Explainable AI in Healthcare: Interpretable Detection of Parkinson's Disease from Mobile-Collected Handwriting Patterns

Parkinson's disease (PD) is a neurodegenerative disease that developed over time due to the progressive loss of neurons, which are responsible for producing dopamine. Dopamine is a neurotransmitter that helps to regulate motor functions. Deficiency of dopamine causes symptoms such as tremors, rigidity, and slowness of movements. In PD, the brain cells in the substantia nigra region responsible for the production of dopamine slowly die off. There are several ways to identify PD, such as gait analysis, neuroimaging techniques, and voice pattern recognition. Additionally, PD can be effectively detected using handwriting by examining different patterns such as spirals, digits, waves, and other structured drawing tasks. In this work, we proposed a PD detection method using an ensemble machine learning (ML) approach i.e. Extra Tree (ET) classifier, combined with explainability techniques. The model utilized handwriting patterns including spirals, loops, meanders, and digits collected using mobile devices providing accurate predictions and interpretable insights into key features influencing the classification. To increase the model performance and robustness, data augmentation techniques such as rotation, zooming, width and height shifting, and horizontal flipping were applied to the image data, complemented by preprocessing steps to normalize and structure the data. These steps significantly contributed to achieving an accuracy of 89.27%. The combination of preprocessing, augmentation, and the robust ML model ensures reliable and interpretable detection of PD based on handwriting patterns. Early detection may help specialists provide individual treatment plans to reduce the wearing-off (WO) phenomenon.