Disease progression modeling‐based prediction of cognitive decline: Neuroimaging / Optimal neuroimaging measures for tracking disease progression. (7th December 2020)

• Record Type: Journal Article
• Authors: Ghazi, Mostafa Mehdipour; Sørensen, Lauge; Pai, Akshay; Cardoso, Jorge; Modat, Marc; Ourselin, Sebastien; Nielsen, Mads
• Is Part Of: Alzheimer's & dementia. Volume 16(2020)Supplement 4
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-12-07
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.043850
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: The objective of this study is to investigate decline prediction of cognitive test scores in stable and converting mild cognitive impairment (MCI) subjects using both nonparametric and parametric Alzheimer's disease (AD) progression modeling methods trained on data including cognitive tests, CSF measures, and neuroimaging biomarkers. Method: The study dataset consisted of yearly visits (2005‐2017) for 782 Alzheimer's Disease Neuroimaging Initiative (ADNI) subjects with normal cognition, MCI, or AD, including FreeSurfer‐based T1‐weighted brain MRI volumes (ventricles, hippocampus, whole brain, fusiform, and entorhinal cortex, all normalized with intracranial volume), cognitive tests (MMSE, CDR‐SB, ADAS‐Cog‐13, FAQ, and RAVLT‐immediate‐recall), CSF measures (Amyloid‐beta and p‐tau), and FDG‐PET. Two state‐of‐the‐art disease progression modeling methods, a nonparametric using LSTMs (Ghazi, M. M., et al., Medical Image Analysis 53, 39‐46, 2019) and a parametric using regression (Jedynak, B. M., et al., Neurobiology of Aging 36, 178‐184, 2015), were trained on 632 subjects and subsequently applied to predict month 24 to 96 MMSE scores for 150 independent test subjects using only their baseline and month 12 data. Result: The predictive power and prognostic capability of the AD progression modeling methods were assessed using the per‐visit mean absolute error (MAE) and area under the ROC curve (AUC) of predicted MMSE scores for stable MCI (sMCI) and converting MCI (cMCI) test subjects. The average MAE results for month 24 to 96 MMSE scores were as follows: nonparametric 1.39 to 1.04 (sMCI), 2.41 to 3.62 (cMCI); parametric 1.46 to 4.00 (sMCI), 2.42 to 2.53 (cMCI). The average AUC results for month 24 to 96 obtained based on the predicted MMSE scores were as follows (two‐sample t ‐test, p < 0.001 in all cases): nonparametric 0.79 to 0.73; parametric 0.83 to 0.69. Conclusion: In almost all cases, the nonparametric method outperforms the parametric model in predicting MMSE scores. Moreover, predictions from both nonparametric and parametric methods can significantly discriminate between sMCI and cMCI groups. Though, the discrimination capability of the nonparametric method is superior in long‐term prediction of cognitive decline.