Radiomics analysis of contrast-enhanced ct images for detection of human papilloma virus in patients with oropharyngeal cancers.
Bagher-Ebadian H, Liu C, Siddiqui F, Wen N, Movsas B, and Chetty I. Radiomics analysis of contrast-enhanced ct images for detection of human papilloma virus in patients with oropharyngeal cancers. Med Phys 2018; 45(6):e411-e412.
Purpose: Human papilloma virus (HPV)-associated, head/neck cancers have been shown to exhibit increased tumor control with radiotherapy than HPVcancers. We analyzed radiomics features extracted from contrast-enhanced CT images of patients with oropharyngeal cancers with known HPV status confirmed by immunohistochemistry (P16 protein) testing. Methods: Expert-segmented contrast-enhanced computed tomography images of 30 oropharyngeal cancer patients provided by Medical Image Computing and Computer Assisted Intervention (MICCAI), and distributed by the NCI, were studied. 169 radiomics features were extracted from the segmented PTV's on CT image datasets (12 w/HPV+ and 18 w/HPV-) and involved lymph nodes (10 w/HPV- and 15 w/HPV+). Radiomics features were extracted from the following eight categories: Intensity-Based-Histogram-Features (IBHF), Gray-Level-Run-Length (GLRL), Law's-Textural-information (LAWS), Discrete- Orthonormal-Stockwell-Transform (DOST), Local-Binary-Pattern (LBP), Two-Dimensional-Wavelet-Transform (2DWT), Two-Dimensional- Gabor-Filter (2DGF), and Gray-Level-Co-Occurrence-Matrix (GLCM). Levene's testing was performed to test for homogeneity of the variance between the various groups. One-way analysis of variance (ANOVA) was used to analyze statistical significance of the null hypothesis tests between corresponding radiomics features extracted from CT image datasets of the two cohorts. Results: Among 169 features extracted from all image datasets, only 4 feature categories comprising 46 radiomics features were significantly different (pFisher<0.0001) between HPV± groups. The following Mean- Absolute-Percent-Changes (MAPC) were noted: LAWS(7 features with MAPC = 52.50%), LBP(2 features with MAPC = 4.47%), 2DWT(14 features with MAPC = 294.04%), and 2DGF(23 features with MAPC = 41.84%). Among 169 features extracted from involved lymph nodes among all datasets, only 7 feature categories comprising 86 features were significantly different (pFisher<0.0001) between HPV± groups: IBHF (1 feature with MAPC = 30.51%), LAWS(12 features with MAPC = 66.83%), DOST(3 features with MAPC = 10.82%), LBP(1 feature with MAPC = 6.57%), 2DWT(25 features with MAPC = 94.84%), 2DGF (39 features with MAPC = 48.55%), and GLCM(5 features with MAPC = 59.55%). Conclusion: These preliminary results suggest that specific radiomics features extracted from contrast-enhanced CT images datasets may serve as a biomarker for HPV status for patients with oropharyngeal cancers. This work was supported in part by a grant from Varian Medical Systems (Palo Alto, CA). Data was provided by the Medical Image Computing and Computer Assisted Intervention (MICCAI) grand challenge. Thanks to University of Texas MD Anderson Cancer Center, Jayashree Kalpathy-Cramer, PhD, Keyvan Farahani, PhD, and John Freymann, PhD.