Head and Neck Cancer

PV QA 2 - Poster Viewing Q&A 2

MO_24_2612 - FDG-PET Imaging-derived Radiomics Correlates of Human Papilloma Virus status: Connecting the Dots in the Oropharyngeal Cancer Biology, Metabolism, and Imaging Interplay

Monday, October 22
10:45 AM - 12:15 PM
Location: Innovation Hub, Exhibit Hall 3

FDG-PET Imaging-derived Radiomics Correlates of Human Papilloma Virus status: Connecting the Dots in the Oropharyngeal Cancer Biology, Metabolism, and Imaging Interplay
H. Elhalawani1, D. Mackin2, R. B. Ger3, T. Lin4, A. S. Mohamed4, C. Rock5, G. B. Gunn4, A. S. Garden4, D. I. Rosenthal4, L. E. Court6, and C. D. Fuller7; 1Department of Radiation Oncology, University of Texas MD Anderson Cancer Center, Houston, TX, 2University of Texas M.D. Anderson Cancer Center, Houston, TX, 3MD Anderson Cancer Center, Houston, TX, 4The University of Texas MD Anderson Cancer Center, Houston, TX, 5University of Texas Health Science Center San Antonio School of Medicine, San Antonio, TX, 6Department of Radiation Physics, The University of Texas MD Anderson Cancer Center, Houston, TX, 7University of Texas Graduate School of Biomedical Sciences, Houston, TX

Purpose/Objective(s): Growing evidence has established differential clinical, pathologic, molecular, and epidemiologic attributes between human papilloma virus-associated (HPV+) and HPV-negative (HPV-) oropharyngeal cancer (OPC) disease entities. We hypothesize that differences in OPC biology can result in discernible metabolic changes. To test this hypothesis, we attempted to predict HPV status using machine learning models trained using quantifiable metabolic imaging features derived from FDG positron emission tomography (FDG PET).

Materials/Methods: We reviewed the records of 980 oropharyngeal cancer patients from an IRB-approved cohort treated at our institution between 2005 and 2012. Patients with pre-treatment PET imaging and known p16 status (as a surrogate of HPV status) were included in our analysis. Patients were considered p16 positive if testing yielded >70% cytoplasmic and nuclear staining. An in-house built software tool was applied to auto-segment intact primary gross tumor volumes (GTVp) on pre-treatment PET images that was then reviewed and edited. IBEX, an open-source radiomics analysis platform, was used to analyze 122 radiomics features in the following categories: intensity (n=8), grey-level co-occurrence matrix ‘GLCM’ (n=110), and shape (n=4). Features with an absolute Spearman correlation >0.8 with absolute value to volume, mean image intensity, or image intensity standard deviation were removed. The least absolute shrinkage and selection operator (LASSO) method was used for radiomics feature selection. Model performance was evaluated with area under the receiver operator characteristic curve (ROC AUC).

Results: 172 patients with known p16 status (131 HPV+ & 41 HPV-) and available pre-treatment PET imaging were included in the final cohort. Selected patient characteristics by HPV status are given in Table 1. After removing correlated features, there were 36 radiomics features remaining. The fundamental features volume, mean, and standard deviation were selected rather than their correlates. The LASSO fit selected five features for the final model. The most discriminating feature was the shape feature ‘Convex’. The other discriminating features were the GLCM features ‘Information Measure Correlation 1’ and ‘Inverse Variance’, the shape feature ‘Roundness’, and the intensity ‘Mean’. The apparent AUC for the composite radiomics signature was 76.2 [95% CI: 67.5 to 85.0].

Conclusion: A composite radiomics signature solely derived from baseline PET imaging could correlate to intrinsic tumor biology on a subcellular level. Such insight might promote the ongoing initiatives of transitioning RT to metabolically-driven planning/delivery algorithms. Table 1. Patients and disease characteristics
HPV- HPV+
Number (%) 41 (24%) 131 (76%)
Age, median (IQR) 57 (52-67) 58 (52-64)
Gender Male 30 (86%) 120 (92%)
Female 5 (14%) 11 (8%)
Smoking Status, (%)(Current/Former/Never) 34/49/17 20/31/49
AJCC Stage, 8th Edition (Stage X/1/2/3/4) 1/0/1/5/34 0/13/96/22/0

Author Disclosure: H. Elhalawani: None. D. Mackin: None. R.B. Ger: None. T. Lin: None. A.S. Mohamed: None. C. Rock: None. G.B. Gunn: Associate Medical Director; MD Anderson Cancer Center - Proton Therapy. A.S. Garden: None. D.I. Rosenthal: None. L.E. Court: None. C.D. Fuller: Research Grant; National Institutes of Health, National Science Foundation, Elekta AB. Grant funding; Elekta AB. Honoraria; Nederlandse Organisatie voor Wetenschappelijk Onde. Consultant; Elekta AB, Nederlandse Organisatie voor Wetenschappelijk Onde. Travel Expenses; Elekta AB, Nederlandse Organisatie voor Wetenschappelijk Onde. Reviewer; Radiological Society of North America. Associate Editor; Radiographics. Data Management Task Force Committee Member; MR-LinAc Consortium. Member; National Cancer Institute. Task Group Member; American Association of Physicists in Medicine.

Send Email for Hesham Elhalawani


Assets

MO_24_2612 - FDG-PET Imaging-derived Radiomics Correlates of Human Papilloma Virus status: Connecting the Dots in the Oropharyngeal Cancer Biology, Metabolism, and Imaging Interplay



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for FDG-PET Imaging-derived Radiomics Correlates of Human Papilloma Virus status: Connecting the Dots in the Oropharyngeal Cancer Biology, Metabolism, and Imaging Interplay