Segmentation Improvement of Sublobar Gray Matter Using Multi-modal Voxel-based Morphometry
碩士 === 國立陽明大學 === 腦科學研究所 === 101 === Background: Conventional voxel-based morphometry (VBM) protocols rely on T1-weighted image (T1WI) for analyses with proposed major limitation in (1) inaccurate tissue segmentation, (2) signal inhomogeneities and (3) field dependence. Patients with focal cortical...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | en_US |
Published: |
2013
|
Online Access: | http://ndltd.ncl.edu.tw/handle/83831087296103337446 |
id |
ndltd-TW-101YM005659015 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-101YM0056590152016-03-18T04:41:52Z http://ndltd.ncl.edu.tw/handle/83831087296103337446 Segmentation Improvement of Sublobar Gray Matter Using Multi-modal Voxel-based Morphometry 多模態-體素-型態學改善腦葉下灰質組織的分割 Pi-Yu Hsu 許必瑜 碩士 國立陽明大學 腦科學研究所 101 Background: Conventional voxel-based morphometry (VBM) protocols rely on T1-weighted image (T1WI) for analyses with proposed major limitation in (1) inaccurate tissue segmentation, (2) signal inhomogeneities and (3) field dependence. Patients with focal cortical dysplasia (FCD) have involved in several VBM studies because FCD may be very subtle in appearance and might be not clearly visible by conventional clinical diagnosis. The comparison between single patient and normal database was used to detect the potential lesion of FCD. Hypothesis: The VBM algorithms with multi-modal approach provide (1) improved segmentation of sublobar gray-matter (GM) in comparison with single-modal approach, and (2) the potential applications in detecting FCD with the abnormal tissue of sublobar GM. Materials and Methods: With brain image data of 24 normal subjects using both 1.5T and 3T magnetic resonance imaging (MRI), GM segmentation was obtained with single- and multi-modal approaches using customized VBM. After de-scalping and bias-field correction, the images were segmented with single-modal [that using T1WI only, we called them as FAST-1 (FMRIB’s Automated Segmentation Tool-1) and SPM8-1 (Statistical Parametric Mapping 8-1)] and multi-modal [that combing the information of T1WI, proton-density weighted image (PDWI) and T2-weighted image (T2WI), we called them as FAST-3 and SPM8-3] approaches by FSL (FMRIB Software Library) and SPM8 algorithms. Group comparison of GM probabilities was made after the images normalized to the customized template of GM. Additionally, two patients of FCD were compared with the normal database which was created from 24 normal subjects, and were evaluated for the potential detection of brain lesions by single- or multi-modal VBM approaches. Results: (1) Based on optimized and customized VBM protocol, our results demonstrated the order of GM probability as FAST-3 > SPM8-3 > SPM8-1 > FAST-1 by statistical inferences using paired t-test with false discovery rate (FDR), p < 1×10-5. By validation using BrainWeb (simulated brain MRI data) and expert-based (real brain data) as ground truth, multi-modal approach (FAST-3) showed higher sensitivity and similar indices of GM segmentation (GM probability > 50%) in the basal ganglia with expert-based validataion. (2) The results of two FCD patients with z value > 3 showed focal lesions (based on the post-operation T1WI) was detected by both single-modal and multi-modal approaches, but multi-modal approach showed more regional differences as compared with single-modal approach. Conclusion: VBM protocol was optimized by comparing the sensitivity and similarity of GM segmented by single- and multi-modal approaches. FAST-3 demonstrated improved segmentation of sublobar GM using multiple image modalities. VBM approaches may be used to assist or remind neuroradiologists to detect the subtle lesions, e.g. FCD. Tzu-Chen Yeh 葉子成 2013 學位論文 ; thesis 85 en_US |
collection |
NDLTD |
language |
