- Email: [email protected]
Improvement of multislice oxygen-enhanced MRI of the lung by fully automatic non-rigid image registration
European Journal of Radiology 81 (2012) 2900–2906
Contents lists available at SciVerse ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Improvement of multislice oxygen-enhanced MRI of the lung by fully automatic non-rigid image registration Francesco Molinari a,b,∗ , Grzegorz Bauman c , Guglielmo Paolantonio a , Tobias Geisler b , Bernhard Geiger d , Lorenzo Bonomo a , Hans-Ulrich Kauczor b,e , Michael Puderbach b,f a
Department of Bioimaging and Radiological Sciences, Catholic University of Rome, Italy Department of Radiology, Deutsches Krebsforschungszentrum DKFZ, Heidelberg, Germany c Department of Medical Physics in Radiology, Deutsches Krebsforschungszentrum DKFZ, Heidelberg, Germany d Siemens Corporate Research, Princeton, NJ, USA e Department of Radiology, University of Heidelberg, Heidelberg, Germany f Department of Interventional and Diagnostic Radiology, Thoraxklinik Heidelberg at University Hospital, Heidelberg, Germany b
a r t i c l e
i n f o
Article history: Received 17 June 2011 Accepted 3 November 2011 Keywords: Lung Magnetic resonance imaging Oxygen Image processing
a b s t r a c t Purpose: In oxygen-enhanced magnetic resonance imaging of the lung (O2-MRI), motion artifacts related to breathing hamper the quality of the parametric O2-maps. In this study, fully automatic non-rigid image registration was assessed as a post-processing method to improve the quality of O2-MRI. Materials and methods: Twenty healthy volunteers were investigated on a 1.5 T MR system. O2-MRI was obtained in four coronal sections using an IR-HASTE sequence with TE/TI of 12/1200 ms. Each section was repeatedly imaged during oxygen and room-air ventilation. Spatial differences among the images were corrected by fully automatic non-rigid registration. Signal variability, relative enhancement ratio between oxygen and room air images, and spatial heterogeneity of lung enhancement were assessed before and after image registration. Results: Motion artifacts were corrected in 5–10 s. Non-rigid registration reduced signal variability of the source images and heterogeneity of the O2-maps by 1.1 ± 0.2% and 11.2 ± 2.9%, respectively (p < 0.0001). Registration did not influence O2 relative enhancement ratio (p = 0.06). Conclusion: Fully automatic non-rigid image registration improves the quality of multislice oxygenenhanced MRI of the lung. © 2011 Elsevier Ireland Ltd. All rights reserved.
1. Introduction Oxygen-enhanced magnetic resonance imaging of the lung (O2-MRI) has been demonstrated clinically [1–3]. Inhaled oxygen diffuses through the alveolar-capillary barrier and dissolves in the pulmonary capillaries [3]. Dissolved molecules of oxygen cause signal enhancement in the areas of the lung that function normally. Defects of O2 enhancement indicate abnormal lung function [3]. Technically, lung signal in O2-MRI is measured over multiple respiratory and cardiac cycles [3]. Differences of respiratory or cardiac phase among images cause signal variability [4,5], which can be reduced by prospective triggering [6]. However, even combining respiratory and cardiac triggering, artificial signal changes are still visible in O2-MRI [7,8].
∗ Corresponding author at: Department of Bioimaging and Radiological Sciences, Catholic University of Rome, Gemelli Hospital, L.go A. Gemelli 8, 00168 Rome, Italy. Tel.: +39 06 3015 6054/4394; fax: +39 06 3550 1928. E-mail addresses: [email protected], [email protected] (F. Molinari). 0720-048X/$ – see front matter © 2011 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ejrad.2011.11.002
Breathing motion increases signal variability also by misplacing the pulmonary vessels [9]. Respiratory movements of the diaphragm and chest wall degrade the quality of the O2-maps [6]. Eventually, O2-maps hampered by anatomic misregistration represent regional lung function inadequately [6]. Anatomic mismatch among images obtained from the same section can be reduced retrospectively [10]. In particular, image registration improved the analysis of lung enhancement in singleslice O2-MRI [11]. However, the long time required for manual labeling of anatomic landmarks (35 min) prevented further assessment of image registration in multislice O2-MRI. Recently, Chefd’hotel and Faugeras [12] have proposed a method for fully automatic non-rigid image registration that corrects breathing artifacts among images within seconds [13]. We hypothesized that this method might reduce signal variability in multislice O2-MRI and improve the visualization of lung enhancement on the O2-maps. Therefore, the aim of the study was to evaluate fully automatic non-rigid image registration as a postprocessing method to improve the quality of multislice oxygen-enhanced MRI of the lung.
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
2901
2.1.5. Oxygen and room air ventilation After positioning, a standard ventilation mask (Intersurgical Ltd., Berkshire, UK) was applied to the face of the volunteer and 100% oxygen was delivered at a flow rate of 15 L/min [14]. A gas wash-in time of approximately 10 min was allowed during preliminary morphologic imaging. At continuous oxygen ventilation, fifteen IR-HASTE images per section were acquired (15 images × 4 sections = 60 images). After 5 min for oxygen wash-out, the 15 measurements were repeated at room-air ventilation. The imaging time, including the time allowed for the equilibrium of oxygen concentration, was 20–30 min. 2.2. Image post-processing and registration
Fig. 1. Oxygen-enhanced MRI of the lung: localization of the imaging planes. S1 anterior, S2 middle-anterior, S3 middle-posterior, S4 posterior. In all examinations the center of S4 matched the spinal canal at the level of the diaphragmatic dome. Spacing among slices was kept constant.
2. Materials and methods Our hospital’s institutional review board approved this prospective study. Twenty healthy nonsmoking volunteers with normal results of pulmonary function tests were consecutively enrolled in the study and participated by giving their informed consent (8 men, 12 women; mean age 49.4 years; age range 41–63 years). 2.1. Oxygen-enhanced MRI of the lung 2.1.1. System All MRI examinations were performed on a 1.5 T whole-body scanner offering a maximum gradient strength of 30 mT/m and a slew rate of 125 T/m/s (Magnetom Symphony, Siemens, Erlangen, Germany). An 8-channel chest coil was used for signal reception. 2.1.2. Sequence and imaging planes Lung signal was measured using nonselective inversionrecovery half-Fourier acquisition single-shot turbo spin-echo sequence (IR-HASTE). Imaging parameters were: inversion time 1200 ms, effective TE 12 ms, inter-echo spacing 1.98 ms, 68 phase-encoding readouts sampled consecutively, section thickness 20 mm, field of view 480 mm, matrix 128 × 128, pixel size 3.75 mm × 3.75 mm, image acquisition time 134.5 ms. TR varied according to cardiac cycles. Four coronal planes were imaged with the subject in supine position: S1 anterior, S2 middle-anterior, S3 middle-posterior, S4 posterior. Section spacing was 25 mm (Fig. 1). 2.1.3. Triggering method Navigator echoes and ECG were combined in a double triggering technique (2D-PACE, Siemens, Erlangen, Germany) [8]. When the end of the expiratory phase of the breathing curve matched the diastolic phase of the ECG, image acquisition was triggered. The trigger tolerance for diaphragm position was set to 2 mm. 2.1.4. Subject preparation Before the examination the volunteers were briefly instructed to recognize the alternating pattern of noise and silence of navigator echoes and inversion times, respectively. To ensure image acquisition at reproducible diaphragm levels, they were asked to slowly exhale during the noiseless interval of the inversion time (1.2 s) and continue breathing during noise. No breathing instructions were given to the volunteers during the examination.
