Limits...
An Example-Based Multi-Atlas Approach to Automatic Labeling of White Matter Tracts.

Yoo SW, Guevara P, Jeong Y, Yoo K, Shin JS, Mangin JF, Seong JK - PLoS ONE (2015)

Bottom Line: To further enhance the computational efficiency, we implemented our approach on the graphics processing unit (GPU).Through nested cross-validation we demonstrated that our approach yielded high classification performance.The average sensitivities for bundles in the left and right hemispheres were 89.5% and 91.0%, respectively, and their average false discovery rates were 14.9% and 14.2%, respectively.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Engineering, Korea University, Seoul, Republic of Korea; Department of Computer Science, KAIST, Daejeon, Republic of Korea.

ABSTRACT
We present an example-based multi-atlas approach for classifying white matter (WM) tracts into anatomic bundles. Our approach exploits expert-provided example data to automatically classify the WM tracts of a subject. Multiple atlases are constructed to model the example data from multiple subjects in order to reflect the individual variability of bundle shapes and trajectories over subjects. For each example subject, an atlas is maintained to allow the example data of a subject to be added or deleted flexibly. A voting scheme is proposed to facilitate the multi-atlas exploitation of example data. For conceptual simplicity, we adopt the same metrics in both example data construction and WM tract labeling. Due to the huge number of WM tracts in a subject, it is time-consuming to label each WM tract individually. Thus, the WM tracts are grouped according to their shape similarity, and WM tracts within each group are labeled simultaneously. To further enhance the computational efficiency, we implemented our approach on the graphics processing unit (GPU). Through nested cross-validation we demonstrated that our approach yielded high classification performance. The average sensitivities for bundles in the left and right hemispheres were 89.5% and 91.0%, respectively, and their average false discovery rates were 14.9% and 14.2%, respectively.

No MeSH data available.


Sensitivity and FDR histograms for tract group labeling, direct tract labeling, and Guevara’s method.Top left: sensitivities of the three methods for each anatomic bundle, Top right: sensitivities of the three methods for each example subject. Bottom left: FDRs of the three methods for each anatomic bundle, Bottom right: FDRs of the three methods for each example subject.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4520495&req=5

pone.0133337.g008: Sensitivity and FDR histograms for tract group labeling, direct tract labeling, and Guevara’s method.Top left: sensitivities of the three methods for each anatomic bundle, Top right: sensitivities of the three methods for each example subject. Bottom left: FDRs of the three methods for each anatomic bundle, Bottom right: FDRs of the three methods for each example subject.

Mentions: Fig 8 illustrates the sensitivity and FDR of the tract group labeling and direct tract labeling. (Results for the Guevara et al.’s method will be described in Section Comparison with a Guevara et al.’s method.) The histograms in the left column shows that high sensitivities were observed for most anatomic bundles. For the tract group labeling, the sensitivities of the left/right anterior thalamic radiation (ATR), left/right corticospinal tract (CST), right inferior fronto-occipital fasciculus (IFO), right inferior longitudinal fasciculus (ILF), and left/right uncinate fasciculus (UNC) were more than 90%. The majority of anatomic bundles showed FDRs below 20%. Relatively high FDRs (more than 20%) for the ILF(L) and UNC(L,R) are due to their complex shapes and high shape variability across the example subjects. For the majority of subjects, high sensitivities (more than 90%) and low FDRs (less than 16%) were observed. The sensitivity of the direct group labeling was higher than that of tract grouping results, while the precision (=1-FDR) was lower: Specifically, the average sensitivity of tract group labeling results was 90.2% (LH: 89.5%, RH: 91.0%), while the average sensitivity of direct tract labeling results was 91.9% (LH: 91.7%, RH: 92.1%). Also, the average FDR of tract group labeling results was 14.6% (LH: 14.9%, RH: 14.2%), while the average FDR of direct tract labeling results was 17.1% (LH: 17.7%, RH: 16.4%). The group-based method showed smaller sensitivity and FDR because every tracts in a group are simultaneously labeled or considered as outlier. If some tracts in a group have different shapes from the ground-truth, then every tract in the group will have more chance to be labeled as outlier even if other tracts are similar to the ground-truth. We think that these situations could happen more if a tight distance threshold is used. Additionally, the FDR of the group-based result could be lower because the outliers are already removed in the tract grouping stage.


