Objective: To test an automated differential diagnostic algorithm for parkinsonian syndromes based on machine learning and FDG-PET scans using subject scores for sets of validated disease patterns from two independent sites.

Background: FDG PET has been used to identify disease-specific metabolic patterns for Parkinson’s disease (PD) and for the major parkinsonian syndromes: multiple system atrophy (MSA) and progressive supranuclear palsy (PSP). While early differential diagnosis can be improved using the corresponding disease patterns [1], we explored the possibility that accuracy can be further enhanced with the addition of machine learning.

Method: We analysed 265 FDG PET scans from two sites (Slovenia and United States) in patients with parkinsonism for whom the final clinical diagnosis (161 PD, 57 MSA, and 47 PSP) was unknown at the time of imaging. We used a machine learning model based on support vector machine and two sets of features: (1) expression values for three previously validated disease patterns identified in Slovenian cohorts [2, 3] (2) analogous expression values for the original US cohorts [4, 5].

Results: Both SLO-patterns and USA-patterns models achieved high overall diagnostic accuracy (86, and 85% respectively). The two pattern-based models had respective specificity and sensitivity of 82–83% and 94% for PD; 96% and 70–74% for PSP; 94–95% and 72% in MSA.

Conclusion: Both pattern-based classifiers achieved comparably high specificities and sensitivities in PD, MSA and PSP. The two sets of pattern-based classifiers performed similarly, indicating that metabolic patterns identified in one institution can be successfully used at different sites.

References: [1] Perovnik M, Rus T, Schindlbeck KA, Eidelberg D. Functional brain networks in the evaluation of patients with neurodegenerative disorders. Nat Rev Neurol 2023;19:73–90. https://doi.org/10.1038/s41582-022-00753-3.
[2] Tomše P, Jensterle L, Grmek M, Zaletel K, Pirtošek Z, Dhawan V, et al. Abnormal metabolic brain network associated with Parkinson’s disease: replication on a new European sample. Neuroradiology 2017;59:507–15. https://doi.org/10.1007/s00234-017-1821-3.
[3] Tomše P, Rebec E, Studen A, Perovnik M, Rus T, Ležaić L, et al. Abnormal metabolic covariance patterns associated with multiple system atrophy and progressive supranuclear palsy. Phys Medica 2022;98:131–8. https://doi.org/10.1016/j.ejmp.2022.04.016.
[4] Tang CC, Poston KL, Eckert T, Feigin A, Frucht S, Gudesblatt M, et al. Differential diagnosis of parkinsonism: a metabolic imaging study using pattern analysis. Lancet Neurol 2010;9:149–58. https://doi.org/10.1016/S1474-4422(10)70002-8.
[5] Rus T, Tomše P, Jensterle L, Grmek M, Pirtošek Z, Eidelberg D, et al. Differential diagnosis of parkinsonian syndromes: a comparison of clinical and automated – metabolic brain patterns’ based approach. Eur J Nucl Med Mol Imaging 2020;47:2901–10. https://doi.org/10.1007/s00259-020-04785-z.

« Back to 2023 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/machine-learning-diagnosis-of-parkinsonian-syndromes-network-approach-with-two-different-sites/