Title : Radiomics evaluation for the early detection of alzheimer's dementia using T1-weighted MRI
Abstract:
Introduction: Early detection of Alzheimer's disease (AD) is crucial for implementing timely interventions and improving patient outcomes. This study evaluates the efficacy of radiomics features extracted from T1-weighted magnetic resonance imaging (MRI) for the early detection of AD.
Methods: We retrospectively analyzed T1-weighted MRI scans from 427 participants (142 with early AD, 153 with mild cognitive impairment (MCI), and 132 age-matched healthy controls) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. A comprehensive radiomics pipeline was implemented to extract 1,072 quantitative features from segmented regions of interest, including the hippocampus, entorhinal cortex, and posterior cingulate. Features encompassed first-order statistics, shape metrics, textural properties, and wavelet transformations. Feature selection using least absolute shrinkage and selection operator (LASSO) identified 37 radiomics features with optimal discriminatory power.
Results: A random forest classifier trained on these selected features achieved an area under the receiver operating characteristic curve (AUC) of 0.91 (95% CI: 0.88-0.94) for distinguishing early AD from healthy controls and 0.83 (95% CI: 0.79-0.87) for distinguishing MCI from healthy controls in the independent test cohort. Notably, textural heterogeneity features in the hippocampus and shape features in the entorhinal cortex demonstrated the highest discriminative value. The radiomics signature outperformed conventional volumetric analysis (AUC 0.91 vs. 0.82, p<0.001) and performed comparably to a deep learning convolutional neural network model (AUC 0.91 vs. 0.93, p=0.14) while offering greater interpretability. Longitudinal analysis revealed that specific radiomics features showed significant changes up to 3.2 years before clinical diagnosis of AD. Integration of radiomics features with demographic and cognitive assessment data further improved prediction performance (AUC 0.94, 95% CI: 0.91-0.97).
Conclusion: These findings suggest that radiomics analysis of T1-weighted MRI provides valuable imaging biomarkers for early detection of AD, potentially facilitating earlier clinical intervention and enrichment of clinical trials targeting early-stage disease.
Keywords: Alzheimer's Disease, Radiomics, Early Detection, Machine Learning, Neuroimaging Biomarkers.
