A CNN Approach to Detect Parkinson's Disease using T1-Weighted, T2-Weighted, and Flair MRI

Parkinson's disease (PD) has always been under the category of incurable disease to date. This neurodegenerative disease is generally caused by the deficiency of dopamine-developing nerve cells in the brain. The primary symptom of PD is a tremor, or shaking, in one or both hands. Other symptoms may include harshness of the limbs and trunk, slow movement (bradykinesia), poor balance and coordination, and a change in speech or writing (dysarthria or micrographia). As the senior citizen population is mounting, the patients of Parkinson's are also increasing. Unfortunately, there is absolutely no effective treatment for Parkinson's disease. Therefore, the task at hand is to enhance the timely identification of the illness. This research study has developed a novel 2D convolutional neural network to learn the intricate patterns in three different types of MR Images (MRI) to detect Parkinson's disease, which is thought to be brought on by minor, undetected strokes in the brain or the severity of white matter lesions, using a sizable dataset available on Kaggle. Through this model, a high accuracy of 94% has been achieved. Next, transfer learning is applied by using InceptionV3. The proposed model demonstrates an overall accuracy of 95.29% by leveraging a pre-trained model. Moreover, the precision, recall, and F1-score stand at 95%, 100%, and 97% correspondingly and are higher for the large dataset.