Room: AAPM ePoster Library
Purpose: The superior soft tissue contrast of MRI compared to that of x-ray CT improves target delineation in many sites including the abdomen, making MRI-guided adaptive radiation therapy (ART) well-suited to the management of interfractional changes in a patient’s anatomy. Avoiding the error-prone deformable registration of a CT simulation scan to daily MRI setup scans in a move to MRI-only ART is desirable, but achieving synthetic CT (sCT) data in the abdomen remains a challenge due to the highly variable presence of intestinal gas. We present the dosimetric evaluation of our deep learning-based sCT reconstruction method for MRI-only ART in the abdomen.
Methods: Treating the problem of sCT reconstruction as an image-to-image translation task, we utilize a fully convolutional DenseNet to generate sCT data based on a patient’s MRI data. The burden of manually contouring intestinal gas to perform density overrides has been shifted out of the clinic to the preprocessing stages in order to prepare a well-matching training dataset.
Results: The dosimetric accuracy of sCT-based plans was first verified in patients with a limited presence of intestinal gas in both CT and MRI scans. In nine patients with notable differences in bowel filling and gas presence between CT and MR images, the percentage of the PTV covered by 95% of the prescribed dose is 21 ± 13% higher in simulation CT-based clinical plans in which these differences are not accounted for compared to sCT-based plans.
Conclusion: In conventional MRI-guided ART practice, achieving accurate dose calculations in the abdomen requires intensive manual contouring to handle the variable presence of intestinal gas. By generating sCT data directly from a patient’s daily MRI setup scan, an MRI-only ART workflow enables accurate dose calculations when the day’s anatomy is incompatible with that represented at simulation.
Image-guided Therapy, MRI, Treatment Planning