Parkinson’s disease (PD) impacts more than 10 million people worldwide. However, the conventional method of diagnosing PD based on physicians’ decision is time-consuming and often inaccurate. Hence, we aim to develop an accurate PD diagnosis model based on subtle voice impairment. With the help of deep learning, we have successfully increased the classification accuracy and efficiency by eliminating complex feature extraction and selection steps necessary in machine learning. Furthermore, we propose to implement Generative Adversarial Network to standardize parameters such as the volume of voice and the quality of the recording whcih may confound the diagnosis. The revised inputs, when put into a network particularly trained with the standard data, resulted in higher accuracy.

Contestants

Kyuhee Jo

Chaeyoung Lee