2.2.1. Preparation and preliminary assessment of the images Each O2-MRI examination produced 120 images (15 images at oxygen + 15 images at room-air ventilation = 30 images × 4 sections = 120). All 120 images were transferred on a workstation and grouped according to section level. As preliminary assessment of diaphragm movements, before registration, the 30 images of each group were displayed in cine-mode. 2.2.2. Fully automatic non-rigid image registration The images of each group were then registered using software implemented in the system platform (syngo MMWP, Siemens, Erlangen, Germany). The software allowed no changes in the settings and technical parameters of the registration algorithm (fully automatic), which is illustrated in the following paragraph. 2.2.3. Registration algorithm Signal intensity summed over the entire field of view is proportional to diaphragm position. The algorithm [12] identifies a reference image with mean value of integrated signal intensity, corresponding approximately to mean cranial–caudal position of diaphragm. Non-rigid registration aligns remaining images to the reference. Spatial displacement is calculated for every uncorrected image (target). In particular, motion is modeled as a displacement vector field that gives for each pixel on the reference image its corresponding location on the target image, and regularized by a Gaussian low-pass filter. This process reflects the numerical implementation of a transport equation and provides an extensive capture range. The algorithm uses a local cross-correlation criterion as a measure of similarity between the reference and target images. This method of image similarity is well adapted to local variation of intensities observed in dynamic MR acquisitions. Similarity is incrementally maximized while estimating deformation. The rigidity constraint on the deformation field depends on the regularization of the parameters of the filter, in this case variance of a Gaussian function. The chosen regularization model has a simple parameterization and can be efficiently implemented using recursive filters. The algorithm operates at a pixel level, however it is embedded in a multiscale strategy. First, a coarse version of a displacement field is estimated from lower resolution data. The result is upsampled and used as starting point to refine the displacement field at a higher resolution, until the original image resolution is reached. This strategy helps to speed up the performance of image registration and reduces the changes of local minima. The registration algorithm transforms data without correcting signal amplitude. Therefore, signal intensity of corresponding areas of the lung within all transformed images is preserved. Manual setting of landmarks is not required. 2.3. Image analysis After non-rigid registration, an operator with previous experience in O2-MRI (F.M.) selected the regions of interest (ROIs)
2902
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
for the quantitative analysis. In each section, right and left lungs were enclosed manually in two ROIs excluding pulmonary hila, diaphragms, and lung borders. All ROIs were saved and used in the following three steps of the quantitative analysis. 2.3.1. Signal variability at oxygen and room-air ventilation In the first step of the analysis, signal variability of the nonregistered and registered images at oxygen and room air ventilation was calculated. Corresponding pixels from the first to the thirtieth image were assessed as in a time series. For example, the first pixel in the right upper corner of image one was considered with the same pixel of image two, three, etc. The pixel values of signal intensity in this time series were used to compute standard deviation and mean signal intensity. Standard deviation was divided by mean signal intensity and expressed as percentage (coefficient of variation) [15]. The analysis conducted pixelwise produced maps of coefficient of variation at each section level for these datasets: nonregistered oxygen, registered oxygen, nonregistered room air, registered room air. The ROIs were loaded on these maps and mean coefficient of variation of the lung was calculated. 2.3.2. Relative enhancement ratio In the second step of the quantitative analysis, the effect of image registration on oxygen-induced lung enhancement was assessed. Similarly to signal variability, signal enhancement was calculated pixelwise. In each pixel, average signal of the 15 images obtained at room air ventilation was subtracted from average signal of the 15 images obtained at oxygen ventilation. Signal difference was divided by average signal at room air ventilation and expressed as percentage (relative enhancement ratio) [3]. Maps reflecting relative enhancement ratio were automatically obtained from nonregistered and registered datasets (O2-maps). The ROIs were loaded on these maps and relative enhancement ratio of the lung was calculated. 2.3.3. Spatial heterogeneity of lung enhancement In the third step of the quantitative analysis, the effect of image registration on the spatial heterogeneity of lung enhancement was assessed. Coefficients of variation were calculated from the ROIs placed on the O2-maps (coefficient of variation = standard deviation/mean signal intensity × 100) [15]. 2.4. Statistical analysis All numerical data were normally distributed (Kolmogorov–Smirnov test) and were expressed as means with 95% confidence intervals. Registered datasets (test method) were compared to nonregistered ones (reference method) using analysis of variance (ANOVA) [15]. Data from the four sections were assessed using a 2 (nonregistered, registered) by 4 (sections 1, 2, 3, and 4) ANOVA for relative enhancement ratios and spatial heterogeneity of lung enhancement. An additional factor of 2 (oxygen, room air) was considered for signal variability at oxygen and room-air ventilation. All factors were considered as repeated measurements. Pairwise differences between measurements were assessed using Bonferroni post hoc test. A ˛ value of 0.05 was accepted to avoid a type 1 error. 3. Results All twenty healthy volunteers followed without effort the instructions they were given before the examination. All images were eligible for further analysis. At each section level, images were registered in 5–10 s (average time for each O2-MRI examination of
Table 1 Variables measured before and after image registration.a Variables
Nonregistered images
Coefficient of variation at oxygen ventilation, % Slice 1 9 ± 1.25 Slice 2 6.9 ± 0.68 Slice 3 5.7 ± 0.78 4.9 ± 0.55 Slice 4 Mean 6.6 ± 0.66 Coefficient of variation at room air ventilation, % 11.2 ± 1.97 Slice 1 7.8 ± 0.68 Slice 2 Slice 3 6.6 ± 0.76 5.1 ± 0.78 Slice 4 7.6 ± 0.71 Mean Relative enhancement ratio, % 22.9 ± 3.55 Slice 1 25.1 ± 4.29 Slice 2 26.7 ± 5.21 Slice 3 22.1 ± 5.10 Slice 4 Mean 24.2 ± 3.98 Coefficient of variation of the O2 maps, % 56.8 ± 5.67 Slice 1 Slice 2 46.1 ± 5.69 38.9 ± 5.68 Slice 3 39.2 ± 6.06 Slice 4 45.3 ± 5.11 Mean a
Registered images 7 5.7 4.8 4.4 5.5
± ± ± ± ±
0.88 0.61 0.71 0.58 0.51
9.2 6.4 5.9 4.6 6.5
± ± ± ± ±
1.60 0.52 0.71 0.75 0.57
22.2 24.6 26.6 21.7 23.8
± ± ± ± ±
3.65 4.25 5.15 5.00 3.95
42.7 34.9 29.0 29.7 34.1
± ± ± ± ±
6.00 3.87 3.37 5.23 3.75
Data are means ± 95% confidence intervals.
