Impact of radiogenomics in esophageal cancer on clinical outcomes: A pilot study

Table 1 Characteristics of included patients

Clinical characteristic	Value
Age (mean ± SD)	56.8 ± 8.65
Sex, n (%)
Male	12 (20)
Female	3 (80)
Histologic diagnosis, n (%)
EAC	2 (13.3)
ESCC	13 (86.7)
Cancer stage, n (%)
I-II	4 (26.7)
III-IV	11 (73.3)
TNM staging
Primary tumor (T)
T1	3 (20)
T2	2 (13.3)
T3	8 (53.4)
T4	2 (13.3)
Regional lymph nodes (N)
N0	4 (26.7)
N1	6 (40)
N2	3 (20)
N3	2 (13.3)
Distant metastases (M)
M0	12 (80)
M1	3 (20)
Alcohol history, n (%)
Yes	8 (53)
No	7 (47)
Smoking history, n (%)
Lifelong non-smoker	1 (6.6)
Current smoker	7 (46.6)
Current reformed smoker	2 (13.3)
Current reformed smoker for > 15 yr	1 (6.6)
NA	4 (26.6)
Age at starting smoking
Under 18 yr old	6 (40)
Over 18 yr old	3 (20)
NA	6 (40)
Pack-year smoked
Under 10	6 (40)
Over 10	3 (20)
NA	6 (40)
BMI (mean ± SD)	20.8 ± 3.55

BMI: Body mass index; EAC: Esophagus adenocarcinoma; ESCC: Esophagus squamous cell carcinoma; TNM: tumor, node, and metastasis classification (classification of malignant tumors).

Table 2 Selected radiomic features after the three feature selection steps for binary outcomes

Selected radiomic features, stage	Stage I-II, mean ± SD	Stage III-IV, mean ± SD	P value
Wavelet-HLH first order entropy	0.62 ± 0.13	0.21 ± 0.31	0.031
Wavelet-LHL GLDM high gray level emphasis	0.59 ± 0.32	0.21 ± 0.21	0.044
Wavelet-HLL GLDM high gray level run emphasis	0.67 ± 0.37	0.2 ± 0.18	0.019
Wavelet-HLL GLDM gray level variance	0.74 ± 0.29	0.22 ± 0.29	0.007
Wavelet-HLL GLDM cluster prominence	0.59 ± 0.36	0.14 ± 0.3	0.019
Wavelet-HLL GLDM inverse variance	0.27 ± 0.3	0.68 ± 0.23	0.028
Wavelet-HLL GLDM maximum probability	0.14 ± 0.17	0.58 ± 0.32	0.042
Wavelet-HLL GLDM dependence entropy	0.78 ± 0.18	0.42 ± 0.25	0.028
Wavelet-HLL GLDM dependence variance	0.2 ± 0.21	0.48 ± 0.26	0.044
Wavelet-HLL GLSZM size zone non uniformity	0.69 ± 0.39	0.15 ± 0.15	0.028
Wavelet-HLL GLSZM small area low gray level emphasis	0.02 ± 0.03	0.24 ± 0.33	0.041
Wavelet-HLL first order 90 percentile	0.55 ± 0.39	0.17 ± 0.16	0.029
Wavelet-HHH GLRLM gray level non uniformity normalized	0.39 ± 0.27	0.87 ± 0.19	0.007
Wavelet-HHH GLCM joint entropy	0.68 ± 0.22	0.2 ± 0.27	0.021
Wavelet-HHH GLCM Contrast	0.78 ± 0.19	0.45 ± 0.25	0.021
Wavelet-HHH GLSZM gray level variance	0.91 ± 0.11	0.3 ± 0.39	0.021
Wavelet-HHH GLSZM zone percentage	0.83 ± 0.13	0.35 ± 0.34	0.029
Wavelet-HHH first order 10 percentile	0.11 ± 0.08	0.46 ± 0.35	0.029
Wavelet-HHL NGTDM strength	0.66 ± 0.3	0.28 ± 0.3	0.044
Wavelet-HHL GLCM cluster shade	0.51 ± 0.49	0.04 ± 0.03	0.007
Wavelet-HHL first order Entropy	0.68 ± 0.33	0.27 ± 0.21	0.029
Wavelet-HHL first order mean absolute deviation	0.66 ± 0.35	0.25 ± 0.18	0.029
Wavelet-LLH NGTDM complexity	0.23 ± 0.13	0.13 ± 0.29	0.044
Wavelet-LLH GLDM high gray level emphasis	0.34 ± 0.23	0.14 ± 0.29	0.044
Wavelet-LLL GLDM small dependence high gray level emphasis	0.67 ± 0.32	0.25 ± 0.22	0.029
Selected radiomic features (alcohol history)	No alcohol history (mean ± SD)	Alcohol history (mean ± SD)	P value
Original first order kurtosis	0.37 ± 0.32	0.12 ± 0.12	0.037
Wavelet-LHH first order interquartile range	0.24 ± 0.24	0.5 ± 0.25	0.049
Wavelet-LHH first order mean	0.94 ± 0.04	0.75 ± 0.31	0.013
Wavelet-HHL first order median	0.62 ± 0.2	0.37 ± 0.19	0.026
Wavelet-LLH first order uniformity	0.68 ± 0.27	0.41 ± 0.26	0.037

The first column includes the two binary clinical outcomes investigated in this study (stage, alcohol history). GLCM: Gray level co-occurrence matrix; GLDM: Gray level dependence matrix; GLRLM: Gray level run length matrix; GLSZM: Gray level size zone matrix; H: High-pass filter; L: Low-pass filter; NGTDM: Neighboring gray tone difference matrix; SD: Standard deviation.