An Example-Based Multi-Atlas Approach to Automatic Labeling of White Matter Tracts.

Yoo SW, Guevara P, Jeong Y, Yoo K, Shin JS, Mangin JF, Seong JK - PLoS ONE (2015)

Sensitivity and FDR histograms for tract group labeling, direct tract labeling, and Guevara’s method.Top left: sensitivities of the three methods for each anatomic bundle, Top right: sensitivities of the three methods for each example subject. Bottom left: FDRs of the three methods for each anatomic bundle, Bottom right: FDRs of the three methods for each example subject.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4520495&req=5

pone.0133337.g008: Sensitivity and FDR histograms for tract group labeling, direct tract labeling, and Guevara’s method.Top left: sensitivities of the three methods for each anatomic bundle, Top right: sensitivities of the three methods for each example subject. Bottom left: FDRs of the three methods for each anatomic bundle, Bottom right: FDRs of the three methods for each example subject.
Mentions: Fig 8 illustrates the sensitivity and FDR of the tract group labeling and direct tract labeling. (Results for the Guevara et al.’s method will be described in Section Comparison with a Guevara et al.’s method.) The histograms in the left column shows that high sensitivities were observed for most anatomic bundles. For the tract group labeling, the sensitivities of the left/right anterior thalamic radiation (ATR), left/right corticospinal tract (CST), right inferior fronto-occipital fasciculus (IFO), right inferior longitudinal fasciculus (ILF), and left/right uncinate fasciculus (UNC) were more than 90%. The majority of anatomic bundles showed FDRs below 20%. Relatively high FDRs (more than 20%) for the ILF(L) and UNC(L,R) are due to their complex shapes and high shape variability across the example subjects. For the majority of subjects, high sensitivities (more than 90%) and low FDRs (less than 16%) were observed. The sensitivity of the direct group labeling was higher than that of tract grouping results, while the precision (=1-FDR) was lower: Specifically, the average sensitivity of tract group labeling results was 90.2% (LH: 89.5%, RH: 91.0%), while the average sensitivity of direct tract labeling results was 91.9% (LH: 91.7%, RH: 92.1%). Also, the average FDR of tract group labeling results was 14.6% (LH: 14.9%, RH: 14.2%), while the average FDR of direct tract labeling results was 17.1% (LH: 17.7%, RH: 16.4%). The group-based method showed smaller sensitivity and FDR because every tracts in a group are simultaneously labeled or considered as outlier. If some tracts in a group have different shapes from the ground-truth, then every tract in the group will have more chance to be labeled as outlier even if other tracts are similar to the ground-truth. We think that these situations could happen more if a tight distance threshold is used. Additionally, the FDR of the group-based result could be lower because the outliers are already removed in the tract grouping stage.

Bottom Line: To further enhance the computational efficiency, we implemented our approach on the graphics processing unit (GPU).Through nested cross-validation we demonstrated that our approach yielded high classification performance.The average sensitivities for bundles in the left and right hemispheres were 89.5% and 91.0%, respectively, and their average false discovery rates were 14.9% and 14.2%, respectively.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Engineering, Korea University, Seoul, Republic of Korea; Department of Computer Science, KAIST, Daejeon, Republic of Korea.

ABSTRACT
We present an example-based multi-atlas approach for classifying white matter (WM) tracts into anatomic bundles. Our approach exploits expert-provided example data to automatically classify the WM tracts of a subject. Multiple atlases are constructed to model the example data from multiple subjects in order to reflect the individual variability of bundle shapes and trajectories over subjects. For each example subject, an atlas is maintained to allow the example data of a subject to be added or deleted flexibly. A voting scheme is proposed to facilitate the multi-atlas exploitation of example data. For conceptual simplicity, we adopt the same metrics in both example data construction and WM tract labeling. Due to the huge number of WM tracts in a subject, it is time-consuming to label each WM tract individually. Thus, the WM tracts are grouped according to their shape similarity, and WM tracts within each group are labeled simultaneously. To further enhance the computational efficiency, we implemented our approach on the graphics processing unit (GPU). Through nested cross-validation we demonstrated that our approach yielded high classification performance. The average sensitivities for bundles in the left and right hemispheres were 89.5% and 91.0%, respectively, and their average false discovery rates were 14.9% and 14.2%, respectively.

No MeSH data available.