Feasibility of MRI-based reference images for image-guided radiotherapy of the pelvis with either cone-beam computed tomography or planar localization images

Abstract

Purpose. This study introduces methods to conduct image-guided radiotherapy (IGRT) of the pelvis with either cone-beam computed tomography (CBCT) or planar localization images by relying solely on magnetic resonance imaging (MRI)-based reference images.

Material and methods. Feasibility of MRI-based reference images for IGRT was evaluated against kV CBCT (50 scans, 5 prostate cancer patients) and kV & MV planar (5 & 5 image pairs and patients) localization images by comparing the achieved patient position corrections to those obtained by standard CT-based reference images. T1/T2*-weighted in-phase MRI, Hounsfield unit conversion-based heterogeneous pseudo-CT, and bulk pseudo-CT images were applied for reference against localization CBCTs, and patient position corrections were obtained by automatic image registration. IGRT with planar localization images was performed manually by 10 observers using reference digitally reconstructed radiographs (DRRs) reconstructed from the pseudo-CTs and standard CTs. Quality of pseudo-DRRs against CT-DRRs was evaluated with image similarity metrics.

Results. The SDs of differences between CBCT-to-MRI and CBCT-to-CT automatic gray-value registrations were ≤ 1.0 mm & ≤ 0.8° and ≤ 2.5 mm & ≤ 3.6° with 10 cm diameter cubic VOI and prostate-shaped VOI, respectively. The corresponding values for reference heterogeneous pseudo-CT were ≤ 1.0 mm & ≤ 0.7° and ≤ 2.2 mm & ≤ 3.3°, respectively. Heterogeneous pseudo-CT was the only type of MRI-based reference image working reliably with automatic bone registration (SDs were ≤ 0.9 mm & ≤ 0.7°). The differences include possible residual errors from planning CT to MRI registration. The image similarity metrics were significantly (p ≤ 0.01) better in agreement between heterogeneous pseudo-DRRs and CT-DRRs than between bulk pseudo-DRRs and CT-DRRs. The SDs of differences in manual registrations (3D) with planar kV and MV localization images were ≤ 1.0 mm and ≤ 1.7 mm, respectively, between heterogeneous pseudo-DRRs and CT-DRRs, and ≤ 1.4 mm and ≤ 2.1 mm between bulk pseudo-DRRs and CT-DRRs.

Conclusion. This study demonstrated that it is feasible to conduct IGRT of the pelvis with MRI-based reference images.

Computed tomography (CT) is conventionally applied for external radiotherapy planning (RTP). Magnetic resonance imaging (MRI) is increasingly adopted for RT target delineation because of multiple imaging modes in MRI and superior soft tissue contrast compared to CT [1]. Imaging in treatment position, the geometrical distortion and the lack of electron density information pose major challenges for relying solely on MR images throughout the external RTP process. By enabling dose planning and image guidance with reference MR images, the planning CT imaging becomes redundant, thus avoiding co-registration uncertainty stemming from the adoption of two imaging modalities into RTP. Furthermore, exclusion of planning CT imaging saves hospital resources and reduces healthy tissue exposure to the ionizing radiation.

MRI can be performed following the general requirements of RT patient positioning by equipping the MR scanner with such as a laser localization system, flat carbon fibre table top, MR-compatible immobilization and receiver coil frames [2]. Recent studies have shown that the geometric errors in MR images can be within acceptable levels for RTP of the pelvis [2–9]. The lack of electron density information in MR images can be overcome by generating so called pseudo-CT images [8–22]. The first pseudo-CT construction methods of the pelvis were based on bulk density assignments; either by representing the whole patient body as water equivalent or by setting an additional separate Hounsfield unit (HU) for a bone segment [9–14]. Recently introduced dual model HU conversion technique enabled generation of heterogeneous pseudo-CTs by transforming intensity values of a T₁/T₂*-weighted in-phase MR image into HUs by separate conversion models within and outside of bone segment [21]. Previous investigations have shown that dose calculation accuracy in pseudo-CTs of prostate cancer patients is roughly between 1% and 4% compared to that in standard CTs depending on the applied pseudo-CT construction method [8–14,18–21].