less than 1 min). Figs. 2 and 3 show a representative case of the analysis. Table 1 and Figs. 4 and 5 summarize numerical data. 3.1. Signal variability at oxygen and room-air ventilation Mean coefficients of variation decreased by 1.1 ± 0.2% after image registration (p < 0.0001, Table 1, Fig. 4) and were 1 ± 0.4% higher at room air than at oxygen ventilation (p = 0.0009). Post hoc analysis demonstrated significant differences between nonregistered and registered images at all section levels (p < 0.0001). The combined analysis of ventilation phase (oxygen, room air), image processing (nonregistered, registered), and section level resulted in a p-value of 0.73. 3.2. Relative enhancement ratio Mean relative enhancement ratio was not significantly different between registered and nonregistered datasets (0.4 ± 0.6%, p = 0.06; Table 1, Fig. 5). 3.3. Spatial heterogeneity of lung enhancement Mean coefficients of variation of O2-maps decreased by 11.2 ± 2.9% after image registration (p < 0.0001; Table 1, Fig. 5). Post hoc analysis demonstrated significant differences between nonregistered and registered images at all section levels (p < 0.0001). 4. Discussion In this study, we have shown that fully automatic non-rigid image registration improves the quality of multislice oxygenenhanced MRI of the lung. Evidence of improved quality of O2-MRI after image registration is that signal variability among the images acquired during oxygen and room air ventilation decreased, and heterogeneity of lung enhancement in the corresponding O2-maps decreased. In O2-MRI of the lung, signal changes related to inhaled oxygen are demonstrated by acquiring multiple images over different respiratory and cardiac cycles. Breathing and cardiac flow produce physiologic changes of lung signal that interfere
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
2903
Fig. 2. Signal variability at oxygen ventilation. Images represent coefficient of variation. Before registration (a–d), bright bands of 2–6 pixels indicating high signal variability superimpose on the diaphragms (motion artifacts). After registration (e–h), signal variability reduces, notably in the dorsal lung. Lung borders are less pronounced in registered images. Pulmonary vessels appear brighter and more visible in nonregistered images than in registered ones, notably in the middle-posterior and posterior slices (S3, S4).
with the assessment of oxygen-induced enhancement [6]. To our knowledge, all previous studies of O2-MRI used triggering techniques to reduce this signal variability and optimize image quality. In this study, respiratory and cardiac triggering were combined [7,8]. Imaging was performed automatically in four coronal sections, improving lung coverage from most previous studies [1–3]. As readouts occur after a long inversion time while the subject breathes freely, images synchronized prospectively by automatic
triggering may still show some degree of breathing motion [7,8]. To improve reproducibility of diaphragm levels some investigators trained the volunteers before the examination [7]. The volunteers of this study were also instructed to breathe regularly and exhale slowly during the noiseless intervals of the inversion times (1.2 s). All volunteers breathed without effort and other instructions were not required during the examination. Multislice O2-MRI was conducted in a relatively short time (20–30 min). No discomfort was reported from inhaling oxygen with the standard face mask [14].
Fig. 3. O2-maps from the same subject of Fig. 2, reflecting relative enhancement ratio. Lung enhancement is similar between nonregistered (a–d) and registered (e–h) O2maps, however more homogeneous after image registration. Nonregistered maps (a–d) show multiple bright and black bands in the lungs, indicating subtraction artifacts. The artifacts are markedly reduced after non-rigid image registration (e–h). Soft and bone tissues of the chest wall are also less visible after image registration. Overall, the quality of the O2-maps is improved. Pixel resolution is not interpolated.
2904
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
Fig. 4. Signal variability before and after image registration, at oxygen (a, b) and room air ventilation (c, d). Data are presented considering individual subjects (a, c) and the four coronal sections (b, d; S1 anterior, S2 middle-anterior, S3 middle-posterior, S4 posterior; I = 95% confidence intervals). Image registration reduced signal variability in the lung.
Subject in-room time was also optimized by delivering oxygen right after positioning. Despite optimized triggering of the sequence and training of the volunteers, at preliminary assessment, nonregistered images form the same coronal section showed small movements of the diaphragms and chest wall. The maps of signal variability confirmed these artifacts (Fig. 2), which have been also reported in other O2-MRI studies using double triggering [7,8]. Diaphragm mismatch indicates misalignment of the pulmonary structures. Naish et al. corrected misalignment in O2-MRI retrospectively by image registration [11]. In our study, anatomic mismatch of images was corrected using a multidirectional registration algorithm [12]. The approach to image registration in this study differs from that of Naish et al. [11]. The registration algorithm used in this study produces a series of vector fields of deformation calculated applying a similarity criterion between reference and target pixels over multiple directions. The registration method proposed by Naish et al. aligned pixel data in the cranial–caudal direction (spatial changes caused by diaphragm movements). Because breathing depends also on chest wall movements, a multidirectional method is required to register lung borders and vessels displaced in the transversal and oblique planes. Additionally, anatomic landmarks were set in their registration method. This resulted in a time-consuming procedure, which required human interaction with potential limits of technical reproducibility. The registration method in this study aligns images within few seconds and is totally automatic. Finally, in this
study non-rigid image registration was investigated on a wider population sample and using a multislice imaging technique. To assess the effect of image registration on the quality of O2MRI, first we measured signal variability at oxygen and room air ventilation. In both ventilation phases and at all section levels, coefficients of variation of registered images were significantly lower than those of nonregistered ones. The registration algorithm transforms pixel data without correcting amplitude of signal. As demonstrated by Bauman et al. [13], differences of lung signal related to respiratory or cardiac phase are maintained in the registered images. Therefore, signal variability was not reduced by adjusting native lung signal, but by correcting anatomic mismatch of images. Reasonably, misplaced pulmonary structures with high signal intensity such as the macroscopic vessels are responsible for artificial signal changes related to anatomic mismatch in O2-MRI. Signal variability was higher at room air than at oxygen ventilation. This result can be explained considering that room air imaging was performed in the second part of the examination, when breathing may have become less regular [16]. Therefore, pulmonary vessels were shifted to a greater extent, producing wider signal changes at room air than at oxygen imaging. Signal variability also lowered from anterior to posterior section. Because lung borders and diaphragms were excluded from the ROIs, this trend reflects only motion of the pulmonary structures. Considering that the volunteers were lying supine and gravity increases lung density from nondependent to dependent parenchyma [17],
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
2905
Fig. 5. Relative enhancement ratio (a, b) and spatial heterogeneity of lung enhancement (c, d) from the O2-maps, before and after image registration. As in Fig. 4, data are presented considering individual subjects (a, c) and the four coronal sections (b, d; S1 anterior, S2 middle-anterior, S3 middle-posterior, S4 posterior; I = 95% confidence intervals). Relative enhancement ratio did not change after image registration, whereas the homogeneity of the O2 maps improved significantly.