en_US |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立陽明大學 === 腦科學研究所 === 101 === Background: Conventional voxel-based morphometry (VBM) protocols rely on T1-weighted image (T1WI) for analyses with proposed major limitation in (1) inaccurate tissue segmentation, (2) signal inhomogeneities and (3) field dependence. Patients with focal cortical dysplasia (FCD) have involved in several VBM studies because FCD may be very subtle in appearance and might be not clearly visible by conventional clinical diagnosis. The comparison between single patient and normal database was used to detect the potential lesion of FCD. Hypothesis: The VBM algorithms with multi-modal approach provide (1) improved segmentation of sublobar gray-matter (GM) in comparison with single-modal approach, and (2) the potential applications in detecting FCD with the abnormal tissue of sublobar GM. Materials and Methods: With brain image data of 24 normal subjects using both 1.5T and 3T magnetic resonance imaging (MRI), GM segmentation was obtained with single- and multi-modal approaches using customized VBM. After de-scalping and bias-field correction, the images were segmented with single-modal [that using T1WI only, we called them as FAST-1 (FMRIB’s Automated Segmentation Tool-1) and SPM8-1 (Statistical Parametric Mapping 8-1)] and multi-modal [that combing the information of T1WI, proton-density weighted image (PDWI) and T2-weighted image (T2WI), we called them as FAST-3 and SPM8-3] approaches by FSL (FMRIB Software Library) and SPM8 algorithms. Group comparison of GM probabilities was made after the images normalized to the customized template of GM. Additionally, two patients of FCD were compared with the normal database which was created from 24 normal subjects, and were evaluated for the potential detection of brain lesions by single- or multi-modal VBM approaches. Results: (1) Based on optimized and customized VBM protocol, our results demonstrated the order of GM probability as FAST-3 > SPM8-3 > SPM8-1 > FAST-1 by statistical inferences using paired t-test with false discovery rate (FDR), p < 1×10-5. By validation using BrainWeb (simulated brain MRI data) and expert-based (real brain data) as ground truth, multi-modal approach (FAST-3) showed higher sensitivity and similar indices of GM segmentation (GM probability > 50%) in the basal ganglia with expert-based validataion. (2) The results of two FCD patients with z value > 3 showed focal lesions (based on the post-operation T1WI) was detected by both single-modal and multi-modal approaches, but multi-modal approach showed more regional differences as compared with single-modal approach. Conclusion: VBM protocol was optimized by comparing the sensitivity and similarity of GM segmented by single- and multi-modal approaches. FAST-3 demonstrated improved segmentation of sublobar GM using multiple image modalities. VBM approaches may be used to assist or remind neuroradiologists to detect the subtle lesions, e.g. FCD.
|
author2 |
Tzu-Chen Yeh |
author_facet |
Tzu-Chen Yeh Pi-Yu Hsu 許必瑜 |
author |
Pi-Yu Hsu 許必瑜 |
spellingShingle |
Pi-Yu Hsu 許必瑜 Segmentation Improvement of Sublobar Gray Matter Using Multi-modal Voxel-based Morphometry |
author_sort |
Pi-Yu Hsu |
title |
Segmentation Improvement of Sublobar Gray Matter Using Multi-modal Voxel-based Morphometry |
title_short |
Segmentation Improvement of Sublobar Gray Matter Using Multi-modal Voxel-based Morphometry |
title_full |
Segmentation Improvement of Sublobar Gray Matter Using Multi-modal Voxel-based Morphometry |
title_fullStr |
Segmentation Improvement of Sublobar Gray Matter Using Multi-modal Voxel-based Morphometry |
title_full_unstemmed |
Segmentation Improvement of Sublobar Gray Matter Using Multi-modal Voxel-based Morphometry |
title_sort |
segmentation improvement of sublobar gray matter using multi-modal voxel-based morphometry |
publishDate |
2013 |
url |
http://ndltd.ncl.edu.tw/handle/83831087296103337446 |
work_keys_str_mv |
AT piyuhsu segmentationimprovementofsublobargraymatterusingmultimodalvoxelbasedmorphometry AT xǔbìyú segmentationimprovementofsublobargraymatterusingmultimodalvoxelbasedmorphometry AT piyuhsu duōmótàitǐsùxíngtàixuégǎishànnǎoyèxiàhuīzhìzǔzhīdefēngē AT xǔbìyú duōmótàitǐsùxíngtàixuégǎishànnǎoyèxiàhuīzhìzǔzhīdefēngē |
_version_ |
1718207887926362112 |