Previous studies have suggested that also image-guided RT (IGRT) with two-dimensional (2D) planar localization images could be conducted relying merely on MRI by reconstructing the reference digitally reconstructed radiographs (DRR) from the pseudo-CTs [14–19]. However, no published papers have reported the accuracy of MRI-based patient position verification by following a standard clinical workflow such as achieving patient position corrections by registering DRRs and localization images. It would be of particular value to examine feasibility of MRI-based reference images for IGRT with state-of-the-art techniques, such as through automatic registration between the planning images and cone-beam CT (CBCT) localization images [23–25].

This study aims to evaluate whether IGRT of the pelvis could be performed with MRI-based reference images; i.e. T₁/T₂*-weighted in-phase MR, HU conversion-based heterogeneous pseudo-CT, bulk pseudo-CT, and DRRs reconstructed from each of these pseudo-CTs. The images are tested for two established IGRT methods that are currently being used; automatic localization with CBCTs and manual localization with 2D planar images, and the obtained patient position corrections are compared to those achieved with the reference images in current practise; the standard CT and the DRRs reconstructed from the CT.

Material and methods

Patients

This study included 15 prostate cancer patients whose RTP had been conducted applying two imaging modalities; MRI for target delineation, and CT for dose calculation and image guidance. The study did not have any impact on treatment of these patients. The patients were selected retrospectively from the patient database of HUCH Cancer Center in order to form three five-patient groups, each group having different localization imaging method; kV CBCT, kV planar or MV planar localization. The clinical target volume (CTV) of each patient included the prostate and the seminal vesicles.

MRI and CT

CT and MRI were performed during the same day for each patient following clinical protocols of HUCH Cancer Center and general requirements of RT patient positioning [2,21]. The imaging parameters are presented in Supplementary material to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197. The acquired CT and T₁/T₂*-weighted in-phase MR image series of each patient were co-registered by a commercial medical image processing software (rigid registration relying on mutual information, MIM Software Inc., version 5.4, Cleveland, OH, USA), and both image series were set in the same coordinate system. Moreover, the DICOM headers of MR images were modified (set as CT) to enable using MRI-based images with RTP systems.

Construction of pseudo-CTs and DRRs

The pseudo-CTs were constructed based on the tissue presentation in the T₁/T₂*-weighted in-phase MR images. The bulk pseudo-CTs were generated by assigning HU value of 330 for the entire pre-contoured bone segment to represent an average electron density value of bones and by assigning all the tissues outside bone segment as water equivalent [10,19]. The heterogeneous pseudo-CTs were constructed individually for each patient by the dual model HU conversion technique (detailed description in previous studies) [8,21]. These pseudo-CTs represent bony (e.g. cortical bone, bone marrow and spongy bone) and soft tissues (e.g. muscle, fat, prostate, rectal wall and urine) with average local uncertainties of 99 HUs and 11 HUs, respectively [21]. Figure 1 includes examples of the obtained pseudo-CTs.

Figure 1. Example transversal slices of 3D reference images for IGRT with CBCT localization; a T₁/T₂*-weighted in-phase MR image (A), a heterogeneous pseudo-CT that was constructed from the MR image by the dual model HU conversion technique (B), a bulk pseudo-CT (C), and a standard CT (D). Contrast in the images is not comparable (different windowing properties and scaling).

DRRs were reconstructed according to the clinical protocol of HUCH Cancer Center reconstructing the 2D pelvic bony tissue representation in posterior-anterior (PA) and in right-left (RL) directions from 3D image information of HUs over 100. Heterogeneous pseudo-CT, bulk pseudo-CT and standard CT of each patient were applied for the DRR reconstruction separately. Figure 2 illustrates examples of the obtained DRRs.

