An image acquisition and registration strategy for the fusion of hyperpolarized helium-3 MRI and x-ray CT images of the lung

Abstract

The purpose of this ethics committee approved prospective study was to evaluate an image acquisition and registration protocol for hyperpolarized helium-3 magnetic resonance imaging (³He-MRI) and x-ray computed tomography. Nine patients with non-small cell lung cancer (NSCLC) gave written informed consent to undergo a free-breathing CT, an inspiration breath-hold CT and a 3D ventilation ³He-MRI in CT position using an elliptical birdcage radiofrequency (RF) body coil. ³He-MRI to CT image fusion was performed using a rigid registration algorithm which was assessed by two observers using anatomical landmarks and a percentage volume overlap coefficient. Registration of ³He-MRI to breath-hold CT was more accurate than to free-breathing CT; overlap 82.9 ± 4.2% versus 59.8 ± 9.0% (p < 0.001) and mean landmark error 0.75 ± 0.24 cm versus 1.25 ± 0.60 cm (p = 0.002). Image registration is significantly improved by using an imaging protocol that enables both ³He-MRI and CT to be acquired with similar breath holds and body position through the use of a birdcage ³He-MRI body RF coil and an inspiration breath-hold CT. Fusion of ³He-MRI to CT may be useful for the assessment of patients with lung diseases.

1. Introduction

Over the last decade, the technology for magnetic resonance imaging (MRI) with inert gases such as hyperpolarized helium-3 (³He) has emerged for lung ventilation imaging (Wild et al 2002) and has been clinically applied to diseases such as cystic fibrosis, asthma and emphysema (Donnelly et al 1999, Salerno et al 2002, van Beek et al 2004, Fain et al 2006). Hyperpolarized gas imaging enables novel quantitative analysis of pulmonary physiology (Wild et al 2004) and has the potential to provide clinical information superior to single-photon emission computed tomography (SPECT) without the need for radioisotopes (Stavngaard et al 2005).

As with other functional radiological investigations, there may be significant benefits in combining ³He-MR images to the anatomical structure provided by x-ray computed tomography (CT) or ¹H-MRI (Woodhouse et al 2005). Accurate image registration is an essential requirement for such multimodality assessment of patients with lung disease. In this paper, we demonstrate an image acquisition and registration strategy for the fusion of ³He-MRI and x-ray CT and discuss its potential application to lung cancer treatment planning and monitoring.

Before radiotherapy treatment, many patients have ventilation or perfusion defects that impair effective lung function. The concept of functionally weighted treatment planning is to identify regions of defective lung and of viable functioning lung and to allow the radiotherapy inverse planning algorithm to apply different dose constraints to healthy and impaired lung tissue. This strategy has been applied using SPECT with the aim of reducing the incidence of treatment complications by modifying the dose distributions within the lung while maintaining tumour toxicity (Marks et al 1993, 1999, McGuire et al 2006, Lavrenkov et al 2007, Shioyama et al 2007).

For the information provided by ³He-MRI to be used in radiotherapy in a similar manner to SPECT, the MR images need to be accurately registered to the lung CT scans that are routinely used for treatment planning. The only previous work related to such ³He-MR image registration was a recently reported pilot study (Ireland et al 2007) in which the feasibility of acquiring ³He-MRI for non-small cell lung cancer (NSCLC) patients was demonstrated. While the study showed that ³He-MR images could be registered to radiotherapy CT for use in treatment planning, a number of changes to both the MR and CT protocol were identified that could enhance the registration accuracy. The key components required to make these changes have recently been realized. These include a new radiofrequency (RF) body coil, which has been designed, built and tested (de Zanche et al 2008), a 3D ³He-MR volume imaging methodology (Wild et al 2004) and the use of a 16-slice radiotherapy CT scanner. These advances have enabled the implementation of treatment position ³He-MR imaging and breath-hold planning CT. The aim of this study was to utilize the modified ³He-MRI and CT acquisition protocols and quantify the impact of breath hold on the accuracy of ³He-MRI to CT image registration.

