Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu).
The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial MRI, positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment MCI and early AD.
The ADNI3 project was launched in 2016 to determine the relationships between clinical, cognitive, imaging, genetic, and biochemical biomarkers across the spectrum of AD.
Scientists of this institution are working on 59 research centers in the United States and Canada. This project aims to identify the pathological changes in the brain that trigger these disorders.
Based on the available data, 35 cognitively normal and 46 participants with cognitive impairment disorders were included in this study.
The cognitive disorders group is classified into four groups based on cognitive test scores and assessments performed by ADNI specialists.
Inclusion criteria for subject selection include:
1. People should be in one of the five groups CN, SMC, EMCI, LMCI, and AD.
2. Multi-echo GRE and T1W scans of these people should be available.
3. Phase and magnitude images should be available separately.
4. Information about cognitive information such as MMSE scores should be available.
Participants' demographic characteristics information is given in Table 1.
2-3. MRI Acquisition
All patient’s MRI scans were performed using the Siemens Prisma 3.0T MRI scanner equipped with a Head-Neck coil.
3D Accelerated_Sagittal_MPRAGE sequence were acquired with the following parameters: TR (ms)=2300, TE/TI (ms)=2.98/ 900, Slice Thickness (mm)=1, Flip Angle=9, Pixel Bandwidth=240 and Acquisition Matrix and Reconstruction Matrix=240.
GRE multi-echo sequence was obtained with 3 different time of echoes (3TE) with the following parameters: TR (ms)=650, TE1-TE2-TE3 (ms)=6.09-13-20, Slice Thickness (mm)=4, Flip Angle=20, Pixel Bandwidth=260, Matrix size=256×256, Voxel size x (mm)= 0.859375, Voxel size y (mm)=0.859375 and Number of slices=44.
2-4. Image Processing
2-4-1. Quantitative Susceptibility Mapping
QSM reconstruction has four steps: generating tissue mask, phase unwrapping, background field removal, and field-to-susceptibility inversion (Figure1).
Each of these steps is performed with different algorithms and toolboxes(23).
After making QSM images, it is time to segment and determine the region of interest (ROIs) to evaluate the QSM values of each area in terms of ppm.
The basis of this reconstruction is done on the brain tissue, so first, we need to separate the area of the brain tissue from the skull bone in the magnitude image.
This brain tissue extraction was performed using the Brain Extraction Tool from FMRIB Software Library (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/BET)(28).
In the next step, to prevent the occurrence of the aliasing artifact in the phase images and total QSM reconstruction, we performed the phase unwrapping using a Laplacian-based phase unwrapping tool from STI Suite MATLAB toolbox (https://www.eecs.berkeley.edu/~chunlei.liu/software.html)(29).
In order to eliminate the unwanted consequences of the border areas, the background phase removal was done using the V-SHARP tool from the STI suite MATLAB toolbox.
Finally, QSM reconstruction or the susceptibility map was computed by the streaking artifacts reduction (STAR) algorithm(30, 31).
For the accuracy of the between-group analysis, a reference was made to the brain mask during QSM reconstruction(32).
The SEPIA toolbox was used to perform the above steps ( https://github.com/kschan0214/sepia)(33).
2-4-2. Automatic Segmentation
To create the region of interest (ROIs) mask, we performed automatic segmentation using the FMRIB Software Library, a model-based segmentation/registration tool.(https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FIRST)(34).
At this stage, the location of 12 nuclei was determined, including left-thalamus, left-caudate, left-putamen, left-pallidum, left-Hippocampus, left-amygdala, right-Thalamus, right-caudate, right-putamen, right-pallidum, right-hippocampus, and right-amygdala.
2-5. Quantitative Susceptibility Analysis
In this step, we measured the magnetic susceptibility values of each nucleus using 3D Slicer software (http://www.slicer.org)(35).
For each participant, the segmented mask and QSM image were defined as input for the software; and based on the FSL guide (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FIRST/UserGuide), we specified the nuclei name of each segment.
Finally, we saved the mean magnetic susceptibility value for further statistical analysis from the calculated statistical parameters for each nucleus.
2-6. Statistical analysis
Statistical analyzes were performed with the help of IBM Statistic SPSS (Statistical package for social science) software V26.
Comparing the age and MMSE scores variables among the groups was done using one-way analysis of variance (ANOVA).
The chi-squared test was used for the same purpose about the qualitative variable of gender.
After examining the normal distribution of data with the Kolmogorov-Smirnov test, ANOVA with post hoc test (Tukey-Kramer test) was used to assess the significant difference between the mean magnetic susceptibility of brain nuclei in 5 groups for normality values.
Kruskal–Wallis, and Mann–Whitney–Wilcoxon tests were also performed for non-normality values with the same purpose.
Finally, ROC curve analysis was performed on the brain nuclei of participants in all five groups.
P ≤ 0.05 was considered statistically significant in all statistical tests.