Multimodal I-vectors to Detect and Evaluate Parkinson's Disease

Parkinson's Disease (PD) is a neurodegenerative disorder characterized by a variety of motor symptoms. PD patients show several motor deficits, including speech deficits, impaired handwriting and gait disturbances. In this work we propose a methodology to fuse i-vectors extracted from three different bio-signals: speech, handwriting and gait. These i-vectors are used to classify Parkinson's Disease patients and healthy controls and to evaluate the neurological state of the patients. Speech i-vectors are extracted from MFCCs, handwriting i-vectors are extracted from kinematic features and gait i-vectors are extracted from modified MFCCs computed from inertial sensor signals. Two fusion strategies are tested: concatenating the i-vectors of a subject to form a super-i-vector with information from the three bio-signals and score pooling. The proposed fusion methods leads to better classification results respect to the separate analysis with each bio-signal, reaching an accuracy of up to 85%.