Machine learning models for the differential diagnosis of vascular parkinsonism and Parkinson’s disease using [123I]FP-CIT SPECT
- 634 Downloads
The study’s objective was to develop diagnostic predictive models using data from two commonly used [123I]FP-CIT SPECT assessment methods: region-of-interest (ROI) analysis and whole-brain voxel-based analysis.
We included retrospectively 80 patients with vascular parkinsonism (VP) and 164 patients with Parkinson’s disease (PD) who underwent [123I]FP-CIT SPECT. Nuclear-medicine specialists evaluated the scans and calculated bilateral caudate and putamen [123I]FP-CIT uptake and asymmetry indices using BRASS software. Statistical parametric mapping (SPM) was used to compare the radioligand uptake between the two diseases at the voxel level. Quantitative data from these two methods, together with potential confounding factors for dopamine transporter availability (sex, age, disease duration and severity), were used to build predictive models following a tenfold cross-validation scheme. The performance of logistic regression (LR), linear discriminant analysis and support vector machine (SVM) algorithms for ROI data, and their penalized versions for SPM data (penalized LR, penalized discriminant analysis and SVM), were assessed.
Significant differences were found in the ROI analysis after covariate correction between VP and PD patients in [123I]FP-CIT uptake in the more affected side of the putamen and the ipsilateral caudate. Age, disease duration and severity were also found to be informative in feeding the statistical model. SPM localized significant reductions in [123I]FP-CIT uptake in PD with respect to VP in two specular clusters comprising areas corresponding to the left and right striatum. The diagnostic predictive accuracy of the LR model using ROI data was 90.3 % and of the SVM model using SPM data was 90.4 %.
The predictive models built with ROI data and SPM data from [123I]FP-CIT SPECT provide great discrimination accuracy between VP and PD. External validation of these methods is necessary to confirm their applicability across centres.
KeywordsVascular parkinsonism Parkinson’s disease [123I]FP-CIT SPECT Statistical parametric mapping Predictive models
This work was supported by grants from the Ministerio de Economía y Competitividad de España (SAF2007-60700), the Instituto de Salud Carlos III (PI10/01674, CP08/00174, PI13/01461), the Consejería de Economía, Innovación, Ciencia y Empresa de la Junta de Andalucía (CVI-02526, CTS-7685), the Consejería de Salud y Bienestar Social de la Junta de Andalucía (PI-0377/2007, PI-0741/2010, PI-0437-2012), the Sociedad Andaluza de Neurología, the Fundación Alicia Koplowitz, the Fundación Mutua Madrileña and the Jaques and Gloria Gossweiler Foundation.
Conflicts of interest
- 13.Navarro-Otano J, Gaig C, Muxi A, Lomeña F, Compta Y, Buongiorno MT, et al. (123)I-MIBG cardiac uptake, smell identification and (123)I-FP-CIT SPECT in the differential diagnosis between vascular parkinsonism and Parkinson’s disease. Parkinsonism Relat Disord. 2014;20:192–7.PubMedCrossRefGoogle Scholar
- 19.Nocker M, Seppi K, Donnemiller E, Virgolini I, Wenning GK, Poewe W, et al. Progression of dopamine transporter decline in patients with the Parkinson variant of multiple system atrophy: a voxel-based analysis of [123I]β-CIT SPECT. Eur J Nucl Med Mol Imaging. 2012;39:1012–20.PubMedCrossRefGoogle Scholar
- 23.Varrone A, Dickson JC, Tossici-Bolt L, Sera T, Asenbaum S, Booij J, et al. European multicentre database of healthy controls for [123I]FP-CIT SPECT (ENC-DAT): age-related effects, gender differences and evaluation of different methods of analysis. Eur J Nucl Med Mol Imaging. 2013;40:213–27.PubMedCrossRefGoogle Scholar
- 25.Friedman J, Hastie T, Tibshirani R. The Elements of Statistical Learning - Data Mining, Inference, and Prediction, 2nd Edition. Springer Series in Statistics. Berlin: Springer; 2009. p. 1–745.Google Scholar