Objective: To use a deep learning based framework to classify MRI brain scans as PD or control cases.

Background: The Parkinson’s Progression Markers Initiative provides a large MRI dataset from a population of PD patients and control individuals. MRI is not currently considered a robust imaging test for PD, although MRI changes have previously been detected in those with PD compared to controls. The subset of artificial intelligence and machine learning known as deep learning has recently shown great promise in diagnostic medical imaging, outperforming practitioners by detecting patterns that are not visible to the human eye. A number of exploratory imaging techniques have suggested that dedicated experimental MRI sequences may be able to detect early brain pathological changes in PD. We aimed to use a machine learning approach to explore whether such changes might be detectable on routine MRI imaging data.

Method: A convolutional neural network architecture has been developed to classify MRI brain images. The model was assessed and the hyper-parameters tuned using k-fold cross-validation.

Results: When tested on a holdout set, the developed model achieved an accuracy of 93.6%, with precision of 94.1% and recall of 98.2%. The Receiver Operator Characteristic Area-Under-the-Curve was 0.948.

Conclusion: The developed model provides a high diagnostic performance, which may potentially be further enhanced by the use of additional clinical data.

« Back to MDS Virtual Congress 2021

MDS Abstracts - https://www.mdsabstracts.org/abstract/deep-learning-based-detection-of-parkinsons-disease-pd-in-magnetic-resonance-imaging-mri-brain-scans/