2. Materials and methods

2.1. NSCLC patients

Patients with histologically confirmed NSCLC due to undergo CT for radiotherapy planning were recruited to the study by consultant oncologists. All patients gave written informed consent to participate, and the prospective study was approved by the Local Research Ethics Committee. Each patient underwent hyperpolarized ³He-MRI in addition to two x-ray CTs.

2.2. MR image acquisition

Hyperpolarized ³He-MR ventilation imaging was performed on a 1.5 T whole body Eclipse system (Philips Medical Systems, Cleveland, OH, USA), which was fitted with a second RF amplifier (2 kW, Analogic Corporation, Peabody, MA, USA) and a transmit–receive circuit tuned to 48.5 MHz for ³He (Wild et al 2004).

In the previous work (Ireland et al 2007), patients were imaged with a twin saddle quadrature transmit–receive RF coil (IGC Medical Advances, Milwaukee, WI, USA), which required the patients to be imaged supine with their arms down. More recently, a volume coil has been developed (de Zanche et al 2008) that provides ³He-MR images which can be acquired with the patients in treatment position (immobilized with their arms up). The coil design is an elliptical high pass birdcage that operates in quadrature mode. For ease of patient access, the coil splits into two halves and rests on the MR scanner patient bed. To mitigate interactions with the proton (¹H) RF body coil, the outer surface of the coil is screened. Tests on the coil indicate a higher B1 homogeneity and sensitivity in terms of image SNR; both are important considerations for high spatial resolution imaging of ³He in the lungs (de Zanche et al 2008).

The birdcage coil was used with a 3D acquisition sequence (Wild et al 2004) that consists of a low flip angle (θ = 4°), [96, 24] phase encodes in the [y, z] axes, 256 samples in the read encoding direction [x], 9.33 mm slice thickness with no gap, FOV = 42.6 cm, TE = 3.25 ms, TR = 6 ms and bandwidth = 31.25 kHz. Each image was interpolated to 256 × 256 pixels with 1.66 mm pixel size. The ³He gas (Spectra Gases, Huntingdon, UK) was polarized on site to approximately 30% by optical pumping with rubidium spin exchange apparatus (GE Healthcare, Princeton, NJ, USA). The investigators hold a license for ³He gas polarization as an Investigational Medicinal Product from the national regulatory body. In vivo imaging was performed during a single 16 s breath hold of a 500 ml ³He plus 500 ml N₂ mixture inhaled from a Tedlar bag (Jensen Inert Products, Coral Springs, FL, USA).

The ³He gas was inhaled from a starting point of functional residual capacity and no additional inhalation of room air was permitted. Blood oxygen saturation was monitored via an MRI compatible pulse oximeter throughout the procedure. Patients were coached in the breathing manoeuvre and were allowed a trial breath hold with a dummy bag containing room air prior to imaging.

2.3. CT image acquisition

On the same day as ³He-MR imaging, all patients underwent two transaxial CTs with a 16-slice Lightspeed CT (GE Healthcare, Princeton, NJ, USA). Initially, the patients were imaged in the conventional treatment planning position, which is supine with arms supported above the head while breathing freely. Subsequently, a second CT was acquired during a 530 ms inspiration breath hold performed with a 1 L Tedlar bag filled with room air that enabled simulation of the ³He-MRI breathing manoeuvre. CT imaging (120 kVp) was performed at 512 × 512 pixels per slice with pixel size 0.9766 mm and 2.5 mm slice thickness.

2.4. Image registration

Using in-house Matlab (MathWorks Inc., Natick, MA, USA) DICOM editing software, distance measurement DICOM tags of the ³He-MR images were scaled by a factor of 1.3 to compensate for the lower ³He gyromagnetic ratio than that of ¹H, and the Scanning Sequence tag (0018,0020) was modified to ‘EP’ to enable the ³He-MR images to be imported into the Eclipse treatment planning system (Varian Medical Systems, Palo Alto, CA, USA).

