Improved motor outcome prediction in Parkinson's disease applying deep learning to DaTscan SPECT images

Dopamine transporter (DAT) SPECT imaging is routinely used in the diagnosis of Parkinson's disease (PD). Our previous efforts demonstrated the use of DAT SPECT images in a data-driven manner by improving prediction of PD clinical assessment outcome using radiomic features. In this work, we develop a convolutional neural network (CNN) based technique to predict clinical motor function evaluation scores directly from longitudinal DAT SPECT images and non-imaging clinical measures.

Data of 252 subjects from the Parkinson's Progression Markers Initiative (PPMI) database were used in this work. The motor part (III) score of the unified Parkinson's disease rating scale (UPDRS) at year 4 was selected as outcome, and the DAT SPECT images and UPDRS_III scores acquired at year 0 and year 1 were used as input data. The specified inputs and outputs were used to develop a CNN based regression method for prediction. Ten-fold cross-validation was used to test the trained network and the absolute difference between predicted and actual scores was used as the performance metric. Prediction using inputs with and without DAT images was evaluated.

Using only UPDRS_III scores at year 0 and year 1, the prediction yielded an average difference of 7.6 ± 6.1 between the predicted and actual year 4 motor scores (range [5, 77]). The average difference was reduced to 6.0 ± 4.8 when longitudinal DAT SPECT images were included, which was determined to be statistically significant via a two-sample t-test, and demonstrates the benefit of including images.

This study shows that adding DAT SPECT images to UPDRS_III scores as inputs to deep-learning based prediction significantly improves the outcome. Without requiring segmentation and feature extraction, the CNN based prediction method allows easier and more universial application.