pulmonary vessels may have been shifted less in posterior lung regions compared to anterior ones. Additionally, cardiac pulsation also increased signal variability predominantly in the anterior lung. Secondly, we assessed the effect of image registration on oxygen-induced signal enhancement. Because registration is applied retrospectively, it is not expected to alter the value of signal enhancement. Relative enhancement ratio [3] was not statistically different after image registration. Therefore, non-rigid image registration can be safely applied to O2-MRI. Finally, we evaluated the effect of image registration on the quality of the O2-maps by calculating spatial heterogeneity of lung enhancement. In normal lungs, oxygen is expected to distribute evenly and produce homogeneous signal enhancement [3]. Conversely, defects of enhancement indicate nonfunctioning lung parenchyma [3]. In order to detect these pathologic defects and differentiate them from normal enhancement, the O2-maps of healthy lungs should appear constantly homogeneous. However, due to anatomic mismatch, subtraction artifacts appear as bright or black areas in the O2-maps mimicking pathologic condition. Therefore, heterogeneity of enhancement in healthy lungs suggests the presence of artifacts, generally indicates poor image quality, and overall reduces the potential of using O2-MRI for diagnostic purposes. Coefficients of variations of the O2-maps obtained from registered images decreased significantly, indicating more homogeneous lung enhancement. Therefore, the quality of the O2-maps after image registration improved.
4.1. Limitations This study has some limitations. We investigated image registration in healthy volunteers; our results might not apply to O2-MRI performed in patients. Considering all possible statistical interactions of ventilation phase, processing method and section level, analysis of variance was not significant. Most likely, statistical differences from data samples arranged by multiple categories were diluted. Image registration used a low-pass filter. Although previously used in O2-MRI of the lung [18], this filter may have influenced the assessment of signal variability in our study. The imaging protocol was set to optimize subject in-room time, which determined the order of acquisition of the oxygen phase before room-air phase. Inter-phase differences of variability were not assessed by switching the order of the two phases. Oxygen-induced lung enhancement has been also evaluated by correlation analysis [19]. Finally, coefficients of variation provided a global index of heterogeneity. To locate the areas of heterogeneity, analysis of fractal dimension is suggested [20].
5. Conclusion In conclusion, this study shows that fully automatic non-rigid image registration improves the quality of multislice oxygenenhanced MRI of the lung.
2906
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
Conflict of interest All the authors declare no financial or other relationship with other people or organizations that may inappropriately influence the work or that could be viewed as representing potential conflicts of interest. References [1] Ohno Y, Iwasawa T, Seo JB, et al. Oxygen-enhanced magnetic resonance imaging versus computed tomography – multicenter study for clinical stage classification of smoking-related chronic obstructive pulmonary disease. American Journal of Respiratory and Critical Care Medicine 2008;177:1095–102. [2] Ohno Y, Koyama H, Matsumoto K, et al. Oxygen-enhanced MRI vs. quantitatively assessed thin-section CT: pulmonary functional loss assessment and clinical stage classification of asthmatics. European Journal of Radiology 2011;77:85–91. [3] Ohno Y, Hatabu H. Basics concepts and clinical applications of oxygenenhanced MR imaging. European Journal of Radiology 2007;64:320–8. [4] Bankier AA, O’Donnell CR, Mai VM, et al. Impact of lung volume on MR signal intensity changes of the lung parenchyma. Journal of Magnetic Resonance Imaging 2004;20:961–6. [5] Knight-Scott J, Keilholz-George SD, Mai VM, Christopher JM. Temporal dynamics of blood flow effects in half-Fourier fast spin echo H-1 magnetic resonance imaging of the human lungs. Journal of Magnetic Resonance Imaging 2001;14:411–8. [6] Vaninbroukx J, Bosmans H, Sunaert S, et al. The use of ECG and respiratory triggering to improve the sensitivity of oxygen-enhanced proton MRI of lung ventilation. European Radiology 2003;13:1260–5. [7] Dietrich O, Losert C, Attenberger U, et al. Fast oxygen-enhanced multislice imaging of the lung using parallel acquisition techniques. Magnetic Resonance in Medicine 2005;53:1317–25. [8] Molinari F, Eichinger M, Risse F, et al. Navigator-triggered oxygen-enhanced MRI with simultaneous cardiac and respiratory synchronization for the
[9] [10] [11]
[12]
[13]
[14]
[15] [16]
[17]
[18]
[19]
[20]
assessment of interstitial lung disease. Journal of Magnetic Resonance Imaging 2007;26:1523–9. Oechsner M, Pracht ED, Staeb D, et al. Lung imaging under free-breathing conditions. Magnetic Resonance in Medicine 2009;61:723–7. Crum WR, Hartkens T, Hill DLG. Non-rigid image registration: theory and practice. British Journal of Radiology 2004;77:S140–53. Naish JH, Parker GJM, Beatty PC, et al. Improved quantitative dynamic regional oxygen-enhanced pulmonary imaging using image registration. Magnetic Resonance in Medicine 2005;54:464–9. Chefd’hotel C, Faugeras O. Flows of diffeomorphisms for multimodal image registration. In: Proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI’2002). 2002. Bauman G, Puderbach M, Deimling M, et al. Non-contrast-enhanced perfusion and ventilation assessment of the human lung by means of Fourier decomposition in proton MRI. Magnetic Resonance in Medicine 2009;62:656–64. Molinari F, Puderbach M, Eichinger M, et al. Oxygen-enhanced magnetic resonance imaging: influence of different gas delivery methods on the T1-changes of the lungs. Investigative Radiology 2008;43:427–32. Hill T, Lewicki P. STATISTICS Methods and Applications. StatSoft, Tulsa, OK; 2007. Taylor AM, Jhooti P, Wiesmann F, Keegan J, Firmin DN, Pennell DJ. MR navigator-echo monitoring of temporal changes in diaphragm position: implications for MR coronary angiography. Journal of Magnetic Resonance Imaging 1997;7:629–36. Hopkins SR, Henderson AC, Levin DL, et al. Vertical gradients in regional lung density and perfusion in the supine human lung: the Slinky effect. Journal of Applied Physiology 2007;103:240–8. Muller CJ, Schwaiblmair M, Scheidler J, et al. Pulmonary diffusing capacity: assessment with oxygen-enhanced lung MR imaging – preliminary findings. Radiology 2002;222:499–506. Dietrich O, Attenberger UI, Ingrisch M, et al. Analysis of signal dynamics in oxygen-enhanced magnetic resonance imaging. Investigative Radiology 2010;45:165–73. Levin DL, Buxton RB, Spiess JP, Arai T, Balouch J, Hopkins SR. Effects of age on pulmonary perfusion heterogeneity measured by magnetic resonance imaging. Journal of Applied Physiology 2007;102:2064–70.
