Four-dimensional computed tomography pulmonary ventilation ... Four-dimensional computed tomography

  • View
    218

  • Download
    0

Embed Size (px)

Text of Four-dimensional computed tomography pulmonary ventilation ... Four-dimensional computed tomography

  • Four-dimensional computed tomography pulmonary ventilation imagesvary with deformable image registration algorithms and metrics

    Tokihiro Yamamotoa!Department of Radiation Oncology, Stanford University School of Medicine,Stanford, California 94305-5847

    Sven KabusDepartment of Digital Imaging, Philips Research Europe, 22335 Hamburg, Germany

    Tobias KlinderClinical Informatics, Interventional, and Translational Solutions, Philips Research North America,Briarcliff Manor, New York 10510

    Jens von Berg and Cristian LorenzDepartment of Digital Imaging, Philips Research Europe, 22335 Hamburg, Germany

    Billy W. Loo, Jr. and Paul J. KeallDepartment of Radiation Oncology, Stanford University School of Medicine, Stanford, California 94305-5847

    !Received 18 August 2010; revised 17 December 2010; accepted for publication 5 January 2011;published 16 February 2011"

    Purpose: A novel pulmonary ventilation imaging technique based on four-dimensional !4D" CT hasadvantages over existing techniques and could be used for functional avoidance in radiotherapy.There are various deformable image registration !DIR" algorithms and two classes of ventilationmetric that can be used for 4D-CT ventilation imaging, each yielding different images. The purposeof this study was to quantify the variability of the 4D-CT ventilation to DIR algorithms and metrics.Methods: 4D-CT ventilation images were created for 12 patients using different combinations oftwo DIR algorithms, volumetric !DIRvol" and surface-based !DIRsur", yielding two displacementvector fields !DVFs" per patient !DVFvol and DVFsur", and two metrics, Hounsfield unit !HU"change !VHU" and Jacobian determinant of deformation !VJac", yielding four ventilation image sets!VHU

    vol , VHUsur , VJac

    vol, and VJacsur". First, DVFvol and DVFsur were compared visually and quantitatively to

    the length of 3D displacement vector difference. Second, four ventilation images were comparedbased on voxel-based Spearmans rank correlation coefficients and coefficients of variation as ameasure of spatial heterogeneity. VHU

    vol was chosen as the reference for the comparison.Results: The mean length of 3D vector difference between DVFvol and DVFsur was 2.0!1.1 mmon average, which was smaller than the voxel dimension of the image set and the variations.Visually, the reference VHU

    vol demonstrated similar regional distributions with VHUsur ; the reference,

    however, was markedly different from VJacvol and VJac

    sur. The correlation coefficients of VHUvol with VHU

    sur ,VJac

    vol, and VJacsur were 0.77!0.06, 0.25!0.06, and 0.15!0.07, respectively, indicating that the metric

    introduced larger variations in the ventilation images than the DIR algorithm. The spatial hetero-geneities for VHU

    vol , VHUsur , VJac

    vol, and VJacsur were 1.8!1.6, 1.8!1.5 !p=0.85", 0.6!0.2 !p=0.02", and

    0.7!0.2 !p=0.03", respectively, also demonstrating that the metric introduced larger variations.Conclusions: 4D-CT pulmonary ventilation images vary widely with DIR algorithms and metrics.Careful physiologic validation to determine the appropriate DIR algorithm and metric is neededprior to its applications. 2011 American Association of Physicists in Medicine.#DOI: 10.1118/1.3547719$

    Key words: lung, functional imaging, four-dimensional !4D" CT, deformable image registration

    I. INTRODUCTION

    Imaging techniques of regional pulmonary function !i.e.,ventilation or perfusion" could be used for functional avoid-ance in lung cancer radiotherapy17 and would also furtherour understanding of pathophysiological characteristics ofpulmonary diseases. There are several techniques for venti-lation imaging, which includes nuclear medicine imaging!the current clinical standard of care",810 hyperpolarized gasmagnetic resonance imaging !MRI",11,12 and Xe-CTimaging.1315 These techniques have drawbacks such as low

    resolution, high cost, long scan time, and/or low accessibil-ity. Ventilation images can be created by a novel four-dimensional !4D" CT-based technique.1623 The 4D-CT-derived ventilation can be considered as free informationfor lung cancer radiotherapy patients because 4D-CT scansare in routine use for treatment planning at many centers#42.3% !Ref. 24"$ and ventilation computation involves onlyimage processing. Moreover, 4D-CT ventilation imaging hashigher resolution, lower cost, shorter scan time, and higheraccessibility from radiotherapy centers than existing tech-

    1348 1348Med. Phys. 38 3, March 2011 0094-2405/2011/383/1348/11/$30.00 2011 Am. Assoc. Phys. Med.

    http://dx.doi.org/10.1118/1.3547719http://dx.doi.org/10.1118/1.3547719http://dx.doi.org/10.1118/1.3547719

  • niques and could potentially be used routinely for functionalavoidance.6,7 Its physiologic accuracy has been investigatedby comparison with the Xe-CT ventilation for anesthetizedsheep, which has demonstrated reasonablecorrelations.20,25,26 Also, the single photon emission CT!SPECT" ventilation has been used in another study for tho-racic cancer patients, which has reported low Dice similaritycoefficients but relatively high in low-functional regions.22

    Physiologically accurate 4D-CT ventilation imaging has notbeen achieved in patients and further studies are necessary.

