Prospective Validation of ¹⁸F-FDG Brain PET Discriminant Analysis Methods in the Diagnosis of Amyotrophic Lateral Sclerosis

Abstract

An objective biomarker for early identification and accurate differential diagnosis of amyotrophic lateral sclerosis (ALS) is lacking. 18 F-FDG PET brain imaging with advanced statistical analysis may provide a tool to facilitate this. The objective of this work was to validate volume-ofinterest (VOI) and voxel-based (using a support vector machine [SVM] approach) 18 F-FDG PET analysis methods to differentiate ALS from controls in an independent prospective large cohort, using a prioriderived classifiers. Furthermore, the prognostic value of 18 F-FDG PET was evaluated. Methods: A prospective cohort of patients with a suspected diagnosis of a motor neuron disorder (n 5 119; mean age ± SD, 61 ± 12 y; 81 men and 38 women) was recruited. One hundred five patients were diagnosed with ALS (mean age ± SD, 61.0 ± 12 y; 74 men and 31 women) (group 2), 10 patients with primary lateral sclerosis (mean age ± SD, 55.5 ± 12 y; 3 men and 7 women), and 4 patients with progressive muscular atrophy (mean age ± SD, 59.2 ± 5 y; 4 men). The mean disease duration of all patients was 15.0 ± 13.4 mo at diagnosis, with PET conducted 15.2 ± 13.3 mo after the first symptoms. Data were compared with a previously gathered dataset of 20 screened healthy subjects (mean age ± SD, 62.4 ± 6.4 y; 12 men and 8 women) and 70 ALS patients (mean age ± SD, 62.2 ± 12.5 y; 44 men and 26 women) (group 1). Data were spatially normalized and analyzed on a VOI basis (statistical software (using the Hammers atlas) and voxel basis using statistical parametric mapping. Discriminant analysis and SVM were used to classify new cases based on the classifiers derived from group 1. Results: Compared with controls, ALS patients showed a nearly identical pattern of hypo-and hypermetabolism in groups 1 and 2. VOI-based discriminant analysis resulted in an 88.8% accuracy in predicting the new ALS cases. For the SVM approach, this accuracy was 100%. Brain metabolism between ALS and primary lateral sclerosis patients was nearly identical and not separable on an individual basis. Extensive frontotemporal hypometabolism was predictive for a lower survival using a Kaplan-Meier survival analysis (P , 0.001). Conclusion: On the basis of a previously acquired training set, 18 F-FDG PET with advanced discriminant analysis methods is able to accurately distinguish ALS from controls and aids in assessing individual prognosis. Further validation on multicenter datasets and ALS-mimicking disorders is needed to fully assess the general applicability of this approach.

Finger extension weakness and downbeat nystagmus motor neuron disease syndrome: A novel motor neuron disorder?

“…Fluorodeoxyglucose-positron emission tomography was performed in 1 patient, revealing mild hypometabolism in the motor cortex, as can be seen in ALS. [13][14][15] A trial with plasma exchange (patient 1) and intravenous immunoglobulins (patient 2) did not alter the disease course.…”

Positron emission tomography in amyotrophic lateral sclerosis: Towards targeting of molecular pathological hallmarks

“…On the other hand, hypermetabolism was observed in the mesiotemporal cortex, cerebellum and upper brain stem [65]. Overall, diagnostic value analysis of [ 18 F]-FDG PET versus a control population showed 90-95% sensitivity [66][67][68]. In the future, implementation of [ 18 F]-FDG PET in the standard diagnostic work-up will result in large datasets, allowing multivariate analysis and identification of spatially distinct brain networks affected by neurodegeneration.…”

Cerebral 18F-FDG PET in macrophagic myofasciitis: An individual SVM-based approach

“…Applied to 18 F-FDG PET brain imaging, SVM classifier recently showed high accuracy to distinguish patients with amyotrophic lateral sclerosis from controls and assess individual prognosis [13]. Indeed, SVM method is a supervised classification machine-learning algorithm commonly used in neuroimaging for multi-voxel pattern analysis either for PET [13] or MRI [14].…”

“…Indeed, SVM method is a supervised classification machine-learning algorithm commonly used in neuroimaging for multi-voxel pattern analysis either for PET [13] or MRI [14]. SVM classifiers are also more and more used in various bioinformatics fields and at different scales such as dicer cleavage sites prediction [15] for miRNA, protein homology detection [16], and in biomedical imaging for example for decoding [17].…”