Prediction of Human Papillomavirus Status and Overall Survival in Patients with Untreated Oropharyngeal Squamous Cell Carcinoma: Development and Validation of CT-Based Radiomics

Article Figures & Data

Download figure
Open in new tab
Download powerpoint
FIG 1.
Representative contrast-enhanced CT images with a 3D ROI in a 73-year-old female patient. Volumetric rendering of segmentation is illustrated as a yellow contour (lower right).
Download figure
Open in new tab
Download powerpoint
FIG 2.
A flow diagram illustrating the patient-selection process.
Download figure
Open in new tab
Download powerpoint
FIG 3.
Receiver operating characteristic curves with AUC values for HPV-status prediction in the training, test, and TCIA cohorts (A). The P values calculated from the DeLong test between a pair of AUCs are presented above each line segment (B).
Download figure
Open in new tab
Download powerpoint
FIG 4.
A nomogram of the 2- and 5-year overall survival. Sur. Prob. indicates survival probability.
Download figure
Open in new tab
Download powerpoint
FIG 5.
Prediction error curves with respect to the overall survival of the training (black line) and TCIA cohorts (red line).

Table 1:

Clinical characteristics of the 2 study cohorts

Table 2:

Relevant features for the prediction of human papillomavirus status and overall survival

	Mean Variable Importance	Norm Hits^{^a}/Coefficients
Selected features by Boruta for prediction (Norm Hits)^{^a}
original_shape_Flatness	4.753	0.812
original_shape_SphericalDisproportion	5.012	0.828
wavelet_HLH_firstorder_Mean	9.017	0.965
wavelet_HLH_firstorder_Uniformity	3.606	0.678
wavelet_HLH_glcm_ClusterShade	4.692	0.804
wavelet_LHH_glcm_Idm	3.296	0.600
wavelet_LHH_glcm_Imc1	4.201	0.753
wavelet_LHL_glszm_SmallAreaHighGrayLevelEmphasis	11.378	0.989
wavelet_LLH_glcm_Imc2	4.503	0.792
Selected features by LASSO-Cox for OS (coefficient)
original_shape_SphericalDisproportion		3.20E-01
original_firstorder_Minimum		–2.36E-03
original_firstorder_10Percentile		–1.75E–05

Note:—LASSO indicates least absolute shrinkage and selection operator; glcm, gray-level co-occurrence matrix; glszm, gray-level size-zone matrix; idm, inverse difference moment; imc1, informational measure of correlation 1; imc2, informational measure of correlation 2.
↵a Fraction of random forest runs (Norm Hits) in which they were more important than the most important shadow value.

Table 3:

Sensitivity, specificity, and accuracy of radiomics features in predicting human papillomavirus status

Cohort	Sensitivity (95% CI)	Specificity (95% CI)	Accuracy (95% CI)	AUC (95% CI)
Training (n = 61)	76.3% (55.3%–100%)	91.3% (56.5%–100%)	80.3% (70.5%–90.2%)	0.865 (0.777–0.953)
Test (n = 25)	100% (40%–100%)	60.0% (30.0%–100%)	80.0% (60.0%–92.0%)	0.747 (0.533–0.961)
TCIA (n = 78)	82.6% (60.9%–100%)	80.0% (52.7%–94.6%)	80.8% (65.4%–89.7%)	0.834 (0.738–0.930)

Table 4:

Cox proportional hazards analyses and model validation

	HR	95%CI/C-Index	P Value
Clinical variable (95% CI)
HPV status, positive	0.26	(0.09–0.7)	.008^{^a}
Smoking history, smoker	1.9	(0.66–5.44)	.23
Age, older than 65 years	0.97	(0.34–2.78)	.96
Sex, female	0.51	(0.07–3.83)	.51
T-stage, ≥III	3.66	(1.34–9.99)	.01^{^a}
N-stage, ≥II	1.62	(0.62–4.19)	.32
Overall stage, >IV	2.7	(0.95–7.68)	.06
Rad-score, high-risk	3.72	(1.21–11.46)	.02^{^a}
Multivariate model (C-index)
Cancer stage (model 1)		0.621	.03
Cancer stage + HPV status (model 2)		0.702	.01
Cancer stage + HPV status + rad-score (model 3)		0.733	.01
Validation (C-index)
Internal validation on test set		0.866	.01
External validation on TCIA set		0.72	<.001