Contents lists available at SciVerse ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Improvement of multislice oxygen-enhanced MRI of the lung by fully automatic non-rigid image registration Francesco Molinari a,b,∗ , Grzegorz Bauman c , Guglielmo Paolantonio a , Tobias Geisler b , Bernhard Geiger d , Lorenzo Bonomo a , Hans-Ulrich Kauczor b,e , Michael Puderbach b,f a
Department of Bioimaging and Radiological Sciences, Catholic University of Rome, Italy Department of Radiology, Deutsches Krebsforschungszentrum DKFZ, Heidelberg, Germany c Department of Medical Physics in Radiology, Deutsches Krebsforschungszentrum DKFZ, Heidelberg, Germany d Siemens Corporate Research, Princeton, NJ, USA e Department of Radiology, University of Heidelberg, Heidelberg, Germany f Department of Interventional and Diagnostic Radiology, Thoraxklinik Heidelberg at University Hospital, Heidelberg, Germany b
a r t i c l e
i n f o
Article history: Received 17 June 2011 Accepted 3 November 2011 Keywords: Lung Magnetic resonance imaging Oxygen Image processing
a b s t r a c t Purpose: In oxygen-enhanced magnetic resonance imaging of the lung (O2-MRI), motion artifacts related to breathing hamper the quality of the parametric O2-maps. In this study, fully automatic non-rigid image registration was assessed as a post-processing method to improve the quality of O2-MRI. Materials and methods: Twenty healthy volunteers were investigated on a 1.5 T MR system. O2-MRI was obtained in four coronal sections using an IR-HASTE sequence with TE/TI of 12/1200 ms. Each section was repeatedly imaged during oxygen and room-air ventilation. Spatial differences among the images were corrected by fully automatic non-rigid registration. Signal variability, relative enhancement ratio between oxygen and room air images, and spatial heterogeneity of lung enhancement were assessed before and after image registration. Results: Motion artifacts were corrected in 5–10 s. Non-rigid registration reduced signal variability of the source images and heterogeneity of the O2-maps by 1.1 ± 0.2% and 11.2 ± 2.9%, respectively (p < 0.0001). Registration did not influence O2 relative enhancement ratio (p = 0.06). Conclusion: Fully automatic non-rigid image registration improves the quality of multislice oxygenenhanced MRI of the lung. © 2011 Elsevier Ireland Ltd. All rights reserved.
1. Introduction Oxygen-enhanced magnetic resonance imaging of the lung (O2-MRI) has been demonstrated clinically [1–3]. Inhaled oxygen diffuses through the alveolar-capillary barrier and dissolves in the pulmonary capillaries [3]. Dissolved molecules of oxygen cause signal enhancement in the areas of the lung that function normally. Defects of O2 enhancement indicate abnormal lung function [3]. Technically, lung signal in O2-MRI is measured over multiple respiratory and cardiac cycles [3]. Differences of respiratory or cardiac phase among images cause signal variability [4,5], which can be reduced by prospective triggering [6]. However, even combining respiratory and cardiac triggering, artificial signal changes are still visible in O2-MRI [7,8].
∗ Corresponding author at: Department of Bioimaging and Radiological Sciences, Catholic University of Rome, Gemelli Hospital, L.go A. Gemelli 8, 00168 Rome, Italy. Tel.: +39 06 3015 6054/4394; fax: +39 06 3550 1928. E-mail addresses: [email protected], [email protected] (F. Molinari). 0720-048X/$ – see front matter © 2011 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ejrad.2011.11.002
Breathing motion increases signal variability also by misplacing the pulmonary vessels [9]. Respiratory movements of the diaphragm and chest wall degrade the quality of the O2-maps [6]. Eventually, O2-maps hampered by anatomic misregistration represent regional lung function inadequately [6]. Anatomic mismatch among images obtained from the same section can be reduced retrospectively [10]. In particular, image registration improved the analysis of lung enhancement in singleslice O2-MRI [11]. However, the long time required for manual labeling of anatomic landmarks (35 min) prevented further assessment of image registration in multislice O2-MRI. Recently, Chefd’hotel and Faugeras [12] have proposed a method for fully automatic non-rigid image registration that corrects breathing artifacts among images within seconds [13]. We hypothesized that this method might reduce signal variability in multislice O2-MRI and improve the visualization of lung enhancement on the O2-maps. Therefore, the aim of the study was to evaluate fully automatic non-rigid image registration as a postprocessing method to improve the quality of multislice oxygen-enhanced MRI of the lung.
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
2901
2.1.5. Oxygen and room air ventilation After positioning, a standard ventilation mask (Intersurgical Ltd., Berkshire, UK) was applied to the face of the volunteer and 100% oxygen was delivered at a flow rate of 15 L/min [14]. A gas wash-in time of approximately 10 min was allowed during preliminary morphologic imaging. At continuous oxygen ventilation, fifteen IR-HASTE images per section were acquired (15 images × 4 sections = 60 images). After 5 min for oxygen wash-out, the 15 measurements were repeated at room-air ventilation. The imaging time, including the time allowed for the equilibrium of oxygen concentration, was 20–30 min. 2.2. Image post-processing and registration
Fig. 1. Oxygen-enhanced MRI of the lung: localization of the imaging planes. S1 anterior, S2 middle-anterior, S3 middle-posterior, S4 posterior. In all examinations the center of S4 matched the spinal canal at the level of the diaphragmatic dome. Spacing among slices was kept constant.
2. Materials and methods Our hospital’s institutional review board approved this prospective study. Twenty healthy nonsmoking volunteers with normal results of pulmonary function tests were consecutively enrolled in the study and participated by giving their informed consent (8 men, 12 women; mean age 49.4 years; age range 41–63 years). 2.1. Oxygen-enhanced MRI of the lung 2.1.1. System All MRI examinations were performed on a 1.5 T whole-body scanner offering a maximum gradient strength of 30 mT/m and a slew rate of 125 T/m/s (Magnetom Symphony, Siemens, Erlangen, Germany). An 8-channel chest coil was used for signal reception. 2.1.2. Sequence and imaging planes Lung signal was measured using nonselective inversionrecovery half-Fourier acquisition single-shot turbo spin-echo sequence (IR-HASTE). Imaging parameters were: inversion time 1200 ms, effective TE 12 ms, inter-echo spacing 1.98 ms, 68 phase-encoding readouts sampled consecutively, section thickness 20 mm, field of view 480 mm, matrix 128 × 128, pixel size 3.75 mm × 3.75 mm, image acquisition time 134.5 ms. TR varied according to cardiac cycles. Four coronal planes were imaged with the subject in supine position: S1 anterior, S2 middle-anterior, S3 middle-posterior, S4 posterior. Section spacing was 25 mm (Fig. 1). 2.1.3. Triggering method Navigator echoes and ECG were combined in a double triggering technique (2D-PACE, Siemens, Erlangen, Germany) [8]. When the end of the expiratory phase of the breathing curve matched the diastolic phase of the ECG, image acquisition was triggered. The trigger tolerance for diaphragm position was set to 2 mm. 2.1.4. Subject preparation Before the examination the volunteers were briefly instructed to recognize the alternating pattern of noise and silence of navigator echoes and inversion times, respectively. To ensure image acquisition at reproducible diaphragm levels, they were asked to slowly exhale during the noiseless interval of the inversion time (1.2 s) and continue breathing during noise. No breathing instructions were given to the volunteers during the examination.
