نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Frequency-Domain Hippocampal MRI Analysis Using Distance Correlation for Early Alzheimer Detection

Subject:

Advanced Techniques in Signal Acquisition and Image Processing

Presenter Author:
Name:	Amirhosein Nasrollahi

Author:
Name 1:	Amirhosein Nasrollahi - Corresponding - Presenter
Affiliation 1:	Department of Energy Engineering and Physics, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran.

Name 2:	Shahyar Saramad
Affiliation 2:	Department of Energy Engineering and Physics, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran.

Name 3:	Saeed Setayeshi
Affiliation 3:	Department of Energy Engineering and Physics, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran.

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Early identification of Alzheimer’s disease (AD) benefits from imaging markers that capture subtle hippocampal alterations before advanced atrophy. However, similarity measures computed directly on spatial-domain MRI can show overlap between cognitively normal (CN) individuals and early mild cognitive impairment (EMCI). We investigated whether adding a frequency-domain representation of the hippocampus and using distance correlation (a dependence measure sensitive to both linear and nonlinear relationships) can improve separation of CN, EMCI, and AD groups when each subject is compared to a standard atlas.

Methods:

T1-weighted 3T MRI scans (MP-RAGE; 176 slices; 1.2-mm thickness) were selected from the ADNI cohort for three groups (n=50 each: CN, EMCI, AD; age 50–70). Images were denoised using gradient anisotropic diffusion and intensity-normalized, then registered to the MNI152 template using a 12-DOF transform. The hippocampus was segmented in 3D Slicer and extracted as a 3D region of interest (ROI). Each ROI was analyzed in the spatial domain and after a 3D fast Fourier transform (3D-FFT) to obtain magnitude spectra, focusing on a mid-frequency band (5–20% of the normalized spectrum). Similarity to the atlas ROI was quantified using Pearson, Spearman, Kendall, and distance correlation. Decision ranges were calibrated on a training split (30 subjects per group) to define non-overlapping confidence intervals, and performance was assessed on a held-out test split (20 subjects per group) using precision, specificity, sensitivity, F1-score, accuracy, and an uncertainty rate (percentage of subjects falling outside predefined ranges).

Results:

Between-group differences were significant across measures (ANOVA p<0.01), with large F-values (F>1000), suggesting substantial between-group variance relative to within-group variance after preprocessing. In the image domain, conventional correlations separated CN from AD more clearly than CN from EMCI, with marked CN–EMCI overlap (e.g., Pearson means: CN 0.821 vs EMCI 0.820). Frequency-domain features improved CN–EMCI separation (Pearson means: CN 0.810 vs EMCI 0.585) and generally reduced variability. Distance correlation showed the most distinct separation in both domains; in the frequency domain, mean distance-correlation similarity to the atlas was 0.960 (CN), 0.550 (EMCI), and 0.532 (AD), with low standard deviations (0.008–0.010). In this split-sample evaluation, distance correlation yielded the widest non-overlapping confidence intervals (75% in the image domain and 95% in the frequency domain). On the held-out test split, discrimination was very high (metrics approached 1.00 in this experimental setup), and uncertainty rates were reduced substantially—from 23.08% for EMCI (Pearson-image) to 0% for distance correlation in both domains; these findings should be interpreted cautiously given the controlled cohort and sample size.

Conclusion:

Frequency-domain hippocampal analysis may better capture distributed textural alterations relevant to early impairment, and distance correlation appears to provide a robust atlas-based similarity metric for group discrimination within this cohort. Further external validation on larger and more diverse datasets (including multi-scanner acquisitions and longitudinal follow-up) will be important to assess generalizability prior to clinical translation

Keywords:

Alzheimer Disease - Mild Cognitive Impairment - Magnetic Resonance Imaging - Hippocampus - Fourier Analysis - Image Processing