Using AI-derived hippocampal metrics offers limited diagnostic value in patients with cognitive impairment

cortechs-ai-neuroquant-brain (8).png

Product: NeuroQuant Company: Cortechs.ai


The Clinical Value of NeuroQuant Hippocampal Volumes and Hippocampal Occupancy Scores in an Academic Medical Center Memory Clinic

Cognitive and Behavioral Neurology, 2025

Abstract

Background

NeuroQuant volumetric data proposes to assist clinical diagnoses of mild cognitive impairment (MCI) and Alzheimer disease (AD) through normative hippocampal volume measurements (HVs) and hippocampal occupancy scores (HOCs). We investigated the diagnostic utility of NeuroQuant HVs and HOCs within an academic memory disorders clinic.

Methods

Participants (n = 181) underwent neurologic examination, MRI with NeuroQuant, and neuropsychological assessment. Clinical diagnoses included subjective cognitive concerns (n = 39), non-amnestic MCI (n = 33), amnestic MCI (n = 74), AD (n = 35). Objective memory impairment (n = 125) was determined using actuarial criteria (≥2 memory measures ≤1.5 SD below the mean) for each participant. Differences in HVs and HOCs were assessed using ANOVA and χ2, while Pearson correlations examined their relationship with the California Verbal Learning Test-2nd Edition (CVLT-II). Receiver operator characteristic curves were used to calculate sensitivity, specificity, predictive values, Youden's Index, and predictive summary index scores.

Results

Participants with amnestic MCI, AD, and objective memory impairment had lower HVs and HOCs and a higher frequency of hippocampal atrophy (≤5th percentile) than participants with subjective cognitive concerns, non-amnestic MCI, and no objective memory impairment (P's <0.05). However, the sensitivity (0.31-0.77) and specificity (0.68-0.91) of HVs and HOCs were variable for determining objective memory impairment, and correlations with the CVLT-II were modest (r = 0.20-0.44).

Discussion

NeuroQuant HVs and HOCs were lower in participants with amnestic MCI, AD, and memory impairment, but had poor sensitivity and specificity to predict memory impairment, and only a modest association with measurable memory impairment. These measures should be used cautiously as supporting diagnostic tools.

Read full study