Dynamic 2D Cine From Beam Eye’s View (BEV) with Projected Tumor Volume for MR-Guided Radiotherapy
X Nie1, G Li2, J Jeong2*, (1) University of Kentucky, Lexington, KY, (2) Memorial Sloan Kettering Cancer Center, New York, NY
Presentations
WE-C1000-IePD-F2-4 (Wednesday, 7/13/2022) 10:00 AM - 10:30 AM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 2
Purpose: To minimize the latency in projecting tumor volume contour on magnetic resonance (MR) 2D beam eye’s view (BEV) cine using a predictive strategy to identify matched tumor volume for MR-guided intensity-modulated radiotherapy (IMRT) or volumetric-modulated arc therapy (VMAT).
Methods: A previously-developed 2D BEV-cine strategy with volumetric tumor projection for MR-guided IMRT via 2D-3D library matching (Nie, et al, PMB, 2021) has been modified to identify tumor volume by predicting for both IMRT and VMAT treatments. Three time-resolved (TR) 4DMRI series (3x80 images) in each of eight lung cancer patients were used to create two patient-specific libraries in exhalation and inhalation (~0.2mm interval) to account for motion hysteresis effect. Three motion waveforms were extracted from TR-4DMRI and interpolated (b-spline) from 2Hz to 4Hz to match the 2Dcine frame rate. An autoregressive (AR) modeling and a long short-term memory (LSTM) deep-learning network were applied to predict tumor position on the next 2D BEV-cine in 250ms. A total of 3x(40-1)=117 predictions per patient were produced by moving the 120 training timepoints along with the 160-timepoint waveforms. A matched tumor volume was identified using both motion direction and amplitude. The accuracy of positional prediction and tumor volume projection was assessed against the embedded ground truth, including the position difference and Dice similarity index.
Results: The accuracy of the prediction at 250ms is 0.4±0.1mm (AR) and 0.6±0.2mm (LSTM). The error in identifying the closest match in the libraries is negligible due to the canceling effect in the small random error. The accuracy of projected volumetric tumor contours on 2D BEV-cine images is DICE=0.97±0.02 for IMRT and VMAT.
Conclusion: This study demonstrates that it is feasible to accurately predict tumor position, identify a matched tumor volume accounting for tumor-motion hysteresis, and project tumor volume contour on 2D BEV-cine, potentially applicable for MR-guided IMRT and VAMT.
Keywords
Image-guided Therapy, MRI, Modeling
Taxonomy
IM/TH- MRI in Radiation Therapy: MRI/Linear accelerator combined- IGRT and tracking
Contact Email