Once imported, the rigid control point (Zitová and Flusser 2003) registration tool implemented in Varian Eclipse was used to register the ³He-MRI to the planning CT. A median of 6 (range 4-7) anatomical landmarks (‘control points’) were used and the mean and maximum error of landmark accuracy were calculated. The registration was performed by two independent observers (both experienced consultant oncologists) who identified landmarks around the lung apex, mediastinum, main and segmental carina and the lung bases. Lung volumes including the trachea were automatically segmented from the CT using standard threshold levels. The ³He-MRI was segmented from the noise using a threshold set at three times the standard deviation of the background noise. In addition to anatomical landmarks, registration accuracy was quantified using an overlap coefficient (Ω), which is calculated as the proportion of the segmented ³He-MR slice volume (V_MRI) that intersects the segmented CT lung slice volume (V_CT) expressed as a percentage of V_MRI:

$$\Omega =100\times \frac{V_{\mathrm{MRI}}\cap V_{\mathrm{CT}}}{V_{\mathrm{MRI}}}$$

where the higher the overlap value the better the registration. Values for the overlap are calculated from the entire MR and CT lung volumes.

As the ³He-MR ventilated volume may be different than the CT defined volume due to the possibility of ventilation defects detected in the functional images, the overlap coefficient is a more appropriate measure of ³He-MR to CT registration accuracy rather than, for example, the intersection-union ratio, which might have been used if the two volumes were expected to be similar.

After an interval of 3 months, the observers repeated the registration procedure and both inter- and intra-observer variability were assessed.

2.4.1. Effect of image resolution

The ³He-MRI is acquired at a lower resolution than the planning CT. Once registered, within the treatment planning system the MRI is displayed overlaid with the planning CT. To evaluate the limit on image registration accuracy that results from the mismatch in image resolution alone, a low-resolution inspiration planning CT was created in Matlab using cubic decimation to the dimensions of the ³He-MRI. The low-resolution CT was imported back into the treatment planning system and registered to the original high-resolution image volume. CT lung volumes were automatically segmented for both low- and high-resolution image sets. An overlap coefficient was then calculated from the low- and high-resolution lung volumes. This procedure was conducted for all of the nine patients to provide a theoretical limit of overlap performance taking into account the differing image resolution.

2.5. Statistical analysis

To test the statistical significance of differences between the registration accuracy of ³He-MRI to conventional and breath-hold planning CT, analysis of variance tests were used with the mean error, maximum error and overlap coefficient as a dependent variable, the observers and CT type (free breathing and breath hold) as fixed factors and the subjects as a random factor.

Inter-observer variability was evaluated using analysis of variance applied to each parameter and CT type, with the observers as a fixed factor and the subjects as a random factor. Intra-observer variability was evaluated using one way analysis of variance applied to each parameter and CT type, with the subjects as a random factor. The Bland and Altman repeatability coefficient was used to summarize the inter- and intra-observer agreements (Bland and Altman 1999).

Statistical analysis was performed using SPSS for Windows (SPSS Version 15.0; Chicago, IL, USA). A p value of less than 0.05 was considered statistically significant. Data normality was verified with the Kolmogorov-Smirnov test. Bland–Altman plots were used to check the independence of the mean and difference values between observers for each parameter to ensure that a logarithmic transformation was not required (Bland and Altman 1999).

3. Results

3.1. NSCLC patients

Between April 2006 and March 2007, ten NSCLC patients underwent ³He-MRI and CT with the new imaging equipment and protocol. There were 7 men and 3 women with mean age 71 years and age range 56-85 years. None of the patients had been previously treated with radiotherapy of the thorax. Patients had WHO performance status of 0-2 and forced expiratory volume (FEV1) > 1.5 L. During ³He-MR imaging, no clinically significant decrease in blood oxygen saturation was observed and no ill effects were reported following inhalation of the ³He/N₂ gas mixture. Due to a poor intake of gas, the ³He-MRI for one patient failed to produce ventilation images of sufficient quality for further registration analysis; therefore, nine MR scans were available for evaluation.

