Computed tomography radiomics-based cross-sectional detection of mandibular osteoradionecrosis in head and neck cancer survivors

Abstract

Purpose: This study aims to identify radiomic features extracted from contrast-enhanced CT scans that differentiate osteoradionecrosis (ORN) from normal mandibular bone in patients with head and neck cancer (HNC) treated with radiotherapy (RT).

Materials and methods: Contrast-enhanced CT (CECT) images were collected for 150 patients (80% train, 20% test) with confirmed ORN diagnosis at The University of Texas MD Anderson Cancer Center between 2008 and 2018. Using PyRadiomics, radiomic features were extracted from manually segmented ORN regions and the corresponding automated control regions, the later defined as the contralateral healthy mandible region. A subset of pre-selected features was obtained based on correlation analysis (r > 0.95) and used to train a Random Forest (RF) classifier with Recursive Feature Elimination. Model explainability SHapley Additive exPlanations (SHAP) analysis was performed on the 20 most important features identified by the trained RF classifier.

Results: From a total of 1316 radiomic features extracted, 810 features were excluded due to high collinearity. From a set of 506 pre-selected radiomic features, the optimal subset resulting on the best discriminative accuracy of the RF classifier consisted of 67 features. The RF classifier was well calibrated (Log Loss 0.296, ECE 0.125) and achieved an accuracy of 88% and a ROC AUC of 0.96. The SHAP analysis revealed that higher values of Wavelet-LLH First-order Mean and Median were associated with ORN of the jaw (ORNJ). Conversely, higher Exponential GLDM Dependence Entropy and lower Square First-order Kurtosis were more characteristic of normal mandibular tissue.

Conclusion: This study successfully developed a CECT-based radiomics model for differentiating ORNJ from healthy mandibular tissue in HNC patients after RT. Future work will focus on the detection of subclinical ORNJ regions to guide earlier interventions.

Figure 1.

Patient exclusion and dataset split workflow.

Figure 2.

Axial view of a contrast-enhanced CT image with segmented ORN (red) and contralateral mirror-image control normal bone (yellow).

Figure 3.

Workflow for the radiomic feature extraction and modeling steps. The number of features at each step is denoted by N.

Figure 4.

A) The cumulative sum of feature importance as determined by the Random Forest model. The horizontal line represents the threshold where cumulative importance reaches 100%, and the vertical line indicates the number of features required to achieve this cumulative importance. B) The top 20 features ranked by their importance in the classification process of the RF model. The length of each bar reflects the relative importance of each feature in the model’s decision-making process, highlighting which features are most influential in differentiating between ORNJ and healthy mandibular tissues.

Figure 5.

Model performance evaluation. A) Receiver operating characteristic (ROC) curves (mean and per CV fold) of the Random Forest trained on the selected extracted radiomic features (n=67) from ORN and healthy mandible VOIs on CECT. B) Reliability curve for the Random Forest classifier. Actual outcome probabilities are plotted against predicted probabilities. The thick grey diagonal line represents an ideal calibration, where predicted probabilities align perfectly with the observed outcome frequencies. Deviations from this line indicate overconfidence (points below the diagonal) or underconfidence (points above the diagonal) in the model’s predictions. The tick marks along the x-axis show the distribution of predicted probabilities. A high concentration of ticks in a certain region indicates that many predictions fall within that probability range.

Figure 6.

SHAP values for the top 20 most influential radiomic features. The y-axis lists each feature, ordered by the average magnitude of their SHAP values. Each dot represents the SHAP value for each individual subject in the dataset illustrating the extent of each feature’s impact on the model’s prediction for differentiating between ORNJ and healthy mandibular tissue. The color gradient, ranging from blue to red, indicates the range of feature values, with red signifying higher values.