An Ensemble-Dense X-Net model for detection of complex regions in Parkinson’s disease using high-resolution MRI scans

Abstract Parkinson’s disease (PD) is a chronic neurological disorder that mainly affects daily life. The aim of this research was primarily to detect PD in its early stages based on abnormal behavior such as cognitive impairment, rapid changes in emotions, and self-control disorders. In this research, a fine-tuned pre-trained DenseNet-based deep learning (DL) model that reliably retrained on PD MRI images. The preprocessing techniques, such as affine transformations and Patch Extraction, are used to enhance the input images. Advanced tissue segmentation is another process that segments the brain output images into specific regions. Finally, the Ensemble-Dense X-Net (EDX-Net) model is used to detect PD based on significant brain regions like substantia-nigra and classifies the samples. The proposed model is developed as a Cross-modal system and was effectively evaluated on two neuroimaging datasets: the Parkinson’s disease functional magnetic resonance imaging (fMRI) Images dataset (D1) and the Parkinson’s disease Dementia (PDD) MRI dataset (D2), both collected from Kaggle. This research also focused on identifying affected regions using both fMRI and MRI images. These two datasets are two different imaging modalities such as fMRI and MRI. Experimental results show that the proposed approach achieves performance of Sn of 97.78, Sp of 98.34, P of 97.89, Acc of 98.99, F1S of 96.23. For D1, and Sn-98.31, Sp-96.99, P-97.78, Acc-98.45, and F1S-97.88 for D2 with significantly less processing time. Thus, we can say that the proposed approach works effectively on fMRI and MRI images.