Automated morphological classification of lung cancer subtypes using H&E tissue images

Abstract

Patient-targeted therapies have recently been highlighted as important. An important development in the treatment of metastatic non-small cell lung cancer (NSCLC) has been the tailoring of therapy on the basis of histology. A pathology diagnosis of “non-specified NSCLC” is no longer routinely acceptable; an effective approach for classification of adenocarcinoma (AC) and squamous carcinoma (SC) histotypes is needed for optimizing therapy. In this study, we present a robust and objective automatic computer vision system for real-time classification of AC and SC based on the morphological tissue patterns of hematoxylin and eosin (H&E) staining images to assist medical experts in the diagnosis of lung cancer. Various original and extended densitometric and Haralick’s texture features are used to extract image features, and a boosting algorithm is utilized to train the classifier, together with alternative decision tree as the base learner. For evaluation, two types of data with 653 tissue samples were tested, including 369 samples from tissue microarray data set and 284 samples from full-face tissue sections. Regarding the data distribution, 45 % are AC samples (288) and 55 % are SC samples (365), which is considerably well balanced for each class. Using tenfold cross-validation, the technique achieved high accuracy of $92.41~\%$ on tissue microarray cores and $95.42~\%$ on full tissue sections. We also found that the two boosting algorithms (cw-Boost and AdaBoost.M1) perform consistently well in comparison with other popularly adopted machine learning methods, including support vector machine, neural network and decision tree. This approach offers a robust, objective and rapid procedure for optimized patient-targeted therapy.

Appendix

In the calculation of Haralick features, four matrices are needed to describe different orientations ($0^\circ , 45^\circ , 90^\circ , 135^\circ$), and each matrix is a symmetric matrix with the dimensionality of $N_\mathrm{{g}} \times N_\mathrm{{g}}$ where $N_\mathrm{{g}}$ is the number of possible gray levels for a particular image. The co-occurrence matrices can be represented by a three-dimensional data structure; the third dimension is $d$, which varies depending on the dimensions of the original input image. The statistical properties of co-occurrence matrix can be formulated as follows. $p(i,j)$ is defined as the $(i,j)$th entry in a normalized single-channel/tone spatial-dependence matrix.

$$\begin{aligned} p(i,j)=P(i,j)/R \end{aligned}$$

(3)

where $R=\sum _{i=1}^{N_\mathrm{{g}}}\sum _{j=1}^{N_\mathrm{{g}}}P(i,j)$.

$p_x(i)$: the i-th entry in the marginal probability matrix obtained by summing the rows of $p(i,j)$.

$$\begin{aligned} p_x(i)=\sum _{j=1}^{N_\mathrm{{g}}}p(i,j) \end{aligned}$$

(4)

The 11 Haralick features are defined as follows.

$$\begin{aligned} f_1&= \sum _{i=1}^{N_\mathrm{{g}}}\sum _{j=1}^{N_\mathrm{{g}}}(p(i,j))^2 \end{aligned}$$

(5)

$$\begin{aligned} f_2&= \sum _{k=0}^{N_\mathrm{{g}}-1}k^2 p_{x-y}(k) \end{aligned}$$

(6)

$$\begin{aligned} f_3&= \frac{\sum _{i=1}^{N_\mathrm{{g}}}\sum _{j=1}^{N_\mathrm{{g}}}(ij)p(i,j)-\upmu _x\upmu _y}{\sigma _x\sigma _y} \end{aligned}$$

(7)

where $\upmu _x, \upmu _y,\sigma _x, \sigma _y$ are the means and standard deviations of $p_x$ and $p_y$.

$$\begin{aligned} f_4&= \sum _{i=1}^{N_\mathrm{{g}}}\sum _{j=1}^{N_\mathrm{{g}}}(i-\upmu )^2p(i,j) \end{aligned}$$

(8)

$$\begin{aligned} f_5&= \sum _{i=1}^{N_\mathrm{{g}}}\sum _{j=1}^{N_\mathrm{{g}}}\frac{1}{1+(i-j)^2}p(i,j) \end{aligned}$$

(9)

$$\begin{aligned} f_6&= \sum _{i=2}^{2N_\mathrm{{g}}}ip_{x+y}(i) \end{aligned}$$

(10)

$$\begin{aligned} f_7&= \sum _{i=2}^{2N_\mathrm{{g}}}(i-f_8)^2p_{x+y}(i) \end{aligned}$$

(11)

$$\begin{aligned} f_8&= -\sum _{i=2}^{2N_\mathrm{{g}}}p_{x+y}(i)\log (p_{x+y}(i)) \end{aligned}$$

(12)

$$\begin{aligned} f_9&= -\sum _{i=1}^{N_\mathrm{{g}}}\sum _{j=1}^{N_\mathrm{{g}}}p(i,j)\log (p(i,j)) \end{aligned}$$

(13)

$$\begin{aligned} f_{10}&= \sigma _{p_{x-y}} \end{aligned}$$

(14)

$$\begin{aligned} f_{11}&= -\sum _{i=0}^{N_\mathrm{{g}}-1}p_{x-y}(i)\log (p_{x+y}(i)) \end{aligned}$$

(15)