Artifacts affect 4D CTimages due to breathing irregularities or incorrect breathing phase identification. The purpose of this study is the reduction of artifacts in sorted 4D CTimages. The assumption is that the use of multiple respiratory related signals may reduce uncertainties and increase robustness in breathing phase identification.Methods:
Multiple respiratory related signals were provided by infrared 3D localization of a configuration of markers placed on the thoracoabdominal surface. Multidimensional K-means clustering was used for retrospective 4D CTimage sorting, which was based on multiple marker variables, in order to identify clusters representing different breathing phases. The proposed technique was tested on computational simulations, phantom experimental acquisitions, and clinical data coming from two patients. Computational simulations provided a controlled and noise-free condition for testing the clustering technique on regular and irregular breathing signals, including baseline drift, time variant amplitude, time variant frequency, and end-expiration plateau. Specific attention was given to cluster initialization. Phantom experiments involved two moving phantoms fitted with multiple markers. Phantoms underwent 4D CT acquisition while performing controlled rigid motion patterns and featuring end-expiration plateau. Breathing cycle period and plateau duration were controlled by means of weights leaned upon the phantom during repeated 4D CT scans. The implemented sorting technique was applied to clinical 4D CT scans acquired on two patients and results were compared to conventional sorting methods.Results:
For computational simulations and phantom studies, the performance of the multidimensional clustering technique was evaluated by measuring the repeatability in identifying the breathing phase among adjacent couch positions and the uniformity in sampling the breathing cycle. When breathing irregularities were present, the clustering technique consistently improved breathing phase identification with respect to conventional sorting methods based on monodimensional signals. In patient studies, a qualitative comparison was performed between corresponding breathing phases of 4D CTimages obtained by conventional sorting methods and by the described clustering technique. Artifact reduction was clearly observable on both data set especially in the lower lung region.Conclusions:
The implemented multiple point method demonstrated the ability to reduce artifacts in 4D CTimaging. Further optimization and development are needed to make the most of the availability of multiple respiratory related variables and to extend the method to 4D CT-PET hybrid scan.
This work was partly supported by the ULICE EU FP7 program. The authors would like to express their thanks to Dr. Jef Vandemeulebroucke (Laboratorie CREATIS, CNRS, Lyon, FR) for the interesting discussions about this work.
II. MATERIALS AND METHODS
II.A. Clustering technique
II.A.1. Initialization techniques
II.A.2. Reassignment techniques
II.B. Computational simulations
II.C. Clinical setup
II.D. Phantom and clinical data analysis
III. RESULTS AND DISCUSSION
III.A. Computational simulations
III.A.1. Stationary signal
III.A.2. Time variant baseline signal and time variant amplitude signal
III.A.3. Time variant frequency signal
III.B. Phantom data analysis
III.C. Clinical data analysis
Data & Media loading...
Article metrics loading...
Full text loading...