2.2.1. Preparation and preliminary assessment of the images Each O2-MRI examination produced 120 images (15 images at oxygen + 15 images at room-air ventilation = 30 images × 4 sections = 120). All 120 images were transferred on a workstation and grouped according to section level. As preliminary assessment of diaphragm movements, before registration, the 30 images of each group were displayed in cine-mode. 2.2.2. Fully automatic non-rigid image registration The images of each group were then registered using software implemented in the system platform (syngo MMWP, Siemens, Erlangen, Germany). The software allowed no changes in the settings and technical parameters of the registration algorithm (fully automatic), which is illustrated in the following paragraph. 2.2.3. Registration algorithm Signal intensity summed over the entire field of view is proportional to diaphragm position. The algorithm [12] identifies a reference image with mean value of integrated signal intensity, corresponding approximately to mean cranial–caudal position of diaphragm. Non-rigid registration aligns remaining images to the reference. Spatial displacement is calculated for every uncorrected image (target). In particular, motion is modeled as a displacement vector field that gives for each pixel on the reference image its corresponding location on the target image, and regularized by a Gaussian low-pass filter. This process reflects the numerical implementation of a transport equation and provides an extensive capture range. The algorithm uses a local cross-correlation criterion as a measure of similarity between the reference and target images. This method of image similarity is well adapted to local variation of intensities observed in dynamic MR acquisitions. Similarity is incrementally maximized while estimating deformation. The rigidity constraint on the deformation field depends on the regularization of the parameters of the filter, in this case variance of a Gaussian function. The chosen regularization model has a simple parameterization and can be efficiently implemented using recursive filters. The algorithm operates at a pixel level, however it is embedded in a multiscale strategy. First, a coarse version of a displacement field is estimated from lower resolution data. The result is upsampled and used as starting point to refine the displacement field at a higher resolution, until the original image resolution is reached. This strategy helps to speed up the performance of image registration and reduces the changes of local minima. The registration algorithm transforms data without correcting signal amplitude. Therefore, signal intensity of corresponding areas of the lung within all transformed images is preserved. Manual setting of landmarks is not required. 2.3. Image analysis After non-rigid registration, an operator with previous experience in O2-MRI (F.M.) selected the regions of interest (ROIs)
2902
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
for the quantitative analysis. In each section, right and left lungs were enclosed manually in two ROIs excluding pulmonary hila, diaphragms, and lung borders. All ROIs were saved and used in the following three steps of the quantitative analysis. 2.3.1. Signal variability at oxygen and room-air ventilation In the first step of the analysis, signal variability of the nonregistered and registered images at oxygen and room air ventilation was calculated. Corresponding pixels from the first to the thirtieth image were assessed as in a time series. For example, the first pixel in the right upper corner of image one was considered with the same pixel of image two, three, etc. The pixel values of signal intensity in this time series were used to compute standard deviation and mean signal intensity. Standard deviation was divided by mean signal intensity and expressed as percentage (coefficient of variation) [15]. The analysis conducted pixelwise produced maps of coefficient of variation at each section level for these datasets: nonregistered oxygen, registered oxygen, nonregistered room air, registered room air. The ROIs were loaded on these maps and mean coefficient of variation of the lung was calculated. 2.3.2. Relative enhancement ratio In the second step of the quantitative analysis, the effect of image registration on oxygen-induced lung enhancement was assessed. Similarly to signal variability, signal enhancement was calculated pixelwise. In each pixel, average signal of the 15 images obtained at room air ventilation was subtracted from average signal of the 15 images obtained at oxygen ventilation. Signal difference was divided by average signal at room air ventilation and expressed as percentage (relative enhancement ratio) [3]. Maps reflecting relative enhancement ratio were automatically obtained from nonregistered and registered datasets (O2-maps). The ROIs were loaded on these maps and relative enhancement ratio of the lung was calculated. 2.3.3. Spatial heterogeneity of lung enhancement In the third step of the quantitative analysis, the effect of image registration on the spatial heterogeneity of lung enhancement was assessed. Coefficients of variation were calculated from the ROIs placed on the O2-maps (coefficient of variation = standard deviation/mean signal intensity × 100) [15]. 2.4. Statistical analysis All numerical data were normally distributed (Kolmogorov–Smirnov test) and were expressed as means with 95% confidence intervals. Registered datasets (test method) were compared to nonregistered ones (reference method) using analysis of variance (ANOVA) [15]. Data from the four sections were assessed using a 2 (nonregistered, registered) by 4 (sections 1, 2, 3, and 4) ANOVA for relative enhancement ratios and spatial heterogeneity of lung enhancement. An additional factor of 2 (oxygen, room air) was considered for signal variability at oxygen and room-air ventilation. All factors were considered as repeated measurements. Pairwise differences between measurements were assessed using Bonferroni post hoc test. A ˛ value of 0.05 was accepted to avoid a type 1 error. 3. Results All twenty healthy volunteers followed without effort the instructions they were given before the examination. All images were eligible for further analysis. At each section level, images were registered in 5–10 s (average time for each O2-MRI examination of
Table 1 Variables measured before and after image registration.a Variables
Nonregistered images
Coefficient of variation at oxygen ventilation, % Slice 1 9 ± 1.25 Slice 2 6.9 ± 0.68 Slice 3 5.7 ± 0.78 4.9 ± 0.55 Slice 4 Mean 6.6 ± 0.66 Coefficient of variation at room air ventilation, % 11.2 ± 1.97 Slice 1 7.8 ± 0.68 Slice 2 Slice 3 6.6 ± 0.76 5.1 ± 0.78 Slice 4 7.6 ± 0.71 Mean Relative enhancement ratio, % 22.9 ± 3.55 Slice 1 25.1 ± 4.29 Slice 2 26.7 ± 5.21 Slice 3 22.1 ± 5.10 Slice 4 Mean 24.2 ± 3.98 Coefficient of variation of the O2 maps, % 56.8 ± 5.67 Slice 1 Slice 2 46.1 ± 5.69 38.9 ± 5.68 Slice 3 39.2 ± 6.06 Slice 4 45.3 ± 5.11 Mean a
Registered images 7 5.7 4.8 4.4 5.5
± ± ± ± ±
0.88 0.61 0.71 0.58 0.51
9.2 6.4 5.9 4.6 6.5
± ± ± ± ±
1.60 0.52 0.71 0.75 0.57
22.2 24.6 26.6 21.7 23.8
± ± ± ± ±
3.65 4.25 5.15 5.00 3.95
42.7 34.9 29.0 29.7 34.1
± ± ± ± ±
6.00 3.87 3.37 5.23 3.75
Data are means ± 95% confidence intervals.