Figure 2. Example reference DRRs reconstructed from different 3D image series; the heterogeneous pseudo-DRRs (A) and (B), the bulk pseudo-DRRs (C) and (D), and the CT-DRRs (E) and (F) (PA and RL, respectively).

IGRT with localization CBCT and automatic position verification method

The 3D localization images were obtained by a kV CBCT system that is integrated with the gantry of linear accelerator (x-ray volumetric imaging [XVI^®] integrated with Elekta Axesse^®, Elekta AB, Stockholm, Sweden) [23,26,27]. The imaging parameters are presented in Supplementary material. The XVI^® software provides patient position corrections by automatic gray-value or bone rigid registration (based on gray level correlation ratio or chamfer matching algorithm, respectively) between localization CBCT and reference CT within user specified volume-of-interest (VOI, either cubic-shaped ‘ClipBox’ or contour-shaped ‘Mask’) [24–30]. Visual inspection is required to detect potential outliers (i.e. the registrations that have clearly failed, but however, the system have reported position corrections) and to verify the registration result before performing patient position corrections [24–30].

In this study, the heterogeneous pseudo-CT, the bulk pseudo-CT and the T₁/T₂*-weighted in-phase MR images were assigned for reference images in addition to the current practise standard reference CT (Supplementary Table I to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197, presents the summary of investigated reference images with the IGRT methods used). Examinations were conducted with 50 localization CBCTs (10 CBCTs of each patient). The bone- and gray-value registrations were performed separately. The investigations were conducted in 3D and in 6D with either a 10 cm diameter ‘ClipBox’ VOI or a CTV + 5mm ‘Mask’ VOI. The adopted ‘Mask’ size is considered optimal for prostate localization [24,25]. The applied ‘ClipBox’ size was chosen to cover also the tissues surrounding the moving organs. The position corrections obtained by automatic registration between each MRI-based reference image and localization CBCT image were compared to those obtained by applying standard reference CT image. In this study, the contribution of major registration outliers to the results was aimed to decrease as follows; the registration differences were calculated with all the registrations, and based on these results, the registrations with > 3 SD difference in any direction were excluded, and the registration differences were recalculated without these (> 3 SD) major outliers. The recalculated averages and SDs were reported.

IGRT with 2D planar localization images

The feasibility of reference heterogeneous- and bulk pseudo-DRRs for IGRT with planar 100 kV or 6 MV localization images (On-board Imager^® or PortalVision^®, respectively, Varian Medical Systems, Helsinki, Finland) was evaluated by manual 2D image registration (Intuity^®, XVI^® R4.5 software, Elekta AB, Stockholm, Sweden) relying solely on bony tissue presentation in the images (Supplementary Table I to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197). The examinations were conducted applying one pair of localization images (PA and RL) of each patient. Ten clinical physicists performed registrations in three degrees-of-freedom between the localization images and each of the reference DRRs. The observers were blind to the achieved patient position corrections. The position corrections obtained with the pseudo-DRRs were compared to those achieved with the standard CT-DRRs. Statistical hypothesis tests (paired t-test and two-tailed F-test) were conducted to evaluate whether the position corrections with either heterogeneous- or bulk pseudo-DRRs were significantly (p ≤ 0.05) more similar to those achieved with CT-DRRs than with the other. Additionally, the quality of the pseudo-DRRs was evaluated with three image similarity metrics to quantify whether the image quality in either type of pseudo-DRRs was significantly more similar to CT-DRRs than in the other (detailed description in Supplementary material to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197) [31].

