Dataset
The data for my thesis was acquired from the OASIS-3 dataset which is a longitudinal study focusing on AD. OASIS-3 contains longitudinal neuroimaging, clinical and cognitive data sets collected over 15 years for 1098 subjects (487 male and 611 female) aged 42 to 95 years.
The OASIS-3 dataset incorporates over 2000 MR scans, of which only the sessions which
contained both TOF-MRA and T1-weighted images were included in this thesis.
T1-weighted and TOF-MRA images were collected on 2 different Siemens scanner models
(Siemens Medical Solutions USA, Inc): Vision 1.5T and TIM Trio 3T. Participants were
placed in a 16-channel head coil on 1.5T scanners and a 20-channel head coil on 3T scanners
with foam pad stabilizers placed next to the ears to decrease motion.
The TOF and T1-weighted images were reconstructed with the approximate resolution of 0.3×0.3×0.6mm and 1×1×1 mm, respectively.
All T1 weighted images were subjected to cortical reconstruction and volumetric segmentation. T1-weighted images collected on 1.5T scanner were processed with
Freesurfer v5.0 or v5.1 and images from 3T were segmented using Freesurfer 3.5. The Freesurfer results were required to investigate the regional vascular alterations. To each brain region, a colour ID is dedicated which can be found in the Freesurfer Lookup table.
Regarding the vascular changes, people with AD were compared with age-matched controls. In my thesis, 24 AD and 24 age-matched controls were chosen.
To eliminate the MRI field inhomogeneities, N4 bias field correction was applied to all the TOF-MRA datasets using the Python SimpleITK library.
T1-weighted images were subsequently registered on the TOF space to enable region-wise analysis using SPM.
Segmentation
To extract the vascular information, the vessels were segmented using OMELETTE (an open-source library). Using this library, vessels are initially enhanced using Jerman or Frangi filters and consequently segmented using hysteresis thresholding. In order to find the best vessel enhancement filter and the most efficient filtering parameter values, I created ground truths to which the automatically segmented vessels can be compared quantitatively and qualitatively. The process is as follows:
1- Eight TOF datasets were chosen randomly (four from healthy and four from AD subjects). 2- All datasets were segmented semi-manually using ilastik software.
3- The same datasets were also segmented automatically with OMELETTE using both Frangi and Jerman as enhancement filters to investigate which filter has better performance. For each filter, the combination of different parameter values was also used to choose the most efficient parameters.
4- Each semi-manually segmented data was then compared with the corresponding automatically segmented dataset qualitatively using maximum intensity projection and quantitatively using the dice coefficient.
5- Finally, after acquiring the right filter and parameter values, all the TOF datasets were segmented.
Metrics
In my thesis, I investigated the vascular changes in 12 brain regions including the left and right hippocampus, amygdala, caudate, putamen, pallidum, and thalamus. For each of these brain regions, the following metrics were calculated:
1- VDM which was acquired by applying Euclidean distance transform to the segmented vessels. After acquiring the distance values in each brain region, the average value of VDM was calculated.
2-Vessel density which was calculated by dividing the number of vessel voxels by the total number of voxels in each ROI.
3-Vessel volume which was calculated by multiplying the number of vessel voxels by the size of each voxel per ROI.
4-Finally the volume of each brain region was also calculated by multiplying the number of voxels by the size of each voxel per ROI.
To perform the calculations in each ROI, the color-coded segmentation and parcellation
masks (Aseg+Aparc) obtained from Freesurfer were mapped on the TOF-MRA data. The
mask values for each brain region were obtained from the Freesurfer lookup table.
Statistical Analysis
To compare the metric values in all 12 brain regions between healthy and AD subjects, Mann-Whittney U-test (non-parametric statistical approach) was used. The significant test only investigates the availability of the significant difference between groups however, the amount of difference is not specified in the results. The effect size measure (ESM) is a valuable tool that determines how significantly the measured metrics differ due to the presence of AD. In my thesis I used cliff's delta ESM using cliff's_delta Python package to compare healthy and AD subjects.
Комментарии