(Monday, 7/30/2018) 1:15 PM - 1:45 PM
Room: Exhibit Hall | Forum 2
Purpose: Radiomics based on cone beam computed tomography (CBCT) scans acquired for patient alignment during chemoradiotherapy may allow early response assessment. We characterized longitudinal changes in CBCT-based radiomics features for lung cancer.
Methods: Eleven CBCT image datasets at an interval of three fractions were studied for 27 patients with locally advanced lung cancer. Tumor contours were propagated from the planning CT image to the CBCT image by deformable image registration, and then used for feature calculations. 658 features were extracted from each CBCT dataset. Feature selection was performed based on the following criteria: test-retest repeatability, robustness against contouring uncertainties, and non-redundancy. We investigated whether early changes in features can predict features at the end of treatment. Linear models were fitted using all feature values between the first and the nᵗʰ (n=2, 3, …) image, and then were used to predict the value at the end of treatment. The relationship of predicted values with actual values was evaluated by the coefficient of determination (R²). Furthermore, we quantified the Pearson’s correlation coefficient (R) between changes in features and tumor volume changes.
Results: Twelve features were selected based on the three criteria. The actual feature values at the end of treatment were strongly related to the predicted values based on the first 3-7 CBCT scans (corresponding to 2-4 weeks) (R² range: 0.67-0.93). Changes in all features were weakly correlated with tumor volume changes (R range: -0.22-0.08), except for the shape feature that is inversely proportional to the volume by definition.
Conclusion: In a 27-patient study, early changes in CBCT-based radiomics features were found to be strongly related to features at the end of treatment for lung cancer. Moreover, changes in most features were only weakly correlated with changes in tumor volume. These findings justify further investigations into delta-radiomics based on early CBCT imaging.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by the China National Key Research and Development Program (2016YFC0103400).
Keywords
Cone-beam CT, Image Analysis, Lung
Taxonomy
IM/TH- Image Analysis (Single modality or Multi-modality): Imaging biomarkers and radiomics
Contact Email
