Comparison of Quantitative Hippocampal Volumes and Structured Scoring Scales in Predicting Alzheimer Disease Diagnosis

Study Design and Patient Selection

This was a Health Insurance Portability and Accountability Act–compliant retrospective review that was approved by our institutional review board. The need for patient-informed consent was waived. We reviewed imaging studies and medical records of adult patients from 2018 to 2021 at 5 centers that fall under the umbrella of a single health care institution.

The inclusion criteria were adult patients who had undergone MR imaging with the dementia protocol within the search period and who had postscanning International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD-10) diagnoses of AD or MCI, including those having SCD. Exclusion criteria were patients with missing NeuroQuant quantitative parameters, patients with other neurodegenerative ICD-10 diagnoses (such as Parkinson disease, frontotemporal dementia, Pick disease), and those with incomplete demographic data (race and/or highest education). The study participants' flow chart is shown in Fig 1.

• Download figure
• Open in new tab
• Download powerpoint

FIG 1.

Flow chart of study participants. Only diagnoses made after MR imaging were considered for this study. PD indicates Parkinson disease; FTD, frontotemporal dementia.

Clinical Scenarios and Diagnoses

Requests for MR imaging with the dementia protocol were received from neurologists and primary care physicians within the institution's health care network and also from providers outside the network. These included requests from first-line memory clinics, doctor's offices, and tertiary referral centers. All patients were eventually reviewed by board-certified neurologists with additional subspecialty certification in Behavioral Neurology and Neuropsychiatry from the United Council for Neurologic Subspecialties. The final diagnosis was based on a comprehensive neurologic examination, detailed cognitive testing, review of laboratory tests and brain MR imaging scans, and additional CSF biomarker testing as appropriate. The diagnostic guidelines used are consistent with the 2011 National Institute on Aging–Alzheimer's Association revised diagnostic criteria for AD.⁸ In the selection of patients, we considered only those diagnoses made postscanning to be valid for our research.

Imaging Acquisition and Analysis

Brain MR imaging was performed on one of several 1.5T or 3T MR imaging scanners using our institutional dementia MR imaging protocol. All examinations included a volumetric T1-weighted sequence with 1-mm isotropic voxels in addition to standard brain MR imaging sequences. The exact sequence parameters varied among scanners due to scanner differences, but a representative MPRAGE sequence was performed with 1-mm section thickness, 1-mm in-plane resolution, FOV = 256 × 256 mm, TR = 2.3 seconds, TE = 3.17 ms, bandwidth = 210 Hz. Images were acquired in the sagittal plane and reformatted to the axial and coronal planes. Assessment of MTA and ERiCA scores was performed on coronal images by the interpreting board-certified and Certificate of Added Qualification–certified neuroradiologists according to published guides.^6,7 To summarize, the MTA score is a quantification of temporal atrophy using the hippocampal height and width of neighboring CSF space, with scores ranging from 0 (no atrophy) to 4 (severe atrophy).⁶ The ERiCA score is a quantification of entorhinal cortex atrophy assessed at the level of the mammillary bodies, with scores ranging from 0 (normal) to 3 (severe atrophy).⁷ Quantitative assessment was performed using NeuroQuant, which performed volumetric segmentation on the volumetric T1-weighted series and output quantitative volumes and ratios to the PACS. All imaging analyses were performed at the time of clinical interpretation, and no post hoc analysis was performed, to simulate a realistic clinical environment. Figure 2 shows examples of MR images obtained from NeuroQuant for patients with AD, MCI, and SCD.

• Download figure
• Open in new tab
• Download powerpoint

FIG 2.

Coronal T1 noncontrast MR images with automated segmentation overlay obtained from NeuroQuant showing hippocampal volumes (yellow arrow) for patients with AD (A), mild cognitive impairment (B), and SCD (C). Note the marked hippocampal atrophy for AD compared with MCI and SCD.

Outcome and Predictor Variables

The outcome variables used were AD, MCI, and SCD. Normalized total hippocampal volume (NTHV), maximum MTA scores, and maximum ERiCA scores were used as predictor variables. NTHV was calculated by assessing the right and left hippocampal volumes generated separately by NeuroQuant, summing the 2 values, and expressing the resultant value as a percentage of each patient's intracranial volume [(right hippocampal volume + left hippocampal volume)/ICV]. Values generated by NeuroQuant are normalized to an age-specific cohort. MTA and ERiCA scores were assigned to each temporal lobe separately by the interpreting radiologists at the time of clinical interpretation using a structured visual scoring routine. Scores were obtained retrospectively from the radiology report. The higher value of the right and left MTA or ERiCA scores, which we termed maximum MTA or maximum ERiCA, was used for analysis.

Covariates

Covariates were sex, race, highest education attained, age at diagnosis, a history of arterial hypertension, and a history of diabetes mellitus.

Statistical Analysis

The χ² test was used to determine associations between categoric variables and the disease outcomes. The Kruskal-Wallis test was used to determine associations between continuous variables and disease outcomes because observations in these variables assumed a non-normal distribution based on the Shapiro-Wilk test for normality. Differences between comparison groups (AD versus MCI, AD versus SCD, and MCI versus SCD) were calculated in a post hoc analysis. Linear correlational analysis of quantitative and qualitative measures was assessed using the Spearman correlation. Multivariate logistic regression analyses were performed with a stepwise addition of predictors. Diagnostic performances of predictor variables were assessed by receiver operating characteristic (ROC) analyses, and the cutoff value for each parameter was calculated using the maximum Youden index. An α level of .05 was set as the level of significance. SAS software, Version 9.4 (SAS Institute, Cary, North Carolina) was used for all analyses.