Associations of multiple visual rating scales based on structural magnetic resonance imaging with disease severity and cerebrospinal fluid biomarkers in patients with Alzheimer's disease

Abstract

The relationships between multiple visual rating scales based on structural magnetic resonance imaging (sMRI) with disease severity and cerebrospinal fluid (CSF) biomarkers in patients with Alzheimer's disease (AD) were ambiguous. In this study, a total of 438 patients with clinically diagnosed AD were recruited. All participants underwent brain sMRI scan, and medial temporal lobe atrophy (MTA), posterior atrophy (PA), global cerebral atrophy-frontal sub-scale (GCA-F), and Fazekas rating scores were visually evaluated. Meanwhile, disease severity was assessed by neuropsychological tests such as the Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), and Clinical Dementia Rating (CDR). Among them, 95 patients were tested for CSF core biomarkers, including Aβ_1-42, Aβ_1-40, Aβ_1-42/Aβ_1-40, p-tau, and t-tau. As a result, the GCA-F and Fazekas scales showed positively significant correlations with onset age (r = 0.181, p < 0.001; r = 0.411, p < 0.001, respectively). Patients with late-onset AD (LOAD) showed higher GCA-F and Fazekas scores (p < 0.001, p < 0.001). With regard to the disease duration, the MTA and GCA-F were positively correlated (r = 0.137, p < 0.05; r = 0.106, p < 0.05, respectively). In terms of disease severity, a positively significant association emerged between disease severity and the MTA, PA GCA-F, and Fazekas scores (p < 0.001, p < 0.001, p < 0.001, p < 0.05, respectively). Moreover, after adjusting for age, gender, and APOE alleles, the MTA scale contributed to moderate to severe AD in statistical significance independently by multivariate logistic regression analysis (p < 0.05). The model combining visual rating scales, age, gender, and APOE alleles showed the best performance for the prediction of moderate to severe AD significantly (AUC = 0.712, sensitivity = 51.5%, specificity = 84.6%). In addition, we observed that the MTA and Fazekas scores were associated with a lower concentration of Aβ_1-42 (p < 0.031, p < 0.022, respectively). In summary, we systematically analyzed the benefits of multiple visual rating scales in predicting the clinical status of AD. The visual rating scales combined with age, gender, and APOE alleles showed best performance in predicting the severity of AD. MRI biomarkers in combination with CSF biomarkers can be used in clinical practice.

Keywords: Alzheimer’s disease; cerebrospinal fluid; global cerebral frontal atrophy; medial temporal lobe atrophy; posterior atrophy; visual rating scale; white matter lesions.

FIGURE 1

Regression analyses of visual rating scales in the prediction of AD severity (N = 435). OR, odds ratio; 95% CI, 95% confidence interval; B, regression coefficient. P < 0.05 was statistically significant.

FIGURE 2

Receiver operating characteristic (ROC) curve analysis of visual rating scales for disease severity. The MTA exhibited the best predictive efficacy as a single indicator (AUC = 0.622, sensitivity = 74%, specificity = 44.6%). The PA, AUC = 0.59, sensitivity = 44.2%, specificity = 73.2%. The GCA-F, AUC = 0.609, sensitivity = 34.7%, specificity = 84.8%. The Fazekas, AUC = 0.51, sensitivity = 95.5%, specificity = 6.2%. The MTA combined with PA, GCA-F, and Fazekas; the predictive performance significantly improved (AUC = 0.654, sensitivity = 48.3%, specificity = 78.6%). The predictive model combining the MTA, PA, GCA-F, Fazekas, age, gender, and APOE alleles showed the best performance for the identification of moderate to severe AD significantly (AUC = 0.712, sensitivity = 51.5%, specificity = 84.6%). The Delong test was used to compare the difference of predictive models. AUC, area under the curve; ROC, receiver operating characteristic curve.

FIGURE 3

Each visual rating scale and correlation of CSF core biomarkers. (A) MTA (N = 92), (B) PA (N = 95), (C) GCA-F (N = 95), and (D) Fazekas (N = 95). P-values from ANOVA (Kruskal–Wallis H-test for abnormal distribution), and Games–Howell test for post-hoc comparison. *P < 0.05 was statistically significant; ns: no statistical difference.