Attention-Based Gated Recurrent Networks for Parkinson’s Disease Detection Through Voice Patterns

Parkinson's disease, which primarily affects adults over 60 years old, is the second most prevalent neurological condition. It causes an extensive effect on the daily life and healthcare causing severe problems in motor and non-motor control behaviors. It has been a stimulating area of research to correlate non-motor disorders such as voice and speech problems, breathing patterns, mental disorders, sleep disorders, etc. for Parkinson’s disease classification. The motto is to provide in future an efficient checklist of various indicators of Parkinson’s disease to support diagnosis, assessment procedures leading to a Parkinson's disease early diagnosis. Artificial intelligence and machine learning have been efficiently proven to accelerate value-based healthcare provided to the patients by the medical professionals. With this motivation, we have proposed an Attention-based Gated Recurrent Unit model for Parkinson’s Disease Detection (AGRU-PDD) through voice phonations. The comparative analysis demonstrates that the AGRU-PDD model proposed outperforms the learning models LSTM and GRU providing an average accuracy of 0.977 and without considering the voice phonations of digits the accuracy alleviated to 0.982. The proposed model showcases its capabilities and promises to be a further research direction for further refining and customizing the attention-based gated recurrent unit which uses voice patterns toward efficient detection and classification of Parkinson’s disease.