Detection of Parkinson's Disease Through Spiral Drawings Using Convolutional Autoencoders

Parkinson disease is a chronic neurodegenerative disease where parts of the brain become progressively damaged affecting over 10 million people worldwide. It is primarily characterized by the degeneration of dopaminergic neurons in the brain which generate striatal dopamine deficiency, and intracellular inclusions containing aggregates of a-synuclein, leading to motor symptoms such as tremors, bradykinesia, stiffness, and postural instability. These motor impairments are often reflected in tasks, such as handwriting and drawing, making them valuable tools for early diagnosis. Despite advancements in medical imaging and other diagnostic techniques, the early detection of Parkinson's disease remains challenging, often relying on clinical assessments that can be subjective and inconsistent. Our purpose in this paper is to use machine learning to diagnose Parkinson through only spiral and wave drawings. We will utilize Convolutional Autoencoders (CAE) a deep learning model designed for unsupervised feature extraction and representation, to analyze these drawings. The dataset is obtained from open source Kaggle.com, which includes 102 drawings belonging to healthy and Parkinson individuals, offers a comprehensive basis for training and assessing the model. The dataset was also split into 3 sets: training, testing and validating, with 80%, 10%, and 10% of the data. To the best of our knowledge, the exclusive use of spiral and wave drawings for diagnosing Parkinson's disease through Convolutional Autoencoders (CAEs) has not been previously explored, positioning this study as a novel contribution to the field. By leveraging this unique dataset, our research aims to develop a more accessible and efficient diagnostic tool that surpasses traditional methods in terms of simplicity, automation, and scalability.