Adaptive image guided brachytherapy for cervical cancer: A combined MRI-/CT-planning technique with MRI only at first fraction

Abstract

Purpose

To investigate and test the feasibility of adaptive 3D image based BT planning for cervix cancer patients in settings with limited access to MRI, using a combination of MRI for the first BT fraction and planning of subsequent fractions on CT.

Material and methods

For 20 patients treated with EBRT and HDR BT with tandem/ring applicators two sets of treatment plans were compared. Scenario one is based on the “gold standard” with individual MRI-based treatment plans (applicator reconstruction, target contouring and dose optimization) for two BT applications with two fractions each. Scenario two is based on one initial MRI acquisition with an applicator in place for the planning of the two fractions of the first BT application and reuse of the target contour delineated on MRI for subsequent planning of the second application on CT. Transfer of the target from MRI of the first application to the CT of the second one was accomplished by use of an automatic applicator-based image registration procedure. Individual dose optimization of the second BT application was based on the transferred MRI target volume and OAR structures delineated on CT.

DVH parameters were calculated for transferred target structures (virtual dose from MRI/CT plan) and CT-based OAR.

The quality of the MRI/CT combination method was investigated by evaluating the CT-based dose distributions on MRI-based target and OAR contours of the same application (real dose from MRI/CT plan).

Results

The mean difference between the MRI based target volumes (HR CTV_MRI2) and the structures transferred from MRI to CT (HR CTV_CT2) was −1.7 ± 6.6 cm³ (−2.9 ± 20.4%) with a median of −0.7 cm³.

The mean difference between the virtual and the real total D90, based on the MRI/CT combination technique was −1.5 ± 4.3 Gy EQD2. This indicates a small systematic underestimation of the real D90.

Conclusions

A combination of MRI for first fraction and subsequent CT based planning is feasible and easy when automatic applicator-based image registration and target transfer are technically available. The results show striking similarity to fully MRI-based planning in cases of small tumours and intracavitary applications, both in terms of HR CTV coverage and respecting of OAR dose limits. For larger tumours and complex applications, as well as situations with unfavourable OAR topography, especially for the sigmoid, MRI based adaptive BT planning remains the superior method.

Magnetic resonance imaging (MRI) based brachytherapy (BT) allows appropriate delineation of target volumes for brachytherapy treatment planning [1,2]. Several studies could provide detailed dosimetric data for target and organs at risk (OAR) [3–9]. The definition of the High Risk Clinical Target Volume (HR CTV) is often used as the major basis for treatment planning. Proper dose shaping, i.e., escalating the dose for MRI based target volumes while keeping it reasonably low for OAR, seems to be the major factor in successfully improving local control and minimizing late side effects [9,10]. However, the use of an MRI for each individual brachytherapy fraction with the applicator in place is in many countries and centres difficult to organize and is expensive. There is evidence that the use of computed tomography (CT) only for brachytherapy target volume definition is inferior to MRI and introduces major uncertainties in dose assessment and therefore optimal dose planning [11]. The relevant question is, if there are methods, which achieve a good compromise between the full MRI approach with individual MRI for each fraction and the use of CT only. An achievable solution might be the use of at least one MRI for the first fraction and the use of CT for