Premium
TH‐E‐17A‐03: Development and Evaluation of a 4D‐CBCT Scheme Based On Simultaneous Motion Estimation and Image Reconstruction
Author(s) -
Dang J,
Gu X,
Ouyang L,
Pan T,
Wang J
Publication year - 2014
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4889678
Subject(s) - computer vision , iterative reconstruction , artificial intelligence , motion estimation , computer science , projection (relational algebra) , motion (physics) , image registration , mathematics , algorithm , image (mathematics)
Purpose: To develop and evaluate the performance of a novel 4D‐CBCT reconstruction scheme based on simultaneous motion estimation and image reconstruction (SMEIR) through patient studies. Methods: The SMEIR algorithm consists of two alternating steps: 1) motion‐compensated CBCT reconstruction; and 2) motion model estimation from the projections directly. In step 1, projections from all phases are used to reconstruct a reference phase 4D‐CBCT by explicitly considering the motion models between different phases. In step 2, updated inverse consistent motion models are obtained directly from projections by matching the measured projections to the forward projection of the deformed reference phase 4D‐CBCT. Motion model estimation was implemented on GPU to increase the computation efficiency. Two lung cancer patients were scanned for 4.5 and 5.7 minutes. Totally 1679 and 1982 projections were acquired and grouped into 10 phases. To evaluate SMEIR algorithm performance on conventional 1‐minute CBCT scan, projections at each phase were down‐sampled by factors ranging from 2 to 10. 4D‐CBCT were reconstructed from down‐sampled projections using FDK, total variation minimization (TV) and SMEIR. Using 4D‐CBCT reconstructed from fully sampled projections as a reference, relative errors of image and tumor motion trajectory were analyzed to quantify performances of different reconstruction algorithms. Results: SMEIR algorithm outperforms FDK and TV in both reconstruction accuracy and tumor tracking accuracy. When averaged projection number per phase decreases to 18 in two patients, relative reconstruction errors for FDK, TV and SMEIR are 38.00%, 11.72% and 9.59%, respectively. The maximum tumor tracking errors are 2.53 mm, 2.05 mm, and 0.68 mm for FDK, TV and SMEIR, respectively. Conclusion: Patient studies show that the SMEIR algorithm achieves 1‐mm tumor tracking accuracy when the average projection number per phase reduces to 18. SMEIR algorithm enables the use of conventional 1‐minute CBCT for accurate motion modeling and 4D‐CBCT reconstruction.