Parkinson recorded voice is used for voice samples.