Predicting Brain Amyloid Using Multivariate Morphometry Statistics, Sparse Coding, and Correntropy: Validation in...

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Alzheimer's Disease Research

Background:

• Early detection of Alzheimer's disease (AD) is crucial for therapeutic intervention.
• Beta-amyloid (Aβ) plaque accumulation is an early hallmark of AD.
• Current Aβ detection methods are invasive or costly.

Purpose of the Study:

• To develop a non-invasive method for inferring brain Aβ burden at the individual level using MRI.
• To utilize multivariate morphometry statistics (MMS) of the hippocampus for Aβ detection.
• To introduce a novel sparse coding algorithm, PASCS-MP, for low-dimensional representation of hippocampal data.

Main Methods:

• Applied MRI-based hippocampal multivariate morphometry statistics (MMS).
• Developed and employed a sparse coding algorithm, Patch Analysis-based Surface Correntropy-induced Sparse-coding and Max-Pooling (PASCS-MP), for data representation.
• Utilized a binary random forest classifier to predict Aβ positivity in two independent cohorts (ADNI and OASIS).

Main Results:

• The PASCS-MP method achieved high accuracy in discriminating Aβ positivity.
• Accuracy was 0.89 in mild cognitive impairment (MCI) individuals (ADNI).
• Accuracy was 0.79 (ADNI) and 0.81 (OASIS) in cognitively unimpaired (CU) individuals.

Conclusions:

• The proposed MRI-MMS and PASCS-MP method offers a promising non-invasive approach for detecting brain Aβ burden.
• This method demonstrates superior performance compared to traditional algorithms for Aβ detection.
• The findings support the potential of MRI-MMS as a cost-effective biomarker for early AD detection.