3.2. Image registration

3.2.1. ³He-MR and planning CT

Calculated from the mean values for the two observers, the ³He-MRI and conventional planning CT were registered with an overlap (mean ± SD) 59.8 ± 9.0%, mean landmark error 1.25 ± 0.60 cm and maximum landmark error 2.04 ± 1.06 cm (table 1). For the breath-hold CT, the overlap was 82.9 ± 4.2% (p < 0.001), mean landmark error 0.75 ± 0.24 cm (p = 0.002) and maximum landmark error 1.20 ± 0.36 cm (p = 0.004), which are all significantly better than for the registration of ³He-MRI to conventional CT.

Table 1.

Comparison of registration accuracy results (mean ± SD) for ³He-MRI to free-breathing CT and breath-hold CT

	3He to free-breathing CT	3He to breath-hold CT	p value	95% confidence interval for mean difference	Effect size (mean difference)
Mean landmark error	1.25 ± 0.60 cm	0.75 ± 0.24 cm	p = 0.002	−0.81 to −0.20 cm	−0.50 cm
Max landmark error	2.04 ± 1.06 cm	1.20 ± 0.36 cm	p = 0.004	−1.34 to −0.30 cm	−0.84 cm
Overlap	59.8 ± 9.0%	82.9 ± 4.2%	p < 0.001	18.3 to 27.9%	23.1%

Example ³He-MR images registered to conventional free-breathing CT and breath-hold CT are shown for two patients in figure 1.

Figure 1.

Two examples of registered coronal hyperpolarized ³He-MRI (colour) displayed fused with radiotherapy treatment planning conventional, free-breathing CT (left) and inspiration breath-hold CT (right).

The intra-observer and inter-observer variability are summarized in table 2. The analysis of variance for inter-observer variability demonstrated no significant difference between the two observers for each dependent variable or CT type (p ≥ 0.38 for mean error, maximum error and overlap coefficient). There was no consistent pattern for inter-and intra-observer repeatability coefficient, as four values were smaller in the breath-hold case and five for the free-breathing data.

Table 2.

Repeatability coefficients for inter- and intra-observer variability

	Breath-hold CT			Free-breathing CT
	Mean	Max	Overlap	Mean	Max	Overlap
Intra-observer 1	0.90 cm	1.65 cm	4.9%	1.76 cm	2.91 cm	4.0%
Intra-observer 2	0.87 cm	1.09 cm	5.7%	0.70 cm	1.73 cm	1.6%
Inter-observer	0.33 cm	0.92 cm	1.4%	0.97 cm	0 cm	0.9%

3.2.2. Effect of image resolution

A pseudo-planning CT was used to calculate a maximum limit of overlap taking into account the difference in image resolution between the original MR and CTs. Low-resolution CT was registered and overlaid with its original high-resolution version to give a mean error of 0.28 ± 0.10 cm, of maximum error of 0.49 ± 0.21 cm and an overlap value 96.2 ± 2.0%. Dividing the mean overlap for ³He-MRI by this value provides a normalized overlap for the image resolution of 86.2% for the inspiration CT and 62.2% for the free-breathing CT.

4. Discussion

4.1. Image registration

Related studies have investigated the optimal acquisition protocol for SPECT and CT (Gutman et al 2005, Utsunomiya et al 2005, Suga et al 2007), and image registration of perfusion MRI and ³He-MRI using controlled gas administration in pigs (Hong et al 2005, Rizi et al 2003). However, our work is the first to consider optimizing an image acquisition protocol for image registration of in vivo ³He-MRI to CT.

Previously, patients have been imaged with a twin saddle coil for MR imaging and a single-slice scanner for CT (Ireland et al 2007). However, as the tight-fitting coil jacket impeded arm movement, ³He-MRI with the original coil required imaging to be performed with arms down by the side of the patient during a breath hold of 1L of gas. This is a different imaging setup than is commonly used for conventional treatment planning which requires a free-breathing CT with the patient immobilized with their arms up behind the head. In addition, free-breathing treatment planning CT often contains artefacts at the base of the lung. These differences in patient positioning and image acquisition inevitably lead to errors in image registration when a rigid algorithm is applied.

