PV QA 3 - Poster Viewing Q&A 3
TU_14_3248 - A Technique to Rapidly Generate Synthetic CT for MRI-Guided Online Replanning of Lung Tumors
Tuesday, October 23
1:00 PM - 2:30 PM
Location: Innovation Hub, Exhibit Hall 3
Ergun Ahunbay, PhD
Medical College of Wisconsin
Medical College of Wisconsin: Associate Professor: Employee
A Technique to Rapidly Generate Synthetic CT for MRI-Guided Online Replanning of Lung Tumors
E. E. Ahunbay1, R. Thapa2, X. Chen2, and A. Li2; 1Medical College of Wisconsin, Department of Radiation Oncology, Milwaukee, WI, 2Medical College of Wisconsin, Milwaukee, WI
Purpose/Objective(s): MRI-guided online adaptive replanning requires a fast way to generate synthetic CT (sCT) based on the MRI of the day. We have previously proposed a method to quickly generate sCT based on registration of previously acquired patient’s own CT and daily MRI. For lung tumors, it’s critical and difficult to identify accurate density for lung tissue as it affects the dosimetry especially due to the electron return effect of MR-Linac. This work aims to develop and test a practical technique that can deform electron density information from CT for lung for the generation of sCT based MRI.
Materials/Methods: The proposed method consists of following steps: (1) registering previously-acquired patient’s CT with daily MRI, (2) transferring bone contours generated previously from the CT to daily MRI by registering each piece of bone individually, maintaining bone rigidity while considering relative geometric changes between individual bones from the CT to MRI, (3) segmenting air regions on daily MRI based on auto-thresholding determined from previous or the current MRI set for the patient, and forcing air density in these regions, (4) performing restricted deformable image registration (DIR) that restricts the vector field by the bone air contours to transfer CT numbers from CT to MRI for all regions outside air and bone. This method was tested on representative cases of lung cancer and doses were generated on the sCT and original CT images for comparison. Comparisons in terms of mean absolute error (MAE), gamma analysis of 3D dose distributions and dose volume histogram (DVH) of important structures were performed. Integral absolute difference (IAD) of DVHs were performed to combine all difference between two DVHs. As benchmark reference, bulk density assignment CT (bCT) sets were also generated and compared with reference CT images. bCT images used 5 different tissue classifications with rED: air (<0.28), tissue-air mix (0.28-0.86), fat (0.86 – 1.01), soft tissue (1.01-1.1), and bone (>1.1). All voxels in a range were assigned with mean HU of voxels in that range. Also 3 separate simulations were performed where the lung voxels were further binned into 1, 2, and 3 classifications that are evenly separated over the HU range of voxels.
Results: MAE numbers of sCT are lower than bCT for all cases and all tissue types. Multiple lung classifications improved the MAE in the lung tissue but did not reduce the MAE to the level of sCT even with 3 separate distinct levels. Dosimetric results also were better with sCT, with higher gamma passing rates and lower IAD of DVHs.
Conclusion: The proposed sCT method utilizing tissue specific DIR provides accurate HU numbers and dosimetric results for lung tumors and is fast for online replanning.
| || sCT || bCT || bCT + 1 Lung Level || bCT + 2 Lung Levels || bCT + 3 Lung Levels |
| MAE (HU) |
| external || 26.6 || 57.7 || 65.1 || 52.4 || 48.7 |
| bone || 90.6 || 122.3 || || || |
| soft tissue || 20.7 || 39.6 || || || |
| Lung || 30.8 || 103.0 || 131.5 || 69.4 || 52.7 |
| Gamma (%) (1.5% - 2mm) || 98.9 || 90.2 || 95.1 || 97.6 || 97.9 |
| IAD (DVH) (%-Gy) |
| Combined Lung || 2.3 || 57.9 || 17.1 || 5.4 || 4.4 |
| Spinal Cord || 7.8 || 32.1 || 9.6 || 8.8 || 10.2 |
Author Disclosure: E.E. Ahunbay: None. R. Thapa: None. X. Chen: None. A. Li: None.