Table 1. Comparison of information gain ratio of top features from expert-guided and CellProfiler categories.
| Feature | IG | Description |
|---|---|---|
| Expert-guided features | ||
| Total clusters | (0.2014) | The number of nuclei clusters computed in an image tile |
| Average clusters | (0.2014) | The average number of nuclei clusters computed in a 32x32 window |
| Red count | (0.1572) | The number of background pixels in an image tile |
| Red percentage | (0.1572) | The percentage of red pixels among foreground and background pixels |
| Blue count | (0.1523) | The number of foreground pixels in an image tile |
| Blue percentage | (0.1523) | The percentage of blue pixels among foreground and background pixels |
| Area | (0.0874) | The average area of identified nuclei clusters in an image tile |
| Circularity | (0.0425) | The average circularity of identified nuclei clusters in an image tile |
| CellProfiler features | ||
| Texture Gabor | (1.8406) | The Gabor feature measures striped texture in an object |
| Texture sum Entropy | (1.8406) | The sum of entropy of all pixel values in GLCM |
| Texture sum variance | (1.8406) | The variance of entropy values in GLCM |
| Texture variance | (1.8406) | The variance of GLCM elements in the matrix |
| Texture sum average | (1.8406) | Sum of product of entries with probability of co-occurrence of values in GLCM |
| Texture correlation | (1.8406) | Gray level linear independence between pixels |
| Texture angular second moment | (1.8406) | Measure of homogeneity in an image |
IG—Information Gain
GLCM—Gray Level Co-occurrence Matrix