Background:
CSF Aß42 is thought to show AD-related alterations earlier than amyloid-ß PET. Therefore, cognitively unimpaired (CU) individuals with abnormal CSF Aß42 and normal amyloid-ß PET are believed to be in the earliest stages of the AD continuum. In this work, we sought to detect structural cerebral alterations in CU individuals with discordant status in these amyloid-ß biomarkers using Machine Learning techniques.
Method:
We included 498 CU individuals from the ALFA+ and ADNI studies with available MRI, amyloid-ß PET and CSF Aß42 measurements, the latter measured with the exploratory Roche NeuroToolKit assays, a panel of automated robust prototype immunoassays. In addition, we calculated Centiloid (CL) values for the PET measurements. Individuals were categorized as CSF-/PET-, CSF+/PET- and CSF+/PET+ according to established cut-offs (CSF Aß42<1098pg/mL for ALFA+ and <880pg/mL for ADNI, and CL<17 for PET). We trained XGBoost classifiers to predict amyloid-ß positivity using as features age, sex, APOE-¿4 status, brain volumes and cortical thicknesses, obtained with Freesurfer 6.0 and the Desikan-Kiliany atlas. Relevant features for pairwise-group classification were sought (CSF-/PET- vs CSF+/PET-; CSF+/PET- vs CSF+/PET+; CSF-/PET- vs CSF+/PET+), calculating SHAP values to determine the most important features for prediction.
Result:
With respect the CSF-/PET- group, the CSF+/PET- showed decreased gray matter volumes in the anterior and posterior cingulate/precuneus and increases in the lateral ventricles and bilateral parahippocampal gyri, among other regions (Figure 1A). Unexpectedly, the posterior cingulate/precuneus showed the opposite effect in cortical thickness measurements. These patterns were similar but more prominent in the comparison between the CSF-/PET- vs CSF+/PET+ group (Figure 1B). Finally, CSF+/PET- group was characterized, with respect the CSF+/PET+ group by higher volume of the bilateral supramarginal gyri and lower cortical thickness in the posterior cingulate/precuneus (Figure 1C). Regarding the other variables in the model, APOE-¿4 status was the most predictive variable in models with respect the CSF-/PET- group and age in the CSF+/PET- vs CSF+/PET+ comparison.
Conclusion:
Our results show that model-free machine learning techniques can detect complex brain morphological alterations in the earliest stages of the AD continuum. Interestingly, some regions showed increases in volume and/or cortical thickness which may reflect compensatory or inflammatory effects.