less than 1 min). Figs. 2 and 3 show a representative case of the analysis. Table 1 and Figs. 4 and 5 summarize numerical data. 3.1. Signal variability at oxygen and room-air ventilation Mean coefficients of variation decreased by 1.1 ± 0.2% after image registration (p < 0.0001, Table 1, Fig. 4) and were 1 ± 0.4% higher at room air than at oxygen ventilation (p = 0.0009). Post hoc analysis demonstrated significant differences between nonregistered and registered images at all section levels (p < 0.0001). The combined analysis of ventilation phase (oxygen, room air), image processing (nonregistered, registered), and section level resulted in a p-value of 0.73. 3.2. Relative enhancement ratio Mean relative enhancement ratio was not significantly different between registered and nonregistered datasets (0.4 ± 0.6%, p = 0.06; Table 1, Fig. 5). 3.3. Spatial heterogeneity of lung enhancement Mean coefficients of variation of O2-maps decreased by 11.2 ± 2.9% after image registration (p < 0.0001; Table 1, Fig. 5). Post hoc analysis demonstrated significant differences between nonregistered and registered images at all section levels (p < 0.0001). 4. Discussion In this study, we have shown that fully automatic non-rigid image registration improves the quality of multislice oxygenenhanced MRI of the lung. Evidence of improved quality of O2-MRI after image registration is that signal variability among the images acquired during oxygen and room air ventilation decreased, and heterogeneity of lung enhancement in the corresponding O2-maps decreased. In O2-MRI of the lung, signal changes related to inhaled oxygen are demonstrated by acquiring multiple images over different respiratory and cardiac cycles. Breathing and cardiac flow produce physiologic changes of lung signal that interfere
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
2903
Fig. 2. Signal variability at oxygen ventilation. Images represent coefficient of variation. Before registration (a–d), bright bands of 2–6 pixels indicating high signal variability superimpose on the diaphragms (motion artifacts). After registration (e–h), signal variability reduces, notably in the dorsal lung. Lung borders are less pronounced in registered images. Pulmonary vessels appear brighter and more visible in nonregistered images than in registered ones, notably in the middle-posterior and posterior slices (S3, S4).
with the assessment of oxygen-induced enhancement [6]. To our knowledge, all previous studies of O2-MRI used triggering techniques to reduce this signal variability and optimize image quality. In this study, respiratory and cardiac triggering were combined [7,8]. Imaging was performed automatically in four coronal sections, improving lung coverage from most previous studies [1–3]. As readouts occur after a long inversion time while the subject breathes freely, images synchronized prospectively by automatic
triggering may still show some degree of breathing motion [7,8]. To improve reproducibility of diaphragm levels some investigators trained the volunteers before the examination [7]. The volunteers of this study were also instructed to breathe regularly and exhale slowly during the noiseless intervals of the inversion times (1.2 s). All volunteers breathed without effort and other instructions were not required during the examination. Multislice O2-MRI was conducted in a relatively short time (20–30 min). No discomfort was reported from inhaling oxygen with the standard face mask [14].
Fig. 3. O2-maps from the same subject of Fig. 2, reflecting relative enhancement ratio. Lung enhancement is similar between nonregistered (a–d) and registered (e–h) O2maps, however more homogeneous after image registration. Nonregistered maps (a–d) show multiple bright and black bands in the lungs, indicating subtraction artifacts. The artifacts are markedly reduced after non-rigid image registration (e–h). Soft and bone tissues of the chest wall are also less visible after image registration. Overall, the quality of the O2-maps is improved. Pixel resolution is not interpolated.
2904
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
Fig. 4. Signal variability before and after image registration, at oxygen (a, b) and room air ventilation (c, d). Data are presented considering individual subjects (a, c) and the four coronal sections (b, d; S1 anterior, S2 middle-anterior, S3 middle-posterior, S4 posterior; I = 95% confidence intervals). Image registration reduced signal variability in the lung.
Subject in-room time was also optimized by delivering oxygen right after positioning. Despite optimized triggering of the sequence and training of the volunteers, at preliminary assessment, nonregistered images form the same coronal section showed small movements of the diaphragms and chest wall. The maps of signal variability confirmed these artifacts (Fig. 2), which have been also reported in other O2-MRI studies using double triggering [7,8]. Diaphragm mismatch indicates misalignment of the pulmonary structures. Naish et al. corrected misalignment in O2-MRI retrospectively by image registration [11]. In our study, anatomic mismatch of images was corrected using a multidirectional registration algorithm [12]. The approach to image registration in this study differs from that of Naish et al. [11]. The registration algorithm used in this study produces a series of vector fields of deformation calculated applying a similarity criterion between reference and target pixels over multiple directions. The registration method proposed by Naish et al. aligned pixel data in the cranial–caudal direction (spatial changes caused by diaphragm movements). Because breathing depends also on chest wall movements, a multidirectional method is required to register lung borders and vessels displaced in the transversal and oblique planes. Additionally, anatomic landmarks were set in their registration method. This resulted in a time-consuming procedure, which required human interaction with potential limits of technical reproducibility. The registration method in this study aligns images within few seconds and is totally automatic. Finally, in this
study non-rigid image registration was investigated on a wider population sample and using a multislice imaging technique. To assess the effect of image registration on the quality of O2MRI, first we measured signal variability at oxygen and room air ventilation. In both ventilation phases and at all section levels, coefficients of variation of registered images were significantly lower than those of nonregistered ones. The registration algorithm transforms pixel data without correcting amplitude of signal. As demonstrated by Bauman et al. [13], differences of lung signal related to respiratory or cardiac phase are maintained in the registered images. Therefore, signal variability was not reduced by adjusting native lung signal, but by correcting anatomic mismatch of images. Reasonably, misplaced pulmonary structures with high signal intensity such as the macroscopic vessels are responsible for artificial signal changes related to anatomic mismatch in O2-MRI. Signal variability was higher at room air than at oxygen ventilation. This result can be explained considering that room air imaging was performed in the second part of the examination, when breathing may have become less regular [16]. Therefore, pulmonary vessels were shifted to a greater extent, producing wider signal changes at room air than at oxygen imaging. Signal variability also lowered from anterior to posterior section. Because lung borders and diaphragms were excluded from the ROIs, this trend reflects only motion of the pulmonary structures. Considering that the volunteers were lying supine and gravity increases lung density from nondependent to dependent parenchyma [17],
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
2905
Fig. 5. Relative enhancement ratio (a, b) and spatial heterogeneity of lung enhancement (c, d) from the O2-maps, before and after image registration. As in Fig. 4, data are presented considering individual subjects (a, c) and the four coronal sections (b, d; S1 anterior, S2 middle-anterior, S3 middle-posterior, S4 posterior; I = 95% confidence intervals). Relative enhancement ratio did not change after image registration, whereas the homogeneity of the O2 maps improved significantly.
