Detection of Parkinson’s disease using MRI images using Auto encoders and Deep-learning models

Abstract

PD (Parkinson’s disease), according to the World Health Organization (WHO), is a movement-related neurodegenerative brain illness that manifests as tremors, insomnia, behavioral issues, and diminished sensory awareness. Artificial intelligence, machine learning, and deep learning have been applied in a recent study (2015–2024) to improve Parkinson’s disease (PD) diagnosis by classifying patients and healthy controls based on similar clinical presentations. In this study, a number of datasets, modalities, and data preprocessing techniques are investigated utilizing collected data. Future directions in PD research are also mentioned, such as subgrouping and link analysis using data from magnetic resonance imaging (MRI), Dopamine Transporter Scan (DaTSCAN), and single-photon emission computed tomography (SPECT). This study compares the effectiveness of diffusion models, autoencoders, and other methods for creating artificial MRI scans. To validate the generative metrics, a categorization study is incorporated as an extra component of the task. Deep learning architectures called convolutional neural networks (CNNs) performed better in disease identification than Gated Recurrent Units (GRUs). Following training with images produced by diffusion models, CNN and GRU were able to diagnose Parkinson’s disease (PD) with 96.5% and 89.3% of the images, respectively. The dataset utilized in investigations is the PPMI dataset. Performance evaluation is based on F1 metrics, recall, accuracy, and precision.