Table 2 Descriptions of geometric, intensity and texture features

The number of pixels on the contour as well as the pixels enclosed by the contour.

The number of pixels on the nuclear contour.

C=4π A/P ², indicating the roundness of the nucleus.

1−b/a (a and b are the length and width of minimum enclosing rectangle, shown in Fig. 7), measuring how much the nucleus deviates from being circular.

A/A _c (A _c is the nuclear convex area measured by counting the number of pixels in the convex hull, as shown in Fig. 7).

The maximum of curvatures (The curvature at each boundary point is calculated by fitting a circle to that boundary point and the two points 10 boundary points away from it.).

The minimum of curvatures.

The standard deviation of curvatures.

The average absolute value of curvatures.

Mean intensity of all pixels in the nuclei.

Variant of all pixels’ intensity in the nuclei.

\(\frac {1}{N-1}\sum _{i=1}^{N}\left (\frac {x_{i}-\bar x}{\sigma }\right)^{3}\) (N is the number of pixels in the nucleus). The negative or positive skewness means that most of the pixel values are concentrated at the right or left side of the histogram, respectively.

\(\frac {1}{N-1}\sum _{i=1}^{N}\left (\frac {x_{i}-\bar x}{\sigma }\right)^{4}\), describing whether the distribution is platykurtic or leptokurtic.

\(\sum _{i,j}\left |i-j\right |^{2}p\left (i,j\right)\), measuring the intensity contrast between a pixel and its neighbor over the whole nucleus.

\(\sum _{i,j}\frac {(i-\mu _{i})(j-\mu _{j})p(i,j)}{\sigma _{i}\sigma _{j}}\), measuring the dependencies between the nucleus image pixels.

\(\sum _{i,j} p(i,j)^{2}\), measuring the orderliness of texture. When the image is proficient orderly, energy value is high.

\(\sum _{i,j}\frac {p(i,j)}{1+\left |i-j\right |}\), measuring the closeness of the distribution of elements in GLCM to its diagonal.