Category: Parkinson's Disease: Neuroimaging
Objective: The goal of this study was to investigate the feasibility of deep learning in the prediction of disease conversion in RBD patients.
Background: Rapid eye movement (REM) sleep behavior disorder (RBD) is considered a prodromal stage of synucleinopathies such as Parkinson’s disease (PD) and multiple system atrophy (MSA). Despite the clinical significance in early phenotype conversion, there is no adequate method available at the moment. The limited availability of longitudinal data with conversion follow-up hampers the development of sophisticated data analysis strategies such as artificial intelligence (AI).
Method: The study was a pilot study conducted in 26 RBD patients with longitudinal clinical and FDG PET follow-up for 2-11 years. Thirteen of these patients converted to PD and one to MSA. In contrast to development from scratch, a convolutional neural network developed on 820 parkinsonian and 714 non-parkinsonian patients for early parkinsonism differential diagnosis was adapted to derive deep metabolic imaging (DMI) indices, which was used to determine predictive scores of longitudinal RBD data. Differences in baseline DMI indices of converted and non-converted RBD patients were assessed by Mann-Whitney independent samples test.
Results: The PD probability of DMI indices increased with disease progression until conversion in 12 of 13 PD converted patients. In the MSA converted patient, the MSA probability of DMI indices also increased. The baseline probabilities of patients who converted in a 6-year period were higher (0.73±0.32) than of non-converted (0.45±0.41), however there were no significant differences (p=0.058).
Conclusion: Despite limited RBD sample size in this pilot study, the preliminary results confirmed the feasibility of the development of AI technologies for early RBD phenotype conversion. The potential of deep learning in extrapolation to the prodromal stage may accelerate the development of early diagnosis methods for PD. Further network improvements must be attempted with a larger RBD database.
To cite this abstract in AMA style:
L. Lopes, P. Wu, J. Lu, J. Wang, C. Bassetti, A. Rominger, C. Zuo, H. Yu, K. Shi. Feasibility of artificial intelligence in predicting early phenotype conversion of REM Sleep Behavior Disorder [abstract]. Mov Disord. 2022; 37 (suppl 2). https://www.mdsabstracts.org/abstract/feasibility-of-artificial-intelligence-in-predicting-early-phenotype-conversion-of-rem-sleep-behavior-disorder/. Accessed September 7, 2024.« Back to 2022 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/feasibility-of-artificial-intelligence-in-predicting-early-phenotype-conversion-of-rem-sleep-behavior-disorder/