Radiomic Biomarkers of Locoregional Recurrence: Prognostic Insights from Oral Cavity Squamous Cell Carcinoma preoperative CT scans

Abstract

This study aimed to identify CT-based imaging biomarkers for locoregional recurrence (LR) in Oral Cavity Squamous Cell Carcinoma (OSCC) patients. Our study involved a retrospective review of 78 patients with OSCC who underwent surgical treatment at a single medical center. An approach involving feature selection and statistical model diagnostics was utilized to identify biomarkers. Two radiomics biomarkers, Large Dependence Emphasis (LDE) of the Gray Level Dependence Matrix (GLDM) and Long Run Emphasis (LRE) of the Gray Level Run Length Matrix (GLRLM) of the 3D Laplacian of Gaussian (LoG σ = 3), have demonstrated the capability to preoperatively distinguish patients with and without LR, exhibiting exceptional testing specificity (1.00) and sensitivity (0.82). The group with LRE > 2.99 showed a 3-year recurrence-free survival rate of 0.81, in contrast to 0.49 for the group with LRE ≤ 2.99. Similarly, the group with LDE > 120 showed a rate of 0.82, compared to 0.49 for the group with LDE ≤ 120. These biomarkers broaden our understanding of using radiomics to predict OSCC progression, enabling personalized treatment plans to enhance patient survival.

Figure 1

a Schematic flowchart illustrating the steps from CT image acquisition and radiomics feature extraction, through the process of machine learning techniques to risk factors. b Axial contrast-enhanced CT image of the oral cavity with a red region of interest (ROI) indicating the squamous cell carcinoma. The images were filtered by Laplacian of Gaussian (LoG) and analyzed to extract LDE and LRE features.

Figure 2

a The plot showcases a notable increase in AUC at the retention of eight informative features, followed by a decline in performance as non-informative features are incorporated. Left panel provides a comprehensive view of the AUC achieved across varying numbers of variables. Right panel offers a closer examination of the AUC within the range of 0 to 200 features. b This set of box plots presents a comparative analysis of eight distinct radiomics features. The selection process involved repeated 5x10-fold cross-validation RFE. b This frequency bar plot visualizes the counts of various degree-2 logistic models derived from the best-subset of eight radiomics features previously identified. These models were generated through 5000 iterations of data shuffling. d This density plot illustrates the distribution of two radiomic factors across two groups: recurrence and non-recurrence.

Figure 3

a These plots illustrate the logistic ROC, delineating comparisons between the full model, which incorporates previously identified radiomic and clinical features, and alternative combinations. The numbers next to each model in the legend give the AUC. b The Kaplan-Meier survival curves demonstrate a significant contrast (p = 1e-04) in recurrence-free survival (RFS) between high/low-end radiomic risk groups. c Side-by-side comparison of nomograms illustrating RFS probability predictions for the cohort. On the left, the "Full Model" includes both radiomic and clinical factors, while on the right, the "radiomics-Only Model" consists of two identified radiomic risk factors exclusively.

Figure 4

The figure displays two sets of LoG filtered images. The first three rows show OSCC images in the tongue region at varying sigma values, illustrating changes in the primary ROI. The last three rows depict normal tissue in the same region, also filtered at corresponding sigma values, to highlight contrasts between OSCC and normal textures.

Figure 5

Correlation Coefficient Heatmaps: On the left, the diagonal heatmap illustrates the pairwise correlations among radiomics features before pruning. On the right, the diagonal heatmap demonstrates the correlations of clinical features. The color scale represents the strength of the correlation, with blue indicating negative correlation, red indicating positive correlation, and white representing no correlation.