Premium
SU‐D‐17A‐05: A Method to Determine the Accuracy of a Proposed Breathing Motion Model‐Based 4DCT Technique
Author(s) -
Dou T,
Thomas D,
Lamb J,
Low D
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 - 10.1118/1.4887898
Subject(s) - breathing , image registration , artificial intelligence , computer vision , landmark , computer science , metric (unit) , medical imaging , iterative reconstruction , similarity (geometry) , standard deviation , mathematics , pattern recognition (psychology) , image (mathematics) , medicine , statistics , operations management , economics , anatomy
Purpose: To develop a technique that measures the accuracy of breathing motion model based 4DCT images by comparing the CT images used to generate the model with images reconstructed using the model at the same breathing phases. Methods: A novel 4DCT technique is employed that employs free‐breathing fast helical CT images and a simultaneous breathing surrogate measurement. Deformable image registration of the CT images is used to develop a patient‐specific motion model. The model can subsequently generate deformation vector fields that generate a CT image at any user‐selected breathing phase, defined here as breathing amplitude and rate. These images will be used for treatment planning. However, unlike existing 4DCT techniques, breathing irregularity‐caused errors are not evident to the planner: Each image will look excellent, with no sorting artifacts independent of the model accuracy. To measure model accuracy, CT images are reconstructed at the breathing phases encountered during the free‐breathing CT acquisition. In principle, the reconstructed and acquired helical CT images should agree exactly, differences indicate errors in deformable registration and the motion model itself. To evaluate this approach, 5 patient studies were selected and 6 of the original 25 helical CT images compared against the generated images using a semi‐automatic landmark‐based metric. 200 pairs of spatially uniform distributed corresponding landmark points were evaluated and the mean error standard deviation determined for each image pair. Results: The model reconstructed 4DCT images showed excellent similarity to the 6 helical CT images. The mean errors ranged between sub‐millimeter to 2 millimeter with uncertainties mostly in the sub‐millimeter. When divided between upper and lower lungs, the mean errors in the upper and lower lungs were less than 2 mm and 3 mm, respectively. Conclusion: The proposed registration and model accuracy test can be used to indicate image geometry errors to a treatment planner. NIH R01CA096679