Individualized Prediction of Early Alzheimer's Disease Based on Magnetic Resonance Imaging Radiomics, Clinical, and Laboratory Examinations: A 60-Month Follow-Up Study

Abstract

Background: Accurately predicting whether and when mild cognitive impairment (MCI) will progress to Alzheimer's disease (AD) is of vital importance to help developing individualized treatment plans to defer the occurrence of irreversible dementia.

Purpose: To develop and validate radiomics models and multipredictor nomogram for predicting the time to progression (TTP) from MCI to AD.

Study type: Retrospective.

Population: One hundred sixty-two MCI patients (96 men and 66 women [median age, 72; age range, 56-88 years]) were included from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.

Field strength/sequence: T₁ -weighted imaging and T₂ -weighted fluid-attenuation inversion recovery imaging acquired at 3.0 T.

Assessment: During the 5-year follow-up, 68 patients converted to AD and 94 remained stable. Patients were randomly divided into the training (n = 112) and validation datasets (n = 50). Radiomic features were extracted from the whole cerebral cortex and subcortical nucleus of MR images. A radiomics model was established using least absolute shrinkage and selection operator (LASSO) Cox regression. The clinical-laboratory model and radiomics-clinical-laboratory model were developed by multivariate Cox proportional hazard model. The performance of each model was assessed by the concordance index (C-index). A multipredictor nomogram derived from the radiomics-clinical-laboratory model was constructed for individualized TTP estimation.

Statistical tests: LASSO cox regression, univariate and multivariate Cox regression, Kaplan-Meier analysis and Student's t test were performed.

Results: The C-index of the radiomics, clinical-laboratory and radiomics-clinical-laboratory models were 0.924 (95% confidence interval [CI]: 0.894-0.952), 0.903 (0.868-0.938), 0.950 (0.929-0.971) in the training cohort and 0.811 (0.707-0.914), 0.901 (0824-0.977), 0.907 (0.836-0.979) in the validation cohort, respectively. A multipredictor nomogram with 15 predictors was established, which had high accuracy for individual TTP prediction with the C-index of 0.950 (0.929-0.971).

Data conclusion: The prediction of individual TTP from MCI to AD could be accurately conducted using the radiomics-clinical-laboratory model and multipredictor nomogram.

Evidence level: 3 TECHNICAL EFFICACY: 2.

Keywords: Alzheimer's disease; machine learning; magnetic resonance imaging; mild cognitive impairment; radiomics; time to progression.