Murphy Lab

The subcellular location pattern represented by a collection of similar images can be described according to a variety of numeric features. These features have been grouped into more than 10 sets known as SLF sets (Subcellular Location Feature sets). In order to demonstrate the usefulness of these features they have been used to train a classifier, which was able to correctly recognize an average of 92% (2D) and 98% (3D) of previously unseen cells showing one of ten patterns. The SLF sets are described as follows:

2D SLF

• SLF1 - A set of 16 features that are calculated from the processed protein image only. [Boland & Murphy, 2001]
• SLF2 - A set of 22 features which includes SLF1 and a set of six features calculated using both the processed protein image and the corresponding DNA image. [Boland & Murphy, 2001]
• SLF3 - Defined as the combination of SLF1 with Zernike and Haralick features giving a total of 78 features that can be calculated from a protein image only. [Boland & Murphy, 2001]
• SLF4 - Defined as the combination of SLF2 with Zernike and Haralick features giving a total of 84 features that are derived from a protein image and a corresponding DNA image. [Boland & Murphy, 2001]
• SLF5 - A subset of features selected from the SLF4 set via stepwise discriminant analysis (SDA). A total of 37 features are used that are derived from a protein image and a corresponding DNA image. The list of features can be see here. [Boland & Murphy, 2001]
• SLF6 - Defined as the combination of SLF1 with Zernike features giving a total of 65 features that can be calculated from a protein image only. [Roques & Murphy, 2002]
• SLF7 - Normalized features designed to enable description and classification of images from heterogeneous sources. [Murphy, Velliste & Porreca, 2002]
• SLF8 - Subset chosen by SDA from SLF7. [Murphy, Velliste & Porreca, 2002]
• SLF12 - Subset chosen by SDA from SLF7. It is the smallest group to achieve over 80% accuracy. [Huang, Velliste & Murphy, 2002]
• SLF13 - Subset chosen by SDA from SLF7 plus 6 DNA features. [Murphy, Velliste & Porreca, 2003]
• SLF15 - Subset chosen by SDA from 174 features including 84 SLF7, 60 Gabor texture features and 30 wavelet features. [Huang & Murphy, 2004b]
• SLF16 - Subset chosen by SDA from 180 features including 84 SLF7, 60 Gabor texture features, 30 wavelet features and 6 DNA features. [Huang & Murphy, 2004b]
• SLF21 - Subset of 26 rotationally invariant SLF7 features that are also independent of the number of cells in a field. Consists of 13 morphological and 13 texture features. [Huang & Murphy, 2004a]

• SLF9 - A set of 28 features for describing 3D images derived from SLF2. [Velliste & Murphy, 2002]
• SLF10 - Nine features selected by SDA from SLF9. [Huang & Murphy, 2004b]
• SLF11 - Unselected set containing morphological, edge and texture features. [Chen, Velliste, Weinstein, Jarvik & Murphy, 2003]
• SLF14 - Features in SLF9 that do not require a parallel DNA image. [Huang & Murphy, 2004b]
• SLF17 - 7 Features selected by SDA from SLF11. The texture features are calculated at 0.4 microns and 256 gray levels. [Chen & Murphy, 2004]
• SLF18 - 34 Features selected by SDA from SLF11. The texture features are calculated at 0.5 microns and 64 gray levels. [Chen & Murphy, 2005]
• SLF19 - Defined as the combination of SLF11 and 14 DNA features. The texture features are calculated at 0.09 microns and 256 gray levels. [Nair, Schaub, Huang, Chen, Murphy, Griffith, Geuze, & Rohrer, 2005]
• SLF20 - 52 features selected by SDA from SLF19. [Nair, Schaub, Huang, Chen, Murphy, Griffith, Geuze, & Rohrer, 2005]
• SLF22 - 14 Morphological and 2 edge features calculated after downsampling (if necessary) to a resolution of 0.11 x 0.11 x 0.5 microns. 26 texture features calculated after downsampling to 0.5 microns and 64 gray levels.
• SLF23 - 14 Morphological and 2 edge features calculated after downsampling (if necessary) to a resolution of 0.05 x 0.05 x 0.2 microns. 26 texture features calculated after downsampling to 0.4 microns and 256 gray levels.
• SLF24 - SLF23 plus 14 DNA features calculated after downsampling image to a resolution of 0.05 x 0.05 x 0.2 microns.

Last Updated: 26 Jun 2005

Copyright © 1996-2005 by the Murphy Lab