DECISION TREE-BASED PARKINSON'S DISEASE DIAGNOSIS: A MOBILE-FIRST APPROACH

Abstract

Accurate early detection of Parkinson's disease (PD) is essential to halt its progression and provide 
patients with access to disease-modifying medications. This study focuses on monitoring the premotor stage of 
PD to achieve early diagnosis using a novel deep-learning method. The proposed method quickly determines the 
presence of PD based on premotor traits, leveraging a range of indicators such as Rapid Eye Movement (REM) 
sleep behaviour disorder, olfactory loss, cerebrospinal fluid (CSF) data, and dopaminergic imaging markers.A 
comparative analysis was conducted between the proposed deep learning model and twelve other machine 
learning and ensemble learning techniques using a sample size of 183 healthy individuals and 401 early PD 
patients. The results highlight the superior detection performance of the designed model, which achieved an 
average accuracy of 96.45%, the highest among the tested methods.Additionally, the study employs a Boosting 
approach to provide feature importance in the PD detection process, offering insights into the most significant 
indicators of early PD. This information is critical for understanding the model's decision-making process and 
for further refining detection techniques. 
The developed system includes both a PC website and a mobile website to enhance accessibility and usability. 
Utilizing Streamlit, the program provides a local host address and a network host address to facilitate the 
connection and execution of the website on mobile devices. This ensures that users can access the PD detection 
tool seamlessly across different platforms.By integrating these features, the study aims to provide a robust and 
user-friendly solution for the early detection of Parkinson's disease, ultimately contributing to better patient 
outcomes through timely intervention. Accurate early detection of Parkinson's disease (PD) is essential to halt its progression and provide 
patients with access to disease-modifying medications. This study focuses on monitoring the premotor stage of 
PD to achieve early diagnosis using a novel deep-learning method. The proposed method quickly determines the 
presence of PD based on premotor traits, leveraging a range of indicators such as Rapid Eye Movement (REM) 
sleep behaviour disorder, olfactory loss, cerebrospinal fluid (CSF) data, and dopaminergic imaging markers.A 
comparative analysis was conducted between the proposed deep learning model and twelve other machine 
learning and ensemble learning techniques using a sample size of 183 healthy individuals and 401 early PD 
patients. The results highlight the superior detection performance of the designed model, which achieved an 
average accuracy of 96.45%, the highest among the tested methods.Additionally, the study employs a Boosting 
approach to provide feature importance in the PD detection process, offering insights into the most significant 
indicators of early PD. This information is critical for understanding the model's decision-making process and 
for further refining detection techniques. 
The developed system includes both a PC website and a mobile website to enhance accessibility and usability. 
Utilizing Streamlit, the program provides a local host address and a network host address to facilitate the 
connection and execution of the website on mobile devices. This ensures that users can access the PD detection 
tool seamlessly across different platforms.By integrating these features, the study aims to provide a robust and 
user-friendly solution for the early detection of Parkinson's disease, ultimately contributing to better patient 
outcomes through timely intervention.