Objective:

To develop a tool for monitoring quantitative changes in brain-amyloid burden, with Evoked Potential Tomography (EPT), a portable, noninvasive EEG protocol that is capable of reconstructing 2D PET-like images from event-related potentials (ERPs), such that monitoring change between PET scans is possible. Two approaches were tested using visually evoked and resting-state EEG (eyes open/closed).

Background:

Over 50 million people live with dementia worldwide, most due to Alzheimer’s disease (AD). Amyloid assessment is crucial for diagnosis and monitoring; however, PET is limited by its cost, accessibility, and potential for radiation exposure. EPT offers an EEG-based alternative for generating 2D images approximating ground-truth amyloid distribution from PET.

Design/Methods:

Twenty participants spanning subjective cognitive impairment to clinical AD were enrolled. EEG and PET were acquired on separate visits. The PET reconstruction pipeline included: (1) registration to a common template; (2) extraction and cortical masking of the axial mid-slice; (3) global intensity harmonization and pixel standardization; (4) cortical division into six regions (left/right anterior, central, posterior); (5) regional principal component (PC) analysis, retaining 11–12 PCs; (6) selection of EPT or resting-state features with Spearman’s correlation >0.65 and inter-correlation <0.7; (7) training a bootstrapped ensemble regressor to predict PCs; and (8) inverse transformation and regional recombination to reconstruct 2D PET-like images. Reconstruction was validated on a held-out test set using the structural similarity index measure (SSIM), with non-brain background masked.

Results:

PET image reconstruction from EPT features on the held-out test set produced qualitatively similar results to ground truth, with an average SSIM of 72%. In comparison, repeating the same procedures using resting state EEG yielded significantly weaker results.

Conclusions:

EPT can noninvasively yield 2D PET-like images of amyloid burden, representing a feasible, low-burden adjunct to PET for assessment and monitoring. Validation in larger, independent, and longitudinal cohorts is needed to confirm generalizability, enhance fidelity, and define clinical utility.