Rather than using nonrigid image registration, which is difficult to apply and validate for the chest (Gutman et al 2005), lung image registration can be improved by using an imaging strategy that minimizes the differences in breathing protocol and body position between the MR and CT acquisitions. Consistent arm position was provided by an elliptical birdcage ³He body RF coil (de Zanche et al 2008) that made it possible to acquire higher quality ³He-MR images using a 3D volume sequence (Wild et al 2004) in treatment position, and the installation of a 16-slice CT scanner that enabled a breath-hold CT to be acquired within a 5 s interval. This acquisition protocol enabled the thoracic image registration to be performed accurately with commercially available rigid registration software.

The statistical analysis outlined in section 3.2.1 demonstrates that the repeatability of image registration of ³He-MRI to breath-hold CT and free-breathing CT is similar. For each parameter of registration accuracy (table 1), analysis of variance provides p values less than 0.004, 95% confidence intervals that do not contain zero and effect sizes that are large. Therefore, it is reasonable to conclude that there is a statistically and clinically significant difference in ³He-MRI registration accuracy between the free-breathing and breath-hold CT.

4.2. Treatment planning

Image registration is an essential step towards implementing functionally weighted radiotherapy treatment planning. The potential for modifying lung cancer treatment plans with functional information has been investigated for SPECT (Marks et al 1993, 1999, McGuire et al 2006, Lavrenkov et al 2007, Shioyama et al 2007), 4D-CT (Yaremko et al 2007) and ³He-MRI (Ireland et al 2007). Although free-breathing CT has been the most commonly used method of planning lung radiotherapy treatment, there may be benefits from using breath-hold techniques (Keall et al 2006, Duggan et al 2007, Nakamura et al 2007). Image registration provided by the breath-hold image acquisition protocol evaluated here could be beneficial in the implementation of ³He-MRI-influenced treatment planning. Similarly, registration of ³He-MRI and CT could be useful for comparing pre- and post-treatment images for treatment monitoring and long-term follow-up.

4.3. Other applications

As demonstrated in other areas of radiology such as SPECT/CT (Suga et al 2007), accurate fusion of anatomical and functional data can assist image interpretation. Although the current study has focused on the application of ³He-MRI to lung cancer patients, MR ventilation imaging is more established in the study of other lung diseases (van Beek et al 2004). In such cases, registration of ³He-MRI to CT would enable pulmonary ventilation to be assessed against the underlying anatomical CT structure which serves as the gold standard. Potential applications include the analysis of serial studies conducted with ³He-MRI. When the reduction of radiation dose is important, it may also be appropriate to assess lung disease with ³He-MRI in conjunction with anatomical images provided by ¹H-MRI rather than CT. For example, an approach similar to that described in this work could also be applied to the registration of ³He-MRI to ¹H-MRI with the aim of further improving quantification accuracy of the ratio of ventilated to total lung volume (Woodhouse et al 2005).

Furthermore, if ³He-MRI is to become a complementary or superior lung imaging technique in a clinical setting, then quantitative metrics of its performance are needed to allow comparison against existing imaging modalities (McMahon et al 2006). These quantitative comparisons will require accurate spatial registration.

4.4. Limitations

One limitation with matching ³He-MRI to CT is the difference in image resolution. The impact of this on image registration can be estimated from constructing pseudo-CT data sets at the resolution of the MRI. Reproducibility of patient breath hold is another limiting factor that requires further investigation for ³He-MRI though reproducibility of breath-hold CT has been shown to be feasible for selected groups of NSCLC patients (Keall et al 2006, Duggan et al 2007, Nakamura et al 2007).

5. Conclusions

This study has demonstrated a method of registering hyperpolarized ³He-MRI to x-ray CT. Compared with free-breathing CT, application of an inspiration CT acquired at a breath hold similar to that used in the MRI provides a statistically significant improvement of ³He-MRI to CT image registration accuracy as measured by anatomical landmarks and an overlap coefficient.