Skip to main content
ASTRO 2024 Main banner
Icon Legend
Presentation Icons
Event Add-on
Event Add-on
In Person Only
In Person Only
Off-site Event
Off-site Event
Location: Marriott Marquis
Location: Marriott Marquis
Back
Main Session View Poster

PQA 10 - PQA 10 Head & Neck Cancer and Health Services Research/Global Oncology Poster Q&A

3668 - Radiomic Correlates of Mandibular Osteoradionecrosis after Radiation Treatment of Head and Neck Cancer Patients

Wednesday, October 2, 2024
10:30 AM – 11:45 AM ET
Location: Hall C

Screen: 21

S. K. Kamel1, L. Humbert-Vidan2, S. Y. Lai3, L. V. van Dijk4, D. Fuentes5, C. D. Fuller2, and A. S. Mohamed2,6; 1Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, HOUSTON, TX, 2Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, 3The University of Texas MD Anderson Cancer Center, Division of Surgery, Houston, TX, 4Department of Radiation Oncology, University Medical Center Groningen, University of Groningen, Groningen, Netherlands, 5Department of Imaging Physics, The University of Texas MD Anderson Cancer Center, Houston, TX, 6Department of Radiation Oncology, Dan L. Duncan Comprehensive Cancer Center, Baylor College of Medicine, Houston, TX

Purpose/Objective(s): To define radiomic signatures that distinguish mandibular osteoradionecrosis (ORN) from healthy mandibular tissue on contrast-enhanced CT (CECT) scans of head and neck cancer (HNC) patients following radiotherapy (RT). Materials/

Methods: Patients with confirmed diagnosis of HNC who developed unilateral ORN of the mandible post-RT were identified. Post-RT CECT scans were retrieved and ORN ROIs were segmented and mirrored on the contralateral healthy tissue control. Radiomic features were extracted using an open source software. The dataset was divided into training (80%) and testing (20%) sets for model development and validation. Recursive Feature Elimination with Cross-Validation (RFECV) integrated with a Random Forest (RF) classifier were used to select and rank ORN-distinctive features. SHAP values were computed and plotted to explain the features influence on the model decisions.

Results: The mean age of the 150 included patients at the time of CECT acquisition was 62.3 years (range 27-82). The average time to ORN post-RT onset was 32.6 months. A total of 1316 radiomic features were initially extracted from 300 ROIs representing ORN and control ROIs. After excluding features with high collinearity (correlation coefficient > 0.95), 508 features were used in model development. The RF classifier with RFECV identified 67 optimal features for the final model. The 5-fold cross-validation conducted on the training dataset resulted in an average AUC of 0.94. For the independent testing dataset, the model achieved an accuracy of 88% and an AUC of 0.96. The model achieved sensitivity of 93%, and specificity of 83%. A simplified model, utilizing the top 5 features demonstrated an accuracy of 85% and an AUC of 0.92 (Table 1)

Conclusion: This study demonstrates the utility of CECT-based radiomics in differentiating ORN from healthy mandibular tissue in HNC patients post-RT. The developed model, informed by key radiomic features, exhibited high accuracy, sensitivity, and specificity. The findings highlight the potential of radiomic profiling in the early detection of ORN. Future studies are needed to refine predictive models for pre-ORN stages and explore interventions to mitigate ORN risk. Abstract 3668 – Table 1: Top 20 selected features based on rank importance
Class Feature Filter Rank Mean SD P-Value Importance
First Order Median Wavelet-LLH 1 -0.241 3.761 <0.01 0.062
First Order Mean Wavelet-LLH 2 -0.844 4.538 <0.01 0.057
GLDM Dependence Entropy Exponential 3 6.491 0.718 <0.01 0.049
First Order 90 Percentile Wavelet-LLL 4 759.283 91.634 <0.01 0.044