Table 3 Selected radiomic features after the three feature selection steps for body mass index

Selected radiomic features	mean ± SD	ρ	P value
Wavelet-HLH GLCM IMC2	0.46 ± 0.31	0.6	0.017
Wavelet-HLH GLCM MCC	0.33 ± 0.27	0.67	0.006
Wavelet-HLH first order median	0.59 ± 0.28	-0.56	0.029
Wavelet-HLH first order entropy	0.32 ± 0.33	0.63	0.013
Wavelet-HLH first order root mean squared	0.44 ± 0.31	0.71	0.003
Wavelet-HLH first order skewness	0.5 ± 0.33	0.52	0.049
Wavelet-HLH first order mean absolute deviation	0.54 ± 0.29	0.62	0.013
Wavelet-LHL GLCM IDN	0.77 ± 0.29	0.58	0.023
Wavelet-LHL GLDM high gray level emphasis	0.31 ± 0.29	0.54	0.039
Wavelet-HLL GLRLM high gray level run emphasis	0.32 ± 0.32	0.54	0.038
Wavelet-HLL GLRLM gray level variance	0.36 ± 0.37	0.62	0.014
Wavelet-HLL GLCM cluster prominence	0.26 ± 0.36	0.6	0.018
Wavelet-HLL GLCM inverse variance	0.57 ± 0.3	-0.58	0.022
Wavelet-HLL GLCM maximum probability	0.47 ± 0.35	-0.53	0.043
Wavelet-HLL GLDM dependence variance	0.41 ± 0.27	-0.6	0.018
Wavelet-HLL GLSZM size zone non uniformity	0.29 ± 0.33	0.57	0.028
Wavelet-HHH GLRLM gray level non uniformity normalized	0.74 ± 0.3	-0.54	0.039
Wavelet-HHH GLCM difference average	0.64 ± 0.31	0.56	0.03
Wavelet-HHH GLCM contrast	0.53 ± 0.28	0.64	0.009
Wavelet-HHH GLSZM small area emphasis	0.55 ± 0.33	0.69	0.005
Wavelet-HHH GLSZM small area low gray level emphasis	0.55 ± 0.33	0.59	0.022
Wavelet-HHL first order entropy	0.38 ± 0.3	0.57	0.025
Wavelet-HHL first order mean	0.64 ± 0.28	0.59	0.02
Wavelet-HHL first order mean absolute deviation	0.36 ± 0.29	0.55	0.032
Wavelet-LLH NGTDM complexity	0.16 ± 0.26	0.53	0.044
Wavelet-LLL GLDM small dependence high gray level emphasis	0.36 ± 0.3	0.56	0.031
Log-sigma-1-0-mm-3D first order root mean squared	0.43 ± 0.26	-0.51	0.049
Log-sigma-5-0-mm-3D GLRLM short run low gray level emphasis	0.22 ± 0.25	-0.52	0.047
Log-sigma-5-0-mm-3D first order 10 percentile	0.48 ± 0.39	-0.59	0.02

GLCM: Gray level co-occurrence matrix; GLDM: Gray level dependence matrix; GLRLM: Gray level run length matrix; GLSZM: Gray level size zone matrix; H: High-pass filter; L: Low-pass filter; NGTDM: Neighboring gray tone difference matrix; SD: Standard deviation.

Table 4 Results of multivariate analysis for the stage I-II/stage III-IV classification task

Model order	Features involved	AUC ± SE	Sens ± SE	Spec ± SE	ACC ± SE
1	Wavelet-HHH GLCM joint entropy	0.896 ± 0.005	0.513 ± 0.021	0.865 ± 0.007	0.77 ± 0.006
2	Wavelet-LLH NGTDM complexity; Wavelet-HHH GLCM joint entropy	0.869 ± 0.008	0.643 ± 0.021	0.834 ± 0.008	0.79 ± 0.006
3	Wavelet-LLH NGTDM complexity; Wavelet-HHH GLCM joint entropy; Wavelet-HLH first order entropy	0.826 ± 0.009	0.523 ± 0.023	0.815 ± 0.009	0.744 ± 0.006
4	Wavelet-HLL GLCM cluster prominence; Wavelet-HHH GLCM joint entropy; Wavelet-LLH NGTDM complexity; Wavelet-HLH first order entropy	0.799 ± 0.009	0.668 ± 0.022	0.766 ± 0.009	0.752 ± 0.007
5	Wavelet-LLH GLDM high gray level emphasis; Wavelet-HHH GLCM joint entropy; Wavelet HLL GLCM cluster prominence; Wavelet HLH entropy; Wavelet LLH NGTDM complexity	0.724 ± 0.011	0.516 ± 0.025	0.762 ± 0.01	0.708 ± 0.008

For each model (from order 1 to 5), area under the receiver operating characteristic curve, sensitivity, specificity and accuracy were reported with the standard error on a 95%CI over all bootstrap sample. AUC: Area under the receiver operating characteristic curve; GLCM: Gray level co-occurrence matrix; GLDM: Gray level dependence matrix; GLRLM: Gray level run length matrix; GLSZM: Gray level size zone matrix; H: High-pass filter; L: Low-pass filter; NGTDM: Neighboring gray tone difference matrix; SE: Standard error.