Patient-Specific Deep Learning Model for Enhancing 4D-CBCT Image for Radiomic Analysis
Z Zhang*, M Huang, Z Jiang, Y Chang, K Lu, F Yin, L Ren, Duke University Medical Center, Durham, NC
Presentations
SU-IePD-TRACK 1-7 (Sunday, 7/25/2021) 3:00 PM - 3:30 PM [Eastern Time (GMT-4)]
Purpose: 4D-CBCT provides phase-resolved images valuable for radiomics analysis for outcome prediction throughout the treatment course. However, 4D-CBCT suffers from streak artifacts caused by under-sampling, which severely degrades its accuracy for deriving radiomic features. Previously we developed group-patient-trained deep learning methods to enhance the 4D-CBCT quality for radiomics analysis, which was not optimized for individual patients. In this study, we developed a patient-specific model to further improve the robustness and accuracy of 4D-CBCT based radiomics analysis for individual patients.
Methods: The patient-specific model was trained using intra-patient data. Specifically, patient planning 4D-CT was augmented through image translation, rotation and deformation to generate 305 CT volumes from 10 volumes to simulate possible patient positions during the image acquisition. 120 projections were simulated from 4D-CT per phase and were used to reconstruct 4D-CBCT using FDK back-projection algorithm. Breathing motion traces were simulated using a regular breathing curve containing 120 breathing cycles. The patient-specific model was trained using these 305 sets of patient-specific 4D-CT and 4D-CBCT data to enhance the 4D-CBCT to match with 4D-CT. For model testing, 4D-CBCT were simulated from a separate 4D-CT scan acquired from the same patient, and were then enhanced by the patient-specific model. Radiomics features were then extracted from the testing 4D-CT, 4D-CBCT, and enhanced 4D-CBCT for comparison
Results: Compared with a group-based model, patient-specific training improved the accuracy of radiomic features, especially for the first order features. First order improved 75.97% and the total radiomics improved by 31.62% from the group-based model to patient specific model. For example, first-order median feature improved 82.39% and texture non-uniformity-GLDM improved 88.09%.
Conclusion: This study demonstrated that the patient-specific model is more effective than group-based model in improving the accuracy of the 4D-CBCT radiomic features, which can consequently improve the precision for outcome prediction in radiotherapy.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by the National Institutes of Health under Grant No. R01-CA184173 and R01-EB028324.
ePosters
Keywords
Cone-beam CT, Quantitative Imaging
Taxonomy
Contact Email