Poster No:
245
Submission Type:
Abstract Submission
Authors:
Paolo Nucifora1
Institutions:
1Loyola University Chicago, Chicago, IL
First Author:
Introduction:
In Alzheimer's disease, amyloid deposition generally precedes the onset of objective symptoms. The duration of symptom-free amyloid deposition is unclear, but it appears to accumulate faster in APOE ε4 carriers (Gonneaud et al., 2016) and is detectable in cognitively normal elderly individuals (Li et al., 2023).
In this study, the presence of amyloid-β in cognitively normal individuals under 70 years old was evaluated using PET after injection of 18F-florbetapir, a radiopharmaceutical that binds to amyloid-β. Images from APOE ε4 carriers and non-carriers were compared with an AI-assisted method of whole-brain evaluation.
Methods:
Data used in preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative database (adni.loni.usc.edu).
Baseline 18F-florbetapir images from APOE ε4 carriers and non-carriers between 60 and 70 years old were obtained for this study. These images were used to train a 3D convolutional neural network. Classification accuracy was measured with 10-fold cross-validation.
By itself, the classification accuracy value is insufficient to establish a significant difference between the two groups. But as described by Golland and Fischl, the significance of an accuracy value can be determined by permutation testing, i.e. re-measuring accuracy with multiple sets of permuted data that are consistent with the null hypothesis (Golland and Fischl, 2003). If the accuracy obtained from the non-permuted dataset is significant, it is evidence against the null hypothesis and implies that the two populations are not identical.
In this study, "carrier" and "non-carrier" labels were permuted 100 times, and each permutation was used to train a new 3D convolutional network. Classification accuracy of the non-permuted dataset was compared to the accuracy of all permuted datasets in order to assess the hypothesis that amyloid-beta binding is not identical in APOE ε4 carrier and non-carrier populations.
Results:
The "carrier" group consisted of 33 unique individuals aged 67.0 ± 2.0 years. The "non-carrier" group consisted of 56 unique individuals aged 66.9 ± 2.2 years.
The non-permuted dataset was associated with a classification accuracy of 67.4%. This was higher than the classification accuracy of all 100 permuted datasets. Therefore, "carrier" amyloid-beta binding differed from "non-carrier" amyloid-β binding at a significance level of p=0.01.
Conclusions:
This study provides evidence of altered amyloid-β binding in cognitively normal APOE ε4 carriers under the age of 70, which is consistent with a recent report of altered amyloid-β binding in cognitively normal APOE ε4 carriers over 70 (Li et al., 2023). These findings support the hypothesis that amyloid-β binding is altered in APOE ε4 carriers long before they develop objective symptoms of Alzheimer's disease, and therefore suggest that early treatment of APOE ε4 carriers might be beneficial.
In addition, this study illustrates the use of AI in combination with permutation analysis for statistical hypothesis testing. Of note, useful results can be obtained even in datasets that are relatively small and unsuitable for training a high-accuracy classifier. This approach may be helpful alongside exploratory methods such as region-of-interest (ROI) analysis or voxel-based morphometry (VBM). Like VBM it may be sensitive to effects that occur at the scale of individual voxels, and may even be sensitive to complex changes that elude VBM (e.g. those that do not consistently localize to specific voxels). However, like ROI analysis it can only demonstrate whether two datasets differ, not which voxels are responsible for the difference.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Other Methods
Keywords:
ADULTS
Aging
Data analysis
Degenerative Disease
Machine Learning
NORMAL HUMAN
Open Data
Positron Emission Tomography (PET)
1|2Indicates the priority used for review
Provide references using author date format
Golland, P. (2003). "Permutation Tests for Classification: Towards Statistical Significance in Image-Based Studies." In C. Taylor & J. A. Noble (Eds.), Information Processing in Medical Imaging (pp. 330–341). Springer. https://doi.org/10.1007/978-3-540-45087-0_28
Gonneaud, J. (2016). "Relative effect of APOE ε4 on neuroimaging biomarker changes across the lifespan." Neurology, 87(16), 1696–1703. https://doi.org/10.1212/WNL.0000000000003234
Li, W. (2023). "Effect of APOE ε4 genotype on amyloid-β, glucose metabolism, and gray matter volume in cognitively normal individuals and amnestic mild cognitive impairment." European Journal of Neurology, 30(3), 587–596. https://doi.org/10.1111/ene.15656