367 - Improving Robustness of CBCT Radiomics with 2D Antiscatter Grid Based CBCT

Purpose/Objective(s): Predicting treatment response and toxicity using serially acquired CBCT images throughout the treatment course is a desirable yet challenging task in radiotherapy. One major challenge is the poor quality of CBCT images, as changes in image feature characteristics of normal tissues and targets may be masked by artifacts, low accuracy of CT numbers, and low contrast resolution in CBCT. Hence, in this work, a new 2D antiscatter-based quantitative CBCT technique (qCBCT) was developed and implemented. Subsequently, the effect of qCBCT on the robustness of CBCT radiomics was investigated prospectively in a pilot trial.

Materials/Methods: In an IRB-approved protocol, a total of 28 patients treated with radiotherapy, 18 with cancers in the pelvis and abdomen, and 10 in H&N, were accrued. The qCBCT system, consisting of a 2D antiscatter grid prototype and data correction algorithms, was developed and implemented in a Varian medical linear accelerator CBCT system. Each study participant was scanned twice: once with a medical linear accelerator's standard-of-care CBCT protocol and once with the qCBCT. The former scan was reconstructed using standard and iCBCT protocols. For each participant, both scans were done using identical acquisition parameters and dose within a 2-hour window. 14 unique tissue structures were delineated in each CBCT and the reference planning CT. Subsequently, 315 first order statistics and texture-based image features were extracted. The consistency of CBCT features was assessed by calculating Pearson Correlation Coefficients (PCC) between each CBCT and planning CT for each participant and tissue structure. Features with PCC > 0.6 were considered robust for downstream robustness analysis.

Results: Among 315 radiomic features evaluated in each structure, 63 ± 13, 66 ± 12, and 114 ± 11 features met the robustness criterion in standard CBCT, iCBCT, and qCBCT, respectively (p < 0.005). For 12 out of 14 tissue structures evaluated, qCBCT provided a greater number of robust radiomics features than both types of standard-of-care CBCTs. Mean PCC values for standard CBCT, iCBCT, and qCBCT were 0.73 ± 0.01, 0.77 ± 0.01, and 0.81 ± 0.01, indicating that qCBCT has higher level of robustness on average (p < 0.001). In disease site specific analysis, similar trends were observed for pelvis/abdomen and H&N patient cohorts, with the highest PCC values for qCBCT observed in the H&N cohort. iCBCT provided higher PCC values than the standard CBCT (p < 0.04).

Conclusion: In this pilot trial, the qCBCT approach yielded more robust radiomic features than standard-of-care CBCTs for most tissues. The level of robustness, as measured by the Pearson Correlation Coefficient, was also significantly higher for features extracted from qCBCT images. Therefore, qCBCT has the potential to improve the fidelity of CBCT radiomics, which may enhance the utilization of daily acquired CBCT images for treatment response and toxicity prediction in the future.