pulmonary vessels may have been shifted less in posterior lung regions compared to anterior ones. Additionally, cardiac pulsation also increased signal variability predominantly in the anterior lung. Secondly, we assessed the effect of image registration on oxygen-induced signal enhancement. Because registration is applied retrospectively, it is not expected to alter the value of signal enhancement. Relative enhancement ratio [3] was not statistically different after image registration. Therefore, non-rigid image registration can be safely applied to O2-MRI. Finally, we evaluated the effect of image registration on the quality of the O2-maps by calculating spatial heterogeneity of lung enhancement. In normal lungs, oxygen is expected to distribute evenly and produce homogeneous signal enhancement [3]. Conversely, defects of enhancement indicate nonfunctioning lung parenchyma [3]. In order to detect these pathologic defects and differentiate them from normal enhancement, the O2-maps of healthy lungs should appear constantly homogeneous. However, due to anatomic mismatch, subtraction artifacts appear as bright or black areas in the O2-maps mimicking pathologic condition. Therefore, heterogeneity of enhancement in healthy lungs suggests the presence of artifacts, generally indicates poor image quality, and overall reduces the potential of using O2-MRI for diagnostic purposes. Coefficients of variations of the O2-maps obtained from registered images decreased significantly, indicating more homogeneous lung enhancement. Therefore, the quality of the O2-maps after image registration improved.
4.1. Limitations This study has some limitations. We investigated image registration in healthy volunteers; our results might not apply to O2-MRI performed in patients. Considering all possible statistical interactions of ventilation phase, processing method and section level, analysis of variance was not significant. Most likely, statistical differences from data samples arranged by multiple categories were diluted. Image registration used a low-pass filter. Although previously used in O2-MRI of the lung [18], this filter may have influenced the assessment of signal variability in our study. The imaging protocol was set to optimize subject in-room time, which determined the order of acquisition of the oxygen phase before room-air phase. Inter-phase differences of variability were not assessed by switching the order of the two phases. Oxygen-induced lung enhancement has been also evaluated by correlation analysis [19]. Finally, coefficients of variation provided a global index of heterogeneity. To locate the areas of heterogeneity, analysis of fractal dimension is suggested [20].
5. Conclusion In conclusion, this study shows that fully automatic non-rigid image registration improves the quality of multislice oxygenenhanced MRI of the lung.
2906
F. Molinari et al. / European Journal of Radiology 81 (2012) 2900–2906
Conflict of interest All the authors declare no financial or other relationship with other people or organizations that may inappropriately influence the work or that could be viewed as representing potential conflicts of interest. References [1] Ohno Y, Iwasawa T, Seo JB, et al. Oxygen-enhanced magnetic resonance imaging versus computed tomography – multicenter study for clinical stage classification of smoking-related chronic obstructive pulmonary disease. American Journal of Respiratory and Critical Care Medicine 2008;177:1095–102. [2] Ohno Y, Koyama H, Matsumoto K, et al. Oxygen-enhanced MRI vs. quantitatively assessed thin-section CT: pulmonary functional loss assessment and clinical stage classification of asthmatics. European Journal of Radiology 2011;77:85–91. [3] Ohno Y, Hatabu H. Basics concepts and clinical applications of oxygenenhanced MR imaging. European Journal of Radiology 2007;64:320–8. [4] Bankier AA, O’Donnell CR, Mai VM, et al. Impact of lung volume on MR signal intensity changes of the lung parenchyma. Journal of Magnetic Resonance Imaging 2004;20:961–6. [5] Knight-Scott J, Keilholz-George SD, Mai VM, Christopher JM. Temporal dynamics of blood flow effects in half-Fourier fast spin echo H-1 magnetic resonance imaging of the human lungs. Journal of Magnetic Resonance Imaging 2001;14:411–8. [6] Vaninbroukx J, Bosmans H, Sunaert S, et al. The use of ECG and respiratory triggering to improve the sensitivity of oxygen-enhanced proton MRI of lung ventilation. European Radiology 2003;13:1260–5. [7] Dietrich O, Losert C, Attenberger U, et al. Fast oxygen-enhanced multislice imaging of the lung using parallel acquisition techniques. Magnetic Resonance in Medicine 2005;53:1317–25. [8] Molinari F, Eichinger M, Risse F, et al. Navigator-triggered oxygen-enhanced MRI with simultaneous cardiac and respiratory synchronization for the
[9] [10] [11]
[12]
[13]
[14]
[15] [16]
[17]
[18]
[19]
[20]
assessment of interstitial lung disease. Journal of Magnetic Resonance Imaging 2007;26:1523–9. Oechsner M, Pracht ED, Staeb D, et al. Lung imaging under free-breathing conditions. Magnetic Resonance in Medicine 2009;61:723–7. Crum WR, Hartkens T, Hill DLG. Non-rigid image registration: theory and practice. British Journal of Radiology 2004;77:S140–53. Naish JH, Parker GJM, Beatty PC, et al. Improved quantitative dynamic regional oxygen-enhanced pulmonary imaging using image registration. Magnetic Resonance in Medicine 2005;54:464–9. Chefd’hotel C, Faugeras O. Flows of diffeomorphisms for multimodal image registration. In: Proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI’2002). 2002. Bauman G, Puderbach M, Deimling M, et al. Non-contrast-enhanced perfusion and ventilation assessment of the human lung by means of Fourier decomposition in proton MRI. Magnetic Resonance in Medicine 2009;62:656–64. Molinari F, Puderbach M, Eichinger M, et al. Oxygen-enhanced magnetic resonance imaging: influence of different gas delivery methods on the T1-changes of the lungs. Investigative Radiology 2008;43:427–32. Hill T, Lewicki P. STATISTICS Methods and Applications. StatSoft, Tulsa, OK; 2007. Taylor AM, Jhooti P, Wiesmann F, Keegan J, Firmin DN, Pennell DJ. MR navigator-echo monitoring of temporal changes in diaphragm position: implications for MR coronary angiography. Journal of Magnetic Resonance Imaging 1997;7:629–36. Hopkins SR, Henderson AC, Levin DL, et al. Vertical gradients in regional lung density and perfusion in the supine human lung: the Slinky effect. Journal of Applied Physiology 2007;103:240–8. Muller CJ, Schwaiblmair M, Scheidler J, et al. Pulmonary diffusing capacity: assessment with oxygen-enhanced lung MR imaging – preliminary findings. Radiology 2002;222:499–506. Dietrich O, Attenberger UI, Ingrisch M, et al. Analysis of signal dynamics in oxygen-enhanced magnetic resonance imaging. Investigative Radiology 2010;45:165–73. Levin DL, Buxton RB, Spiess JP, Arai T, Balouch J, Hopkins SR. Effects of age on pulmonary perfusion heterogeneity measured by magnetic resonance imaging. Journal of Applied Physiology 2007;102:2064–70.