Background: Early detection of Alzheimer’s disease (AD) remains challenging, particularly when

 

distinguishing early mild cognitive impairment (EMCI) from cognitively normal (CN) individuals. Because

 

hippocampal alterations emerge early in the disease course, hippocampal MRI may provide sensitive

 

neuroimaging biomarkers. We evaluated whether frequency-domain hippocampal features and robust

 

dependence measures improve separation among CN, EMCI, AD.

 

Methods: T1-weighted 3T MP-RAGE MRI scans were obtained from ADNI for CN, EMCI, and AD

 

(n=50/group; age 50-70). Images were denoised and intensity-normalized, registered to the MNI152 atlas

 

(12 DOF), and the hippocampus was segmented and extracted. Hippocampal regions were analyzed in

 

spatial and frequency domains using 3D FFT. Similarity to the atlas was quantified using Pearson,

 

Spearman, Kendall, and distance correlation (capturing linear and nonlinear dependencies). Group

 

differences were assessed with ANOVA, and separability/classification was evaluated using nonoverlapping confidence-interval decision ranges and standard performance metrics

 

Results: Spatial-domain measures separated AD from CN but showed limited sensitivity for EMCI.

 

Frequency-domain features improved EMCI versus CN differentiation, increasing sensitivity and F1-score

 

relative to the spatial domain. Across both domains, distance correlation provided the strongest group

 

separation, yielding statistically significant between-group differences (p<1%) and the largest nonoverlapping confidence intervals, reaching 95% confidence in the frequency domain.

 

Conclusion: Frequency-domain hippocampal MRI analysis enhances detection of subtle early-stage

 

changes and improves EMCI discrimination compared with spatial-domain analysis. Distance correlation,

 

by capturing nonlinear dependencies, offers a robust similarity measure for neuroimaging-based early AD

 

detection and may provide strong features for downstream machine-learning and deep-learning.