Objective: The aim of this study was to build PD diagnostic models based on deep learning approaches using technologies including handwriting identification, automatic speech recognition and video processing.

Background: At present, the diagnosis of Parkinson’s disease (PD) mainly depends on the experience of neurologists. For the early identification of PD, there is still a lack of objective and effective methods. Approaches based on deep learning to classify PD patients and healthy subjects have recently been receiving more and more attention.

Method: A total of 77 participants were enrolled, including 47 PD patients(25females), 7 patients diagnosed with movement disorders except for PD and 27 age-matched healthy controls. A smartphone was used to record the handwriting pictures about the given sentences of patients, the sustained Mandarin Chinese vowels (a, o, e, i, u, ü) and the ambulatory videos, in which the participants were ordered to walk 5 meters forwards in a straight line, then to turn around and to walk back to the beginning position. Deep learning method based on convolution neural networks was adopted to assess  PD patients, controls and patients with other movement disorders.

Results: Using handwriting, audio, or video, the accuracy rates of predictive differential diagnosis(distinguishing PD patients from patients with other movement disorders) were 88.90%, 88.89% and 99.71%, respectively; the accuracy rates of predictive diagnosis(distinguishing PD patients from healthy controls) were 61.50%, 61.54% and 99.64%. Besides, the accuracy for distinguishing patients with movement disorders(including PD patients) from healthy controls was 64.30%, 64.29% and 98.25%, respectively.

Conclusion: The approach based on deep learning we found is highly accurate to classify PD patients and healthy subjects. It can be developed as a screening tool for diagnosis and assessment of PD in the early stage.

« Back to 2022 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/deep-learning-approaches-for-recognition-of-parkinsons-disease-patients-through-handwriting-audio-and-video-analysis/