Room: Exhibit Hall | Forum 6
Purpose: In the stereotactic body radiation therapy (SBRT) procedure, cone-beam CT (CBCT) imaging is routinely performed for patient set-up and positioning verification. CBCT has valuable information of the day-to-day variation of the lesion. Radiomics applied to the CBCT allows for an effective radiation-therapy prognostic analysis. The existence of shading artifacts in CBCT due to scatter contamination reduces the accuracy of the analysis. In this study, we perform Monte Carlo (MC) based scatter correction on CBCT and analyze the radiomics features from corrected CBCT and planning CT (pCT) to explore the feasibility for radiomics analysis in CBCT.
Methods: The dataset consists of 12 patients with two-stage abdominal CBCT and pCT images. The shading artifacts are firstly suppressed using a customized ultrafast MC simulation based scatter correction algorithm. The regions of interest (ROIs) are contoured randomly in the uncorrected and corrected CBCT images and pCT images to extract the features. A total number of 271 radiomics features including 59 unfiltered and 212 wavelet-filtered features are calculated. The wavelet transformation in two orthogonal directions performs the low- or high-pass filtration on the ROIs. The transformed features are referred to as includes LL-, LH-, HL- and HH-wavelet-filtered features. The Spearmanâ€™s correlation coefficient (CC) is calculated to evaluate the variety of radiomics features after scatter correction. The features of CCs greater than 0.8 are defined as strongly correlated features.
Results: The percentages of strongly correlated features (PSCFs) are improved by 11.8% and 9.9% in the two stages respectively after scatter correction. For the LL-wavelet-filtered features, the PSCFs increase from 17.0%, 18.9% to 37.7% and 34.0% respectively.
Conclusion: Effective scatter correction can improve the correlation between the features of CBCT and pCT. The features especially LL-wavelet-filtered features extracted from scatter-corrected CBCT images have the possibility to be used in radiomics.