    There are various deformable image registration !DIR"algorithms and two classes of ventilation metric that can beused for 4D-CT ventilation imaging.1623 Several DIR algo-rithms are currently under investigation and have been de-veloped, of which the transformation model ranges in com-plexity from a simple extension of a global affinetransformation using higher order polynomials with rela-tively few parameters to a completely local or free formmodel with a number of parameters where each voxel in theimage can move independently. Also, there are two classes ofsimilarity metric commonly used for DIR: Geometry-basedand intensity-based. Recently, two multi-institution studieswere conducted to evaluate the accuracy of various DIR al-gorithms using the same CT images of a deformable thoraxphantom with plastic markers27 or a lung cancer patient.28

    They compared the locations of the transformed and actuallandmarks, i.e., markers27 or bronchial bifurcations,28 andfound overall acceptable accuracy with the mean error rang-ing from 1.5 to 3.9 mm !vector length" for the phantom27 orfrom 0.7 to 1.9 mm #superior-inferior !SI" direction$ for thepatient.28 However, both studies showed large variations inthe maximum error ranging from 5.1 to 15.4 mm !vector" forthe phantom27 or from 2.0 to 7.8 mm !SI" for the patient.28

    Furthermore, Kabus et al.29 demonstrated that six differentDIR algorithms, which had similar and small mean landmarkregistration errors ranging from 1.0 to 1.4 mm, yielded vary-ing displacement vector fields !DVFs" in regions apart fromthe landmarks. Such variations in the DIR results may influ-ence 4D-CT ventilation imaging.

    Two classes of ventilation metric have been used for4D-CT ventilation imaging: Hounsfield unit !HU"change16,17,19,22,23,26,30 and Jacobian determinant ofdeformation.19,20,22,23,26,30 Both metrics are based on the as-sumptions that regional ventilation is proportional to the re-gional volume change. However, clear discrepancies be-tween the metrics have been reported by severalinvestigators.22,26,30 Recently, Castillo et al.22 demonstratedlow Dice similarity coefficients between the two metrics forthe segmented functional lung regions in thoracic cancer pa-tients. Castillo et al.22 and Yamamoto et al.30 demonstrated ahigher potential of the HU metric than the Jacobian metric incomparison with the SPECT ventilation and emphysematousvolume for patients, respectively. Moreover, Ding et al.26

    proposed a hybrid metric combining the two metrics anddemonstrated consistently higher correlations with theXe-CT ventilation than the HU metric for anesthetizedsheep. These discrepancies between the metrics obviouslyinfluence 4D-CT ventilation imaging.

    The purpose of this study was to quantify the variabilityof the 4D-CT ventilation to DIR algorithms and metrics.There has been no literature that has comprehensively quan-tified its variability to both DIR algorithms and metrics, eventhough there have been studies reporting the discrepanciesbetween the metrics as described above. We have comparedfour 4D-CT ventilation image sets computed with differentcombinations of two DIR algorithms, volumetric !DIRvol"and surface-based !DIRsur", that are fundamentally differentfrom each other and represent two broad classes of algorithmand two ventilation metrics, HU change !VHU" and JacobianVJac, that are the only two proposed classes of metric.

    II. METHODS AND MATERIALS

    This study was a retrospective analysis approved by Stan-ford Universitys Institutional Review Board. We studied 12patients !8 males and 4 females" with a mean age of 76 yr!range 6290 yr" who underwent 4D-CT scanning and radio-therapy for thoracic cancer. Four 4D-CT ventilation imagesets were created and compared for each patient as shownschematically in Fig. 1. The first step was the acquisition ofa 4D-CT image set for ten respiratory phase-based bins. Thesecond step was DIR for spatial mapping of the peak-exhale4D-CT image to the peak-inhale image using DIRvol orDIRsur. The third step was the creation of a 4D-CT ventila-tion image through the computation of ventilation metric,VHU or VJac. Finally, VJac

    vol, VHUsur , and VJac

    sur were compared to thereference VHU

    vol to quantify the variability of the 4D-CT ven-tilation. VHU

    vol was chosen as the reference as it was found tobe superior when separating ventilation in emphysema andnonemphysema lung regions.30 These steps are described indetail in the following subsections.

    Acquire 4D-CT scan andextract peak-inhale andpeak-exhale images

    Perform Perform surD