Using Artificial Neural Network to Discriminate Parkinson’s Disease from other Parkinsonisms by Focusing on Putamen of Dopamine Transporter SPECT Images: A Retrospective Study

Abstract

It is still difficult to distinguish Parkinson's disease from parkinsonism brought on by other disorders at an early stage. To process images from dopamine transporter single-photon emission computed tomography, we used an artificial neural network (ANN) (DAT-SPECT). Training and test sets of abnormal DAT-SPECT images of individuals with Parkinson's disease and parkinsonism brought on by other disorders were created. To distinguish Parkinson's disease from parkinsonism brought on by other disorders, the striatal regions of the images were segmented using an active contour model and used as the data for transfer learning on a pre-trained ANN. For comparison, a support vector machine was trained using semi-quantitative measurement parameters such as the specific binding ratio and asymmetry index. The ANN classifier's predictive accuracy (86%) was higher than the support vector machine classifier's (68%). The ANN classifier had a sensitivity and specificity of 81.8% and 88.6%, respectively, in predicting Parkinson's disease. The ANN classifier performed better than traditional biomarkers at distinguishing between Parkinson's disease and parkinsonism brought on by other disorders. At an early stage of the parkinsonian disorders, ANN can be a promising biomarker, particularly when it concentrates on the putamen.