Results

Table I presents the differences in automatic patient position corrections between those obtained by standard reference CT and those achieved either with the heterogeneous pseudo-CT or with the T₁/T₂*-weighted in-phase MR image. The reported averages and SDs were resulted from recalculations by omitting the major registration outliers (> 3 SD, with each registration method ≤ 10% of registrations). Supplementary Figure 1 to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197 illustrates the differences with 3D automatic gray-value registrations. Supplementary Figure 2 to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197 shows the percentages of registrations passing certain difference criteria with 6D automatic gray-value registrations. The averages and SDs of registration differences were mostly sub-mms and sub-degrees with the ‘ClipBox’ VOI. With this VOI the maximum position correction differences of all the registrations between CT and heterogeneous pseudo-CT reference images were 2.0 mm (gray-value 3D), 1.7 mm and 1.1° (gray-value 6D), and 1.6 mm and 1.3° (bone 6D), and between CT and MR reference images 4.0 mm (gray-value 3D) and 3.5 mm and 1.6° (gray-value 6D). With the prostate-shaped ‘Mask’ VOI the average registration differences were mainly sub-mms and sub-degrees with SDs ranging up to over 2 mm and 3°. The maximum position correction differences were > 20 mm and > 20°. The bony tissue presentation in MR images was not sufficient for the bone registration (the system reported lack of sufficient tissue and that registrations may be inaccurate). Warnings by the system were detected also with the bulk pseudo-CTs, and thus any of these registration results were not reported.

Table I. Differences in automatic patient position corrections between those obtained by MRI-based reference images and those achieved with standard reference CT images registered against localization CBCTs as mean ± SD (recalculated by excluding major registration outliers, i.e. > 3 SD).

		CT vs. CBCTn – Pseudo^a vs. CBCTn			CT vs. CBCTn – MRI^b vs. CBCTn
Registration method	Direction	Included registrations	Translation (mm)	Rotation (degrees)	Included registrations	Translation (mm)	Rotation (degrees)
Gray-value 6D with ‘ClipBox’	RL IS PA	100%	0.3 ± 1.0 0.2 ± 0.4 −0.3 ± 0.7	0.1 ± 0.4 −0.1 ± 0.7 0.0 ± 0.4	92%	0.6 ± 1.0 −0.6 ± 0.6 −1.4 ± 0.7	0.0 ± 0.5 −0.4 ± 0.8 −0.3 ± 0.4
Gray-value 6D with ‘Mask’	RL IS PA	92%	0.8 ± 1.4 −0.9 ± 1.6 −0.4 ± 2.2	1.1 ± 3.3 0.1 ± 2.4 −0.5 ± 2.3	90%	0.5 ± 1.3 −0.8 ± 1.6 −0.1 ± 2.5	1.6 ± 3.6 0.3 ± 2.6 0.2 ± 2.0
Gray-value 3D with ‘ClipBox’	RL IS PA	98%	0.2 ± 0.9 0.2 ± 0.5 −0.3 ± 0.6	– – –	90%	0.5 ± 1.2 −0.7 ± 0.7 −1.5 ± 0.6	– – –
Gray-value 3D with ‘Mask’	RL IS PA	92%	0.5 ± 1.3 −1.2 ± 1.9 −0.4 ± 2.0	– – –	90%	0.5 ± 1.5 −1.1 ± 1.6 −0.3 ± 2.4	– – –
Bone 6D with ‘ClipBox’	RL IS PA	100%	0.3 ± 0.9 0.1 ± 0.4 −0.3 ± 0.7	0.1 ± 0.3 −0.1 ± 0.7 −0.1 ± 0.3	–	– – –	– – –

^aThe heterogeneous pseudo-CT image series was constructed by the dual model HU conversion technique; ^bThe T₁/T₂*-weighted in-phase MR image series.

Table II presents the differences in manual patient position corrections between those obtained by the reference pseudo-DRRs and those achieved with the reference CT-DRRs. The differences were at their highest in PA direction with means± SDs of −0.3 ± 1.0 mm and 0.3 ± 1.7 mm (kV and MV, respectively), between heterogeneous pseudo-DRRs and CT-DRRs, and −0.5 ± 1.4 mm and −0.3 ± 2.1 mm (kV and MV, respectively), between bulk pseudo-DRRs and CT-DRRs. All the applied image similarity metrics were significantly better in agreement between heterogeneous pseudo-DRRs and CT-DRRs than between bulk pseudo-DRRs and CT-DRRs (p ≤ 0.01).

