Limits...
Standardization of size, shape and internal structure of spinal cord images: comparison of three transformation methods.

Fujiki Y, Yokota S, Okada Y, Oku Y, Tamura Y, Ishiguro M, Miwakeichi F - PLoS ONE (2013)

Bottom Line: To solve this problem, we attempted to standardize transversely sectioned spinal cord images focusing on the laminar structure in the gray matter.In this study, we used neuron-specific marker (NeuN)-stained histological images of transversely sectioned cervical spinal cord slices (21 images obtained from 4 rats) to create the standard atlas and also to serve for benchmark tests.This novel image standardization technique would be applicable to optical recording such as voltage-sensitive dye imaging, and will enable statistical evaluations of neural activation across multiple samples.

View Article: PubMed Central - PubMed

Affiliation: Department of Statistical Science, School of Multidisciplinary Sciences, The Graduate University for Advanced Studies, Tachikawa, Tokyo, Japan.

ABSTRACT
Functional fluorescence imaging has been widely applied to analyze spatio-temporal patterns of cellular dynamics in the brain and spinal cord. However, it is difficult to integrate spatial information obtained from imaging data in specific regions of interest across multiple samples, due to large variability in the size, shape and internal structure of samples. To solve this problem, we attempted to standardize transversely sectioned spinal cord images focusing on the laminar structure in the gray matter. We employed three standardization methods, the affine transformation (AT), the angle-dependent transformation (ADT) and the combination of these two methods (AT+ADT). The ADT is a novel non-linear transformation method developed in this study to adjust an individual image onto the template image in the polar coordinate system. We next compared the accuracy of these three standardization methods. We evaluated two indices, i.e., the spatial distribution of pixels that are not categorized to any layer and the error ratio by the leave-one-out cross validation method. In this study, we used neuron-specific marker (NeuN)-stained histological images of transversely sectioned cervical spinal cord slices (21 images obtained from 4 rats) to create the standard atlas and also to serve for benchmark tests. We found that the AT+ADT outperformed other two methods, though the accuracy of each method varied depending on the layer. This novel image standardization technique would be applicable to optical recording such as voltage-sensitive dye imaging, and will enable statistical evaluations of neural activation across multiple samples.

Show MeSH

Related in: MedlinePlus

Comparisons of the mean error ratio estimates for type 1 error (A) and type 2 error (B).The significance of the differences were evaluated by two-way ANOVA with Bonferroni's post hoc test, and results with significance level <0.05 (Bonferroni-corrected <0.05/3 = 0.017) were marked with “*”.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3818318&req=5

pone-0076415-g006: Comparisons of the mean error ratio estimates for type 1 error (A) and type 2 error (B).The significance of the differences were evaluated by two-way ANOVA with Bonferroni's post hoc test, and results with significance level <0.05 (Bonferroni-corrected <0.05/3 = 0.017) were marked with “*”.

Mentions: Pixels, which were categorized in each lamina in the transformed image but were not categorized in the left out image, induce type 1 error. In the opposite case, pixels induce type 2 error. Then, type 1 and 2 error ratios can be defined as and respectively, where and (: the pixel value at a coordinate in the -th lamina of the -th image) are total numbers of pixels which satisfy conditions and , respectively. Repeating this procedure for all images (k = 1,..n) yields mean error ratio estimates and for type 1 and 2 error, respectively. Two-way analysis of variance (ANOVA) was performed to assess mean error ratio differences due to the methods (AT, ADT and AT+ADT) and laminas for type 1 and 2 error. There were significant main effect of the methods (type 1: F(2,1230) = 26.77, <0.01, type 2: F(2,1230) = 13.14,<0.01) and laminas (type 1: F(9,1230) = 77.96, <0.01, type 2: F(9,1230) = 119.39, <0.01) and also interaction (type 1: F(18,1230) = 7.72, <0.01, type 2: F(18,1230) = 5.33, <0.01). It indicates that the performance of the methods is different depending on the laminas. Subsequently the difference of mean error ratios corresponding to the method for each lamina, which satisfy the significance level, were evaluated by the paired -test with the Bonferroni correction (Figure 6).


