Nodal-based radiomics analysis for identifying cervical lymph node metastasis at levels I and II in patients with oral squamous cell carcinoma using contrast-enhanced computed tomography

Abstract

Objective

Discriminating metastatic from benign cervical lymph nodes (LNs) in oral squamous cell carcinoma (OSCC) patients using pretreatment computed tomography (CT) has been controversial. This study aimed to investigate whether CT-based texture analysis with machine learning can accurately identify cervical lymph node metastasis in OSCC patients.

Methods

Twenty-three patients (with 201 cervical LNs [150 benign, 51 metastatic] at levels I–V) who underwent preoperative contrast-enhanced CT and subsequent cervical neck dissection were enrolled. Histopathologically proven LNs were randomly divided into the training cohort (70%; n = 141, at levels I–V) and validation cohort (30%; n = 60, at level I/II). Twenty-five texture features and the nodal size of targeted LNs were analyzed on the CT scans. The nodal-based sensitivities, specificities, diagnostic accuracy rates, and the area under the curves (AUCs) of the receiver operating characteristic curves of combined features using a support vector machine (SVM) at levels I/II, I, and II were evaluated and compared with two radiologists and a dentist (readers).

Results

In the validation cohort, the AUCs (0.820 at level I/II, 0.820 at level I, and 0.930 at level II, respectively) of the radiomics approach were superior to three readers (0.798–0.816, 0.773–0.798, and 0.825–0.865, respectively). The best models were more specific at levels I/II and I and accurate at each level than each of the readers (p < .05).

Conclusions

Machine learning–based analysis with contrast-enhanced CT can be used to noninvasively differentiate between benign and metastatic cervical LNs in OSCC patients.

Key Points

• The best algorithm in the validation cohort can noninvasively differentiate between benign and metastatic cervical LNs at levels I/II, I, and II.

• The AUCs of the model at each level were superior to those of multireaders.

• Significant differences in the specificities at level I/II and I and diagnostic accuracy rates at each level between the model and multireaders were found.

Appendix 1. The list of Texture features

Histogram features

Histogram features describe simple statistics that are associated with pixel values in images. Spatial patterns of pixel values are not included. Skewness, kurtosis, entropy, and energy were calculated.

Gray-level co-occurrence matrix (GLCM) features

The GLCM is defined as the distribution of co-occurring pixel values that are calculated from 4 directions in 2-dimensional (2D) space or 13 directions in 3-dimensional (3D) space:

$$ \mathrm{Homogeneity}=\sum \limits_{i,j}\frac{p\left(i,j\right)}{1+\mid i-j\mid } $$

$$ \mathrm{Energy}=\sum \limits_{i,j}p{\left(i,j\right)}^2 $$

$$ \mathrm{Contrast}=\sum \limits_{i,j}{\left|i-j\right|}^2p\left(i,j\right) $$

$$ \mathrm{Correlation}=\sum \limits_{i,j}\frac{\left(i-{\upmu}_i\right)\left(j-{\upmu}_j\right)p\left(i,\kern0.5em \mathrm{j}\right)}{\upsigma_i{\upsigma}_j} $$

$$ \mathrm{Entropy}=\sum \limits_{i,j}\log \left(p\left(i,j\right)\right)p\left(i,j\right) $$

$$ \mathrm{Dissimilarity}=\sum \limits_{i,j}\mid i-j\mid p\left(i,j\right) $$

Where p(i,j) represents (i,j) value of the GLCM.

Gray-level run length matrix (GLRLM) features

The GLRLM is defined as the number of consecutive pixels of the same gray-level value for 4 directions in 2D space or 13 directions in 3D space:

Short run emphasis $\left(\mathrm{SRE}\right)=\frac{1}{n_r}\sum \limits_{i,j}\frac{p\left(i,j\right)}{j^2}$

Low gray-level run emphasis $\left(\mathrm{LGRE}\right)=\frac{1}{n_r}\sum \limits_{i,j}\frac{p\left(i,j\right)}{i^2}$

High gray-level run emphasis $\left(\mathrm{HGRE}\right)=\frac{1}{n_r}\sum \limits_{i,j}p\left(i,j\right){i}^2$

Short run low gray-level emphasis $\left(\mathrm{SRLGE}\right)=\frac{1}{n_r}\sum \limits_{i,j}\frac{p\left(i,j\right)}{i^2{j}^2}$

Short run high gray-level emphasis $\left(\mathrm{SRHGE}\right)=\frac{1}{n_r}\sum \limits_{i,j}\frac{p\left(i,j\right){i}^2}{j^2}$

Run percentage $\left(\mathrm{RP}\right)=\frac{n_r}{\sum \limits_{i,j}\left( jp\left(i,j\right)\right)}$Where n_r corresponds to the number of homogenous runs.

Neighborhood gray-level different matrix (NGLDM)

The NGLDM is defined as the difference in gray-levels between adjacent voxels of 8 in 2D space and 26 in 3D space:

$$ \mathrm{Coarseness}=\frac{1}{\sum \limits_ip\left(i,1\right)p\left(i,2\right)} $$

$$ \mathrm{Contrast}=\left[\sum \limits_{i,j}p\left(i,1\right)p\left(j,1\right){\left(i-j\right)}^2\right]\frac{\sum \limits_ip\left(i,2\right)}{EG\left(G-1\right)} $$

Where E is the number of voxels in VOI and G is the number of gray-levels.

Gray-level zone length matrix (GLZLM)

GLZLM is defined as the number of homogenous zones of the same gray-level value in 2D or 3D space:

Short-zone emphasis $\left(\mathrm{SZE}\right)=\frac{1}{n_r}\sum \limits_{i,j}\frac{p\left(i,j\right)}{j^2}$

Low gray-level zone emphasis $\left(\mathrm{LGZE}\right)=\frac{1}{n_r}\sum \limits_{i,j}\frac{p\left(i,j\right)}{i^2}$

High gray-level zone emphasis $\left(\mathrm{HGZE}\right)=\frac{1}{n_r}\sum \limits_{i,j}p\left(i,j\right){i}^2$

Short-zone low gray-level emphasis $\left(\mathrm{SZLGE}\right)=\frac{1}{n_r}\sum \limits_{i,j}\frac{p\left(i,j\right)}{i^2{j}^2}$

Short-zone high gray-level emphasis $\left(\mathrm{SZHGE}\right)=\frac{1}{n_r}\sum \limits_{i,j}\frac{p\left(i,j\right){i}^2}{j^2}$

Zone length non-uniformity $\left(\mathrm{ZLNU}\right)=\frac{1}{n_r}\sum \limits_j{\left(\sum \limits_ip\left(i,j\right)\right)}^2$

Zone percentage $\left(\mathrm{ZP}\right)=\frac{n_r}{\sum \limits_{i,j}\left( jp\left(i,j\right)\right)}$where n_r corresponds to the number of homogenous zones.