Applications of SLFs
| Line 11: | Line 11: | ||
=== Cell Classification by SLFs === | === Cell Classification by SLFs === | ||
| − | We used the SLFs to quantify the structures and associations among five seperate channels of the 3D image for brain tissue [ | + | We used the SLFs to quantify the structures and associations among five seperate channels of the 3D image for brain tissue [1]. The five channels are:(i)cell nuclei (CyQuant); (ii)astrocytes (GFAP); (iii)neurons (Nissl); (iv)blood vessels (EBA); and |
| − | (v)microglia (Iba1). The SLFs computation are based on the nuclei(NUC),Nissl and Iba1 channels(Figure.5). After the segmentation on Nuclei channel, there are 1019 cells which are of four types, namely the Astroaytes(386), Microglia(85), Neuron(365) and the Endothelial(183). Four SLF feature sets are calculated: (i)14 intrinsic morphological features from NUC; (ii)42 morphological features from NUC and Nissl(14 intrinsic and 28 associative); (iii)42 morphological features from NUC and Iba1(14 intrinsic and 28 associative); (iv)the combination of feature set 2 and 3(14 intrinsic and 56 associative). In[ | + | (v)microglia (Iba1). The SLFs computation are based on the nuclei(NUC),Nissl and Iba1 channels(Figure.5). After the segmentation on Nuclei channel, there are 1019 cells which are of four types, namely the Astroaytes(386), Microglia(85), Neuron(365) and the Endothelial(183). Four SLF feature sets are calculated: (i)14 intrinsic morphological features from NUC; (ii)42 morphological features from NUC and Nissl(14 intrinsic and 28 associative); (iii)42 morphological features from NUC and Iba1(14 intrinsic and 28 associative); (iv)the combination of feature set 2 and 3(14 intrinsic and 56 associative). In[1], in order to compute associative features, 3D compartments are defined for each segmented nuclei and are then used to compute the amount of intranulcear and cytoplasmic(approximated as a 5-voxel region surrounding the nuclear compartment) of Iba1 and Nissl signals. Here we just expand the segmented region in NUC channel by 5-voxel and then compute associative on both NUC and Nissl/Iba1 channels. We didn't use the SLF texture and edge features since the single cell image is very small and textureless, while SLF textures are designed for cell images of larger size. |
The obtained feature sets are then fed to Support Vector Machine(SVM) for cell classification. The classification results from 5-fold corss-validation are shown in Table.1. | The obtained feature sets are then fed to Support Vector Machine(SVM) for cell classification. The classification results from 5-fold corss-validation are shown in Table.1. | ||
[[Image:Channels.PNG|thumb|600px| Figure.5 Three Channels used for computing SLFs|center]] | [[Image:Channels.PNG|thumb|600px| Figure.5 Three Channels used for computing SLFs|center]] | ||
[[Image:CellClassification.PNG|thumb|783px|Table.1 Cell Classification Results|center]] | [[Image:CellClassification.PNG|thumb|783px|Table.1 Cell Classification Results|center]] | ||
| + | |||
| + | == Reference == | ||
| + | * [1] Christopher S Bjornsson, Gang Lin, Yousef Al-Kofahi, Arunachalam Narayanaswamy, Karen L Smith, William Shain, Badrinath Roysam (2008). “Associative image analysis: a method for automated quantification of 3D multi-parameter images of brain tissue,” Journal of Neuroscience Methods, 170(1):165-178. | ||
Latest revision as of 21:02, 25 April 2009
Contents |
Applications of SLFs
Cell Analysis by SLFs
We applied the code to compute SLFs on twelve 3D images of two types, one is the wild type in which cells are in their normal state, and the other one is mutated type where cells have been drug treated. The obtained SLFs are then used for further analysis to observe the changes in cells. Each 3D image has one Actin channel and one Nuclei channel(Figure.3). Cell segmentation is based on the Nuclei channel, associative features are computed on both channel. For the first six images belonging to wild type, there are 91 cells, and for the six images of mutated type, there are 111 cells. Therefore, there are totally 202 feature vectors, each one with 56 dimensions.
In order to observe the difference between the cells in normal state, and the cells treated with drug. The obtained 202 cell feature vectors are used for principal component analysis (PCA). The scatter plot for the first and second principal components are displayed by Figure.4, from which it can be noticed that the features of DDR and WT are of slightly different scatter patterns.
Cell Classification by SLFs
We used the SLFs to quantify the structures and associations among five seperate channels of the 3D image for brain tissue [1]. The five channels are:(i)cell nuclei (CyQuant); (ii)astrocytes (GFAP); (iii)neurons (Nissl); (iv)blood vessels (EBA); and (v)microglia (Iba1). The SLFs computation are based on the nuclei(NUC),Nissl and Iba1 channels(Figure.5). After the segmentation on Nuclei channel, there are 1019 cells which are of four types, namely the Astroaytes(386), Microglia(85), Neuron(365) and the Endothelial(183). Four SLF feature sets are calculated: (i)14 intrinsic morphological features from NUC; (ii)42 morphological features from NUC and Nissl(14 intrinsic and 28 associative); (iii)42 morphological features from NUC and Iba1(14 intrinsic and 28 associative); (iv)the combination of feature set 2 and 3(14 intrinsic and 56 associative). In[1], in order to compute associative features, 3D compartments are defined for each segmented nuclei and are then used to compute the amount of intranulcear and cytoplasmic(approximated as a 5-voxel region surrounding the nuclear compartment) of Iba1 and Nissl signals. Here we just expand the segmented region in NUC channel by 5-voxel and then compute associative on both NUC and Nissl/Iba1 channels. We didn't use the SLF texture and edge features since the single cell image is very small and textureless, while SLF textures are designed for cell images of larger size. The obtained feature sets are then fed to Support Vector Machine(SVM) for cell classification. The classification results from 5-fold corss-validation are shown in Table.1.
Reference
- [1] Christopher S Bjornsson, Gang Lin, Yousef Al-Kofahi, Arunachalam Narayanaswamy, Karen L Smith, William Shain, Badrinath Roysam (2008). “Associative image analysis: a method for automated quantification of 3D multi-parameter images of brain tissue,” Journal of Neuroscience Methods, 170(1):165-178.