Standardization of size, shape and internal structure of spinal cord images: comparison of three transformation methods.

Fujiki Y, Yokota S, Okada Y, Oku Y, Tamura Y, Ishiguro M, Miwakeichi F - PLoS ONE (2013)

Comparisons of the mean error ratio estimates for type 1 error (A) and type 2 error (B).The significance of the differences were evaluated by two-way ANOVA with Bonferroni's post hoc test, and results with significance level <0.05 (Bonferroni-corrected <0.05/3 = 0.017) were marked with “*”.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3818318&req=5

pone-0076415-g006: Comparisons of the mean error ratio estimates for type 1 error (A) and type 2 error (B).The significance of the differences were evaluated by two-way ANOVA with Bonferroni's post hoc test, and results with significance level <0.05 (Bonferroni-corrected <0.05/3 = 0.017) were marked with “*”.
Mentions: Pixels, which were categorized in each lamina in the transformed image but were not categorized in the left out image, induce type 1 error. In the opposite case, pixels induce type 2 error. Then, type 1 and 2 error ratios can be defined as and respectively, where and (: the pixel value at a coordinate in the -th lamina of the -th image) are total numbers of pixels which satisfy conditions and , respectively. Repeating this procedure for all images (k = 1,..n) yields mean error ratio estimates and for type 1 and 2 error, respectively. Two-way analysis of variance (ANOVA) was performed to assess mean error ratio differences due to the methods (AT, ADT and AT+ADT) and laminas for type 1 and 2 error. There were significant main effect of the methods (type 1: F(2,1230) = 26.77, <0.01, type 2: F(2,1230) = 13.14,<0.01) and laminas (type 1: F(9,1230) = 77.96, <0.01, type 2: F(9,1230) = 119.39, <0.01) and also interaction (type 1: F(18,1230) = 7.72, <0.01, type 2: F(18,1230) = 5.33, <0.01). It indicates that the performance of the methods is different depending on the laminas. Subsequently the difference of mean error ratios corresponding to the method for each lamina, which satisfy the significance level, were evaluated by the paired -test with the Bonferroni correction (Figure 6).

Bottom Line: To solve this problem, we attempted to standardize transversely sectioned spinal cord images focusing on the laminar structure in the gray matter.In this study, we used neuron-specific marker (NeuN)-stained histological images of transversely sectioned cervical spinal cord slices (21 images obtained from 4 rats) to create the standard atlas and also to serve for benchmark tests.This novel image standardization technique would be applicable to optical recording such as voltage-sensitive dye imaging, and will enable statistical evaluations of neural activation across multiple samples.

View Article: PubMed Central - PubMed

Affiliation: Department of Statistical Science, School of Multidisciplinary Sciences, The Graduate University for Advanced Studies, Tachikawa, Tokyo, Japan.

ABSTRACT
Functional fluorescence imaging has been widely applied to analyze spatio-temporal patterns of cellular dynamics in the brain and spinal cord. However, it is difficult to integrate spatial information obtained from imaging data in specific regions of interest across multiple samples, due to large variability in the size, shape and internal structure of samples. To solve this problem, we attempted to standardize transversely sectioned spinal cord images focusing on the laminar structure in the gray matter. We employed three standardization methods, the affine transformation (AT), the angle-dependent transformation (ADT) and the combination of these two methods (AT+ADT). The ADT is a novel non-linear transformation method developed in this study to adjust an individual image onto the template image in the polar coordinate system. We next compared the accuracy of these three standardization methods. We evaluated two indices, i.e., the spatial distribution of pixels that are not categorized to any layer and the error ratio by the leave-one-out cross validation method. In this study, we used neuron-specific marker (NeuN)-stained histological images of transversely sectioned cervical spinal cord slices (21 images obtained from 4 rats) to create the standard atlas and also to serve for benchmark tests. We found that the AT+ADT outperformed other two methods, though the accuracy of each method varied depending on the layer. This novel image standardization technique would be applicable to optical recording such as voltage-sensitive dye imaging, and will enable statistical evaluations of neural activation across multiple samples.

Show MeSH
Related in: MedlinePlus