Internal respiratory surrogate in multislice 4D CT using a combination of Fourier transform and anatomical features

• Published in: Medical physics (Lancaster) 2015-07, Vol.42 (7), p.4338-4348
• Main Authors: Hui, Cheukkai, Suh, Yelin, Robertson, Daniel, Pan, Tinsu, Das, Prajnan, Crane, Christopher H., Beddar, Sam
• Format: Article
• Language: English
• Subjects: 4D CT; 60 APPLIED LIFE SCIENCES; ALGORITHMS; anatomical feature; Biological material, e.g. blood, urine; Haemocytometers; cluster; Cluster analysis; Computed tomography; Computerised tomographs; computerised tomography; COMPUTERIZED TOMOGRAPHY; Cone beam computed tomography; Datasets as Topic; Digital computing or data processing equipment or methods, specially adapted for specific applications; feature extraction; Four-Dimensional Computed Tomography - methods; Fourier Analysis; Fourier transform; FOURIER TRANSFORMATION; Fourier transforms; Humans; image classification; Image data processing or generation, in general; IMAGE PROCESSING; Integral transforms; internal surrogate; Lungs; Medical image artifacts; medical image processing; Medical image reconstruction; Motion; Multislice computed tomography; PATIENTS; Pneumodyamics, respiration; pneumodynamics; RADIATION PROTECTION AND DOSIMETRY; Radiography, Abdominal - methods; Radiography, Thoracic - methods; Respiration; Retrospective Studies; Time Factors
• ISSN: 0094-2405; 2473-4209
• DOI: 10.1118/1.4922692

Purpose: The purpose of this study was to develop a novel algorithm to create a robust internal respiratory signal (IRS) for retrospective sorting of four‐dimensional (4D) computed tomography (CT) images. Methods: The proposed algorithm combines information from the Fourier transform of the CT images and from internal anatomical features to form the IRS. The algorithm first extracts potential respiratory signals from low‐frequency components in the Fourier space and selected anatomical features in the image space. A clustering algorithm then constructs groups of potential respiratory signals with similar temporal oscillation patterns. The clustered group with the largest number of similar signals is chosen to form the final IRS. To evaluate the performance of the proposed algorithm, the IRS was computed and compared with the external respiratory signal from the real‐time position management (RPM) system on 80 patients. Results: In 72 (90%) of the 4D CT data sets tested, the IRS computed by the authors’ proposed algorithm matched with the RPM signal based on their normalized cross correlation. For these data sets with matching respiratory signals, the average difference between the end inspiration times (Δtins) in the IRS and RPM signal was 0.11 s, and only 2.1% of Δtins were more than 0.5 s apart. In the eight (10%) 4D CT data sets in which the IRS and the RPM signal did not match, the average Δtins was 0.73 s in the nonmatching couch positions, and 35.4% of them had a Δtins greater than 0.5 s. At couch positions in which IRS did not match the RPM signal, a correlation‐based metric indicated poorer matching of neighboring couch positions in the RPM‐sorted images. This implied that, when IRS did not match the RPM signal, the images sorted using the IRS showed fewer artifacts than the clinical images sorted using the RPM signal. Conclusions: The authors’ proposed algorithm can generate robust IRSs that can be used for retrospective sorting of 4D CT data. The algorithm is completely automatic and requires very little processing time. The algorithm is cost efficient and can be easily adopted for everyday clinical use.