Vision Transformer Based Approach for Parkinson's Disease Diagnosis

Parkinson's Disease is one of the most recurrent neurological disorders and have been a topic of research for many years. This neurological disease causes several impairments, that affects the daily activities of the individuals suffering from it. Since there has not been any known cures and proven underlying causes that lead to this disease, diagnosing it at its early stage becomes a hurdle. Solutions powered by AI have been proven useful for potentially detecting the disease at its early occurrence. This can be achieved by observing different patient characteristics such as voice samples, hand-drawings, clinical data and medical images, which are formulated in form of structured datasets. In-spite of research progress, there have been issues that are yet to be addressed: a.lack of models trained on datasets formed from patients' hand-drawings b. lack of attention-based modelling for diagnosing this disease c. lack of study utilising pre-trained powerful vision transformers. This work proposes a solution to all these issues by employing a powerful vision transformer model extracting global features from hand-drawn spiral image dataset and finally classifying it using the linear classification heads. During the evaluation phase the model has outperformed the existing best solutions for the considered performance metrics.