Novel image markers for non-small cell lung cancer classification and survival prediction

BMC Bioinformatics

Table 1 The image features and their descriptions

Name	Description
area1-6	Cell area feature
axis1-6	Major-minor axis ratio
cir1-6	Cell circularity feature
peri1-6	Contour perimeter feature
solidity1-6	Contour solidity feature
mean1-6	Cell intensity mean feature
std1-6	Cell intensity stand deviation feature
kurt1-6	Cell intensity Kurtosis
entr1-6	Cell intensity entropy
energy1-6	Cell intensity energy
contrast1-6	Cell intensity contrast
corr1-6	Cell intensity correlation
engy1-6	Cell intensity energy from co-occurrence matrix
homo1-6	Cell intensity homogeneity
skew1-6	Cell intensity skewness
tfcm1-24	Texture feature coding method (TFCM)
csac1-24	Center symmetric auto correlation (CSAC) feature
lbp1-24	Local binary pattern (LBP) feature
t1-4	Texton histogram feature

The 1-6 in each image feature represent the mean, median, variance and three frequency values of the histogram for each intensity and geometric feature, respectively. Csac, tfcm, lbp and texton histogram features are high dimensional feature vectors, therefore we calculate their moment statistics to reduce the dimensionality. In total we have extracted 166 geometric, pixel intensity, and image texture feature variables for each patient. All variables are normalized before further classification and survival analysis.

ISSN: 1471-2105