Magnetic Resonance Imaging based Texture Analysis Predicts Conversion to Mild Cognitive Impairment and Alzheimer Disease

Existing neuroimaging cannot reliably predict future conversion from cognitively unimpaired (CU) to mild cognitive impairment (MCI) or Alzheimer's disease (AD). This study utilized longitudinal texture analysis of T1-weighted MRI scans to identify CU individuals who converted to MCI/AD dementia versus non-converters. 151 CU individuals from ADNI were analyzed (62 progressed to MCI, 25 to AD). Textural features were extracted from regional brain volumes at baseline and time to conversion follow-up using a combination of FreeSurfer and Pyradiomics software. Multiple machine-learning models based on texture features were developed and tested to evaluate its performance to identify CU individuals who converted to MCI/AD dementia versus non-converters. Hippocampal texture changes showed highest importance for predicting conversion. Models using longitudinal texture changes (AUC 0.7) outperformed baseline only (AUC 0.56). When hippocampal texture heterogeneity increased beyond thresholds, conversion risk reached 90 percent versus 37 percent for others. Longitudinal hippocampal texture analysis better predicted cognitive decline than volumetric measures or static textures analyses. Tracking hippocampal texture alterations on T1 MRI may enable the early diagnosis, prognosis, and differentiation of Alzheimer's and other pathologies causing cognitive decline.