Groupwise segmentation improves neuroimaging classification accuracy

Authors:
Yaping Wang;Hongjun Jia;Pew-Thian Yap;Bo Cheng;Chong-Yaw Wee;Lei Guo;Dinggang Shen
Affiliations:
School of Automation, Northwestern Polytechnical University, Xi’an, Shaanxi Province, China, Department of Radiology and BRIC, University of North Carolina, Chapel Hill;Department of Radiology and BRIC, University of North Carolina, Chapel Hill;Department of Radiology and BRIC, University of North Carolina, Chapel Hill;Department of Radiology and BRIC, University of North Carolina, Chapel Hill;Department of Radiology and BRIC, University of North Carolina, Chapel Hill;School of Automation, Northwestern Polytechnical University, Xi’an, Shaanxi Province, China;Department of Radiology and BRIC, University of North Carolina, Chapel Hill
Venue:
MBIA'12 Proceedings of the Second international conference on Multimodal Brain Image Analysis
Year:
2012

Citing 7
Cited 0

SUSAN—A New Approach to Low Level Image Processing

International Journal of Computer Vision
Bilateral Filtering for Gray and Color Images

ICCV '98 Proceedings of the Sixth International Conference on Computer Vision
Impact of Rician Adapted Non-Local Means Filtering on HARDI

MICCAI '08 Proceedings of the 11th International Conference on Medical Image Computing and Computer-Assisted Intervention, Part II
Rician Noise Removal by Non-Local Means Filtering for Low Signal-to-Noise Ratio MRI: Applications to DT-MRI

MICCAI '08 Proceedings of the 11th International Conference on Medical Image Computing and Computer-Assisted Intervention, Part II
Robust deformable-surface-based skull-stripping for large-scale studies

MICCAI'11 Proceedings of the 14th international conference on Medical image computing and computer-assisted intervention - Volume Part III
Rician noise removal in diffusion tensor MRI

MICCAI'06 Proceedings of the 9th international conference on Medical Image Computing and Computer-Assisted Intervention - Volume Part I
Wavelet-based Rician noise removal for magnetic resonance imaging

IEEE Transactions on Image Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Accurate diagnosis of Alzheimer's disease (AD), especially mild cognitive impairment (MCI), is critical for treatment of the disease. Many algorithms have been proposed to improve classification performance. While most existing methods focus on exploring different feature extraction and selection techniques, in this paper, we show that the pre-processing steps for MRI scans, i.e., registration and segmentation, significantly affect the classification performance. Specifically, we evaluate the classification performance given by a multi-atlas based multi-image segmentation (MABMIS) method, with respect to more conventional segmentation methods. By incorporating tree-based groupwise registration and iterative groupwise segmentation strategies, MABMIS attains more accurate and consistent segmentation results compared with the conventional methods that do not take into account the inherent distribution of images under segmentation. This increased segmentation accuracy will benefit classification by minimizing errors that are propagated to the subsequent analysis steps. Experimental results indicate that MABMIS achieves better performance when compared with the conventional methods in the following classification tasks using the ADNI dataset: AD vs. MCI (accuracy: 71.8%), AD vs. healthy control (HC) (89.1%), progressive MCI vs. HC (84.4%), and progressive MCI vs. stable MCI (70.0%). These results show that pre-processing the images accurately is critical for neuroimaging classification.