Accuracy of Visual Atrophy Scales on MRI to predict Mild Dementia in the Cog‐Aging Cohort Study

TL;DR

This study evaluates how well MRI-based visual atrophy scales predict dementia in older adults, finding that the ERICA scale is most accurate.

Contribution

The study introduces the ERICA scale as a more accurate predictor of dementia compared to other visual atrophy scales in a clinical cohort.

Findings

ERICA average score had the highest AUC of 0.792 with 88.9% sensitivity and 51% specificity at a cut-off of 1.5.

ERICA average score was the best logistic regression model for predicting dementia after adjusting for age and education.

All visual atrophy scales showed significant differences between dementia and non-dementia groups.

Abstract

The Clinical Dementia Rating (CDR) is the most used categorization system for Alzheimer's Disease Dementia (ADD). In clinical practice, Magnetic Resonance Imaging (MRI) remains one of the most used methods for differential diagnosis and to improve accuracy, particularly with Visual Atrophy Scales (VAS).

To conduct an accuracy analysis of two VAS to predict dementia with CDR1 in an Outpatient Memory Clinic.

This is a cross‐sectional study, using data from the Cog‐Aging cohort study from 2019 to 2024. Participants underwent a comprehensive clinical, neuropsychological and neuroimaging assessment to determine CDR scores. Fifty‐two participants had MRI and assessments within 6 months. Participants were divided in Dementia (9) and Non‐Dementia (ND=43) groups. MRIs were assessed by two independent physicians, using The Entorhinal Cortical Atrophy (ERICA) scale and the Medial Temporal Lobe Atrophy (MTA) scale. They scored left and right, leading to an average score and a worst score. CDR Group scores were assessed with Mann‐Whitney. Furtheron, a Receiver Operating Characteristic (ROC) curve was done and cut‐off values were defined by Youden's J statistic. Finally, logistic regression models were done to predict CDR, considering VAS score, age and years of schooling. This study was approved by UFMG Ethics Committee.

Participants had a mean 76.28 years of age (SD 7.09) and median of 4.5 years of schooling (IQR 7.42). All VAS scores were significantly different between Dementia and ND. All scores had an Area Under the Curve (AUC) >0.715. ERICA average score had the highest AUC 0.792, achieving 88.9% sensitivity and 51% specificity at a cut‐off point of 1.5 and over. Finally, all scores were assessed by logistic regression adjusting for age and years of schooling, and only ERICA score's were significant (table 2). ERICA average score was the best model (Nagelkerke R2 0.356; p =  0.006) with an Odds‐Ratio of 17.43.

These findings suggest that ERICA average score was the most accurate method to detect Dementia in this sample, and also represented the best logistic regression model to predict Dementia when adjusted for years of schooling and age.

Figures