Objective: To assess the accuracy of an automated, disease-specific, pattern model to distinguish among PD, MSA and PSP, in comparison with the final clinical diagnosis.

Background: Parkinsonian disorders may be challenging to differentiate, especially early in the disease course. Spatial covariance analysis has previously been applied onFDG-PET to identify and measure disease-specific metabolic patterns associated with PD (1) and atypical parkinsonian syndromes including MSA and PSP (2). Based on these pattern scores, a two-level, automated algorithm has been developed and evaluated for the classification of individuals with PD, MSA and PSP, showing very high specificity (3). No studies have yet attempted to apply this algorithm on further patient-cohorts.

Method: In this pilot study, 24 patients with parkinsonism, enrolled in a hospital-based, prospective cohort, that had performed FDG-PET scans in diagnostic purpose were included. Pattern expression values (i.e. subject scores) of different disease-related metabolic networks were computed for all 24 patients. Finally, an automated differential diagnosis of PD, MSA and PSP was performed for each patient, using their pattern scores and the previously published and validated algorithms (3, 4). The automated diagnoses were compared to the clinical diagnoses according to previously validated cut-off probabilities.

Results: 16/19 PD patients, 3/3 MSA and 2/2 PSP patients were accurately classified. Thus, the automated algorithm provided 84% sensitivity and 89% specificity for PD, and 100% sensitivity and specificity for MSA and PSP. The positive predicted values were 94% for PD, and 100% for MSA and PSP, whereas the negative predictive values were 74% and 100% respectively.

Conclusion: Our study provides further evidence that disease-specific, voxel-based, FDG-PET metabolic patterns can be used in automated classification algorithms to improve differential diagnosis of parkinsonian syndromes. Accurate diagnosis, early in the disease course may have significant impact on treatment-plan decisions, and also on patient-selection for clinical trials. Further investigation of a larger sample from the Stockholm-cohort is ongoing. It will also be important to increase the number of cases where the clinical diagnosis is pathologically confirmed.

References: 1. Ma Y, Tang C, Spetsieris PG, Dhawan V, Eidelberg D. Abnormal metabolic network activity in parkinson’s disease: Test-retest reproducibility. J Cereb Blood Flow Metab. 2007;27:597-605 2. Eckert T, Tang C, Ma Y, Brown N, Lin T, Frucht S, et al. Abnormal metabolic networks in atypical parkinsonism. Mov Disord. 2008;23:727-733 3. 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 neurology. 2010;9:149-158 4. Tripathi M, Tang CC, Feigin A, De Lucia I, Nazem A, Dhawan V, et al. Automated differential diagnosis of early parkinsonism using metabolic brain networks: A validation study. J Nucl Med. 2016;57:60-66

« Back to 2019 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/automated-metabolic-pattern-analysis-in-the-differential-diagnosis-of-parkinsonism-in-a-swedish-cohort/