Deep Learning Based Parkinson’s Disease Classification Using Angular Orientations

Convolutional Neural Networks (CNN) are proficient in extracting essential features for the recognition and classification of various diseases. However, the intricate symptoms associated with changes in brain anatomy present challenges during CNN training. While an ideal approach would involve using a patient’s complete Magnetic Resonance Imaging (MRI) data with minimal preprocessing and human intervention, it doesn’t always yield optimal results. In such cases, researchers often resort to employing larger and more complex networks to enhance CNN performance, but this doesn’t guarantee improvement. In this paper, we introduce an innovative method to enhance performance by incorporating multiple distinct 3D orientations of the data within a multi-classifier framework. The approach involves predictions from networks trained on unique angular orientations of the same dataset, combining them to provide a unified prediction. Results from this proposed method highlight that these minimalistic, computationally efficient adjustments can elevate average accuracy rates from 82.07% to a commendable 88.05%, representing a 6% performance boost.