3.3. Data Preprocessing

3.3.1. rs-fMRI

‘’’★ ADNI 다운로드 형태: DICOM —> dcm2niix —> NIfTI 포맷으로 변환하여 사용 ★’’’

First, we converted the raw rs-fMRI data, downloaded from ADNI in digital imaging and communications in medicine (DICOM) format, to neuroimaging informatics technology initiative (NIfTI/NII) format using an open-source tool known as the dcm2niix software [87].

‘’’★ Brain Extraction: FSL-BET ★’’’

We removed skull and neck voxels considered non-brain regions from the structural T1-weighted imaging data corresponding to each fMRI time course using FSL-BET software [88].

‘’’★ Motion Correction: FSL-MCFLIRT★’’’

Third, using FSL-MCFLIRT [89], we corrected the rs-fMRI data for motion artifact caused by low-frequency drifts, which could negatively impact the time course decomposition.

‘’’Slice Timing Correction: HWSI‘’’

Finally, we applied a standard slice timing correction (STC) method known as Hanning-Windowed Sinc Interpolation (HWSI) to each voxel’s time series. According to the ADNI data acquisition protocol, the brain slices were acquired halfway through the relevant volume’s TR; therefore, we shifted each time series by a proper fraction relative to the middle point of TR period.

‘’’★ Registration: fMRI to T1w —> MNI152★’’’

We registered the fMRI brains to the corresponding high-resolution structural T1-weighted scans using an affine linear transformation with seven degrees of freedom (7 DOF). Subsequently, we aligned the registered brains to the Montreal Neurological Institute standard brain template (MNI152) using an affine linear transformation with 12 DOF [90].

‘’’Resampling: 4mm’’’

We resampled the aligned brains by a 4 mm kernel that generated 45 × 54 × 45 brain slices per time course.

3.3.2. Structural MRI

We preprocessed the structural MRI data from scratch using a 6-step pipeline where we first converted the DICOM raw images to NifTi/NII format using dcm2niix software [87].

Then, using the FSL-VBM library [91], we segmented the brain images into grey matter (GM), white matter (WM), and cerebrospinal fluid (CSF).

We used the GM images to register to the GM ICBM-152 standard template using a linear affine transformation with 6 DOF.

Next, we concatenated the brain images, flipped them along the x-axis, then re-averaged to create a first-pass, study-specific template as a standard approach [88].