[4L1-GS-10] AI application: Medicine / Healthcare

[4L1-GS-10-01] Detection of Parkinson’s Disease from the Audio Modality Based on Ensemble Learning

〇Yuki Ito¹, Shohei Kato¹, Takuto Sakuma¹, Reiko Ohdake², Michihito Masuda³, Shinji Ito², Hirohisa Watanabe² (1. Nagoya Institute of Technology, 2. Fujita Health University, 3. Nagoya University)

Keywords:Parkinson's Disease, Screening, Machine Learning, Voice Analysis, Ensemble Learning

In recent years, there has been a notable increase in the number of Parkinson’s disease (PD) due to rapidly aging population. Early detection is desirable because the progression of the disease makes it difficult to perform daily activities. The current challenge is that the diagnosis is a heavy burden for patients, and early diagnosis is difficult. In this paper, we propose a PD discrimination model from audio analysis, specifically targeting speech disorders associated with PD. Spontaneous speech task responses were recorded from 134 PD patients and 94 healthy controls (HC). In each of the speeches, we extracted 6,373 acoustic features, ComParE 2016 feature set to capture prosodic features, 17 linguistic features and 4 temporal features. Features were selected by forward stepwise selection based on AIC. We constructed an ensemble learning model with SVM as weak learners to discriminate between PD and HC. As a result, our model produced a F-measure of 0.94 and a sensitivity of 0.95. The proposed method using the audio modality has shown applicability to the easy screening of Parkinson’s disease.