Classifying Alzheimer's Disease Neuropathology Using Clinical and MRI Measurements

Abstract

Background: Computer-aided machine learning models are being actively developed with clinically available biomarkers to diagnose Alzheimer's disease (AD) in living persons. Despite considerable work with cross-sectional in vivo data, many models lack validation against postmortem AD neuropathological data.

Objective: Train machine learning models to classify the presence or absence of autopsy-confirmed severe AD neuropathology using clinically available features.

Methods: AD neuropathological status are assessed at postmortem for participants from the National Alzheimer's Coordinating Center (NACC). Clinically available features are utilized, including demographics, Apolipoprotein E(APOE) genotype, and cortical thicknesses derived from ante-mortem MRI scans encompassing AD meta regions of interest (meta-ROI). Both logistic regression and random forest models are trained to identify linearly and nonlinearly separable features between participants with the presence (N = 91, age-at-MRI = 73.6±9.24, 38 women) or absence (N = 53, age-at-MRI = 68.93±19.69, 24 women) of severe AD neuropathology. The trained models are further validated in an external data set against in vivo amyloid biomarkers derived from PET imaging (amyloid-positive: N = 71, age-at-MRI = 74.17±6.37, 26 women; amyloid-negative: N = 73, age-at-MRI = 71.59±6.80, 41 women).

Results: Our models achieve a cross-validation accuracy of 84.03% in classifying the presence or absence of severe AD neuropathology, and an external-validation accuracy of 70.14% in classifying in vivo amyloid positivity status.

Conclusions: Our models show that clinically accessible features, including APOE genotype and cortical thinning encompassing AD meta-ROIs, are able to classify both postmortem confirmed AD neuropathological status and in vivo amyloid status with reasonable accuracies. These results suggest the potential utility of AD meta-ROIs in determining AD neuropathological status in living persons.

Keywords: APOE genotype; Alzheimer’s disease-meta-ROIs; in vivo amyloid status; machine learning; severe AD neuropathology.

Fig. 1

Inclusion/exclusion flow chart. NACC Neuropathology Data Set variables NPADNC, NPLBOD, NPFTDTDP, and NPFTDTAU were utilized to evaluate AD, Lewy body, Frontotemporal lobar degeneration with TPD-43-immunoreactive pathology (FTLD-TDP), and FTLD-tau pathologies, respectively.

Fig. 2

Classification schema. Black box (on the left): Clinically accessible features included in this study; Blue boxes (top part on the right): training and validating machine learning models in classifying the presence or absence of severe AD neuropathological status in NACC participants; Orange boxes (bottom part): External validation of trained models in classifying in vivo amyloid positivity status determined by PET imaging using CNTN subjects.

Fig. 3

LASSO-logistic regression results. A) Feature selection results. Cross-validated (CV) deviance of LASSO-logistic-regression models, trained with NACC participants to classify the presence or absence of severe AD neuropathological status (ADNC3 versus ADNC0&1), as a function of regularization strength in LASSO (lambda). The green circle corresponds to the selected model with a minimum CV deviance. The intersect table lists the beta coefficient in the logistic regression model of each selected feature. B) Model performance with selected features. ROC curve for CV performance of the reduced logistic regression model trained with 6 selected features to classify ADNC3 versus ADNC0&1. The red filled dot indicates the point with the lowest total false rate (false positive rate (FPR) + false negative rate (FNR)). The corresponding threshold s = 0.55 is used to binarize the predicted probability in assigning participants to the ADNC3 group. Using this model with this threshold, the intersect table shows the CV-performance with NACC participants to classify AD neuropathological status (ADNC3 versus ADNC0&1) and external testing results with CNTN participants to classify amyloid positivity status (amyloid positive versus amyloid negative).

Fig. 4

Random forest results. A) Feature selection results. Out-of-box (OOB) permutation based (blue bars) and Gini impurity index (orange curve) based feature importance scores in the random forest model trained using all features from NACC participants to classify the presence or absence of severe AD neuropathological status (ADNC3 versus ADNC0&1). Stars (*) indicate features retained in the final model. B) Model performance with selected features. ROC curve of the random forest model with the 6 selected features (detailed in Table 2). The red filled dot indicates the point with the lowest total false rate (false positive rate + false negative rate). The corresponding threshold s = 0.5863 is used to binarize the predicted probability in assigning participants to the ADNC3 group. Using this model with this threshold, the intersect table shows the OOB-validation-performance with NACC participants to classify AD neuropathological status (ADNC3 versus ADNC0&1) and external testing results with CNTN participants to classify amyloid positivity status (amyloid positive versus amyloid negative).