Table II. Differences in manual patient position corrections between those obtained by reference pseudo-DRRs and those achieved with reference CT-DRRs registered against planar localization images as mean ± SD (range).

Localiz. images	Transl. direct.	DRRCT^a vs. planarn − DRRHETER.^b vs. planarn (mm)	DRRCT^a vs. planarn − DRRBULK^c vs. planarn (mm)	t- and F-tests (pt and pF)
kV	RL	0.1 ± 0.5 (−0.8–1.5)	0.2 ± 0.6 (−1.0–1.4)	0.55 and 0.36
	IS	0.2 ± 0.9 (−1.6–2.8)	0.2 ± 1.0 (−1.9–2.7)	0.72 and 0.14
	PA	−0.3 ± 1.0 (−2.5–1.5)	−0.5 ± 1.4 (−3.9–4.3)	0.33 and 0.01
MV	RL	−0.2 ± 0.6 (−1.5–1.7)	−0.1 ± 0.7 (−2.2–1.2)	0.06 and 0.23
	IS	0.0 ± 1.1 (−2.9–3.0)	0.1 ± 1.2 (−3.6–2.2)	0.25 and 0.40
	PA	0.3 ± 1.7 (−5.2–5.1)	−0.3 ± 2.1 (−6.4–6.1)	0.04 and 0.12

^aDRR_CT was reconstructed from standard CT image series; ^bDRR_HETER. was reconstructed from heterogeneous pseudo-CT image series; ^cDRR_BULK was reconstructed from bulk pseudo-CT image series.

Discussion

The study reached its primary goal by demonstrating that it is feasible to apply MRI-based reference images for IGRT of the pelvis. Consequently, it has been shown that the entire external RTP process of prostate cancer patients can be conducted by relying solely on MRI [2,8,21]. It is essential to recognize that the reported registration differences include uncertainties stemming from ∼1 mm-pixel sizes, 1.2–2.5 mm-slice thickness in the reference images, residual errors from planning CT to MR image registration (causing at least partly the systematic offsets for each patient, Supplementary Figure 1 to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197), inherent registration uncertainty in XVI^® software (reported previously for CT to CBCT registrations), movement of tissues, and limited number of test patients [24–30,32–34]. Furthermore, the reported patient positioning uncertainties (i.e. registration differences) with MRI-based reference images were analyzed by assuming that the current practise registration with the standard CT can be regarded as reference for optimal patient position verification. Hence, it is likely that the actual registration uncertainty with MRI-based reference images is smaller than reported in this study. The observers considered that the automatic registration degree-of-success (i.e. requiring no further re-adjustments based on visual inspection) was roughly similar with any of the applied reference images, and approximately similar to that (65%) quantified in previous studies for CT to CBCT registration with the prostate-shaped VOI [24,25]. With this VOI the registration accuracy was strongly influenced by organ motion and rectal gas thus causing also major registration outliers as recognized in previous studies (previously reported uncertainties for CT to CBCT registration with the prostate-shaped VOI; mean± SDs up to −1.1 ± 2.9 mm and −0.3 ± 2.4°) [24,25,28]. For example, in one of the adopted reference CTs in the current study there was a substantial rectal gas pocket causing registration uncertainty with any of the localization CBCTs when applying the prostate-shaped VOI. The rectal gas is presented as tissue in the heterogeneous pseudo-CTs that can minimize the registration uncertainty stemming from anatomical changes in the rectum [21,24]. Additionally, the prostate position verification by fiducial markers could be conducted with the pseudo-CTs by including an additional marker visualizing sequence into the MRI workflow and subsequently inserting marker segments with a representative HU into the pseudo-CT [2].

Table I, Supplementary Figures 1 and 2 to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197 illustrate that the heterogeneous pseudo-CT as reference image for localization CBCTs by automatic patient position verification works precisely and reliably both with the gray-value and with the bone registration. The T₁/T₂*-weighted in-phase MR reference images can also be concluded feasible for IGRT with automatic gray-value registration against CBCTs. However, the registration with the ‘ClipBox’ VOI was not as reliable with MR images as it was with CTs or heterogeneous pseudo-CTs. The automatic registration was not functioning in eight of the CBCT to MR image registrations. Nevertheless, after few mm ‘ClipBox’ re-adjustments the automatic registrations were functioning, and these registration results were included into the study. In these cases the results were not optimal (i.e. the registration accuracy could have been improved by manual re-adjustments), thus causing the maximum reported position correction differences between reference CT and MR image. The observers did not detect any particular reasons for the failures nor for the systematic shift of over 1 mm to posterior direction with the ‘ClipBox’ VOI (Table I, Supplementary Figures 1 and 2 to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197). It is likely that the malfunctions were due to the fact that the registration cost functions were not tailored for CBCT to MR image registration. Therefore, it is reasonable to underline the importance of mandatory visual inspection before carrying out the patient position corrections. The excellent soft tissue contrast in MR images can aid this task. However, the accuracy of visual inspection is limited more by the image quality of CBCTs than that in the applied heterogeneous reference images. If adopting bulk pseudo-CTs for MRI-based RTP, the image guidance could be done by relying only on the bone outlines in the images. The automatic bone registration may be helpful for this task, because occasionally the CBCT-to-bulk pseudo-CT registrations were functioning without any warnings by the system. Future work could aim to determine the needed level of tissue presentation in the reference images for reliable automatic registration, and additionally evaluate whether the bone segmentation accuracy in pseudo-CT construction influences on the patient position verification accuracy.

The manual patient position verification examinations revealed that by adopting the heterogeneous pseudo-CTs for RTP it is feasible to conduct IGRT also by relying on 2D bony tissue localization with the pseudo-DRRs. The manual registration uncertainty (between 10 observers) with the heterogeneous pseudo-DRRs was similar to that with the CT-DRRs, and thus the reported differences could have resulted from registration uncertainties with either of the DRRs. Moreover, according to Table II, the pseudo-DRRs did not cause relevant systematic errors for patient positioning. The position verification was possible also with the bulk pseudo-DRRs, but a common challenge for the observers was to detect sufficient bony tissue information from lateral bulk pseudo-DRRs, because of the lacks of typical contrast variations in the images. This was revealed also with the image similarity comparisons showing significantly better agreement between the heterogeneous pseudo-DRRs and CT-DRRs than between the bulk pseudo-DRRs and CT-DRRs. However, the effect of pseudo-DRR image quality on the positioning accuracy was mainly insignificant (Table II). According to Table II and received feedback from the observers, the localization image quality and FOV especially with the MV planar images considerably contributed to registration uncertainty.

It is important to recognize that the used reference images in this study were obtained by particular equipment and methods. Any changes to MR platform, sequence parameters or pseudo-CT and DRR construction techniques may result in different patient position verification results. Feasibility of the applied images needs to be verified before introducing any reference MRI, pseudo-CT or pseudo-DRR for image guidance. Further studies could aim to quantify the absolute patient positioning accuracy in reference MRI-based IGRT without comparisons to standard CT-based workflow, with different MRI sequences, with large number of test patients, and in different treatment sites.

Supplementary material available online

Supplementary Table I, Figure 1 and 2 to be found online at http://informahealthcare.com/doi/abs/10.3109/0284186X.2014.958197.

Acknowledgments

The authors acknowledge Kevin Brown for his valuable guidance during the project. The authors want to thank also the physicists Erna Kaleva, Hanna Koivunoro, Antti Kulmala, Juha Peltonen, Jean-Manuel Torrès, Laura Tuomikoski and Elisa Ålander of HUCH for their contribution to the project. The work was financially supported by Elekta Limited.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.