Detection of parkinson’s disease from t2-weighted magnetic resonance imaging scans using efficientnet-v2

Parkinson’s disease (PD) is a multifaceted neurode-generative disorder that primarily disrupts voluntary motor movements by causing an excitation-inhibition imbalance in the brain. Approximately 10 million people worldwide are affected by PD. However, accurate diagnosis of PD is still challenging in the early stages of this disease due to the similarity of the phenotypes of the neurological disorders. Magnetic resonance imaging (MRI) has played an important role in understanding brain function and disease in neuroimaging. In particular, it can detect structural abnormalities in the brain caused by dopamine deprivation in PD patients leading to excitation-inhibition imbalance. Besides, the utilization of deep learning has emerged as a crucial factor in the identification of Parkinson’s disease due to its ability to identify irregularities, and structural changes at a specific location of brain. This study investigates the effectiveness of the deep learning model EfficientNet-V2 in combination with transfer learning for the purpose of identifying the presence or absence of Parkinson’s disease in individuals. In contrast to the arbitrary scaling employed by conventional CNN, the EfficientNet-V2 employs a straightforward and efficient compound factor to modify the network dimensions, which facilitates the identification of the optimal set of parameters. The identification is made by analyzing MRI samples obtained from the Parkinson’s Progression Markers Initiative, an openly accessible dataset. To reduce bias while analyzing the detection performance and to stabilize the overall performance of the architecture, this study has employed 4-fold cross-validation method during data split. This method has obtained an overall 99.13% accuracy, which is substantially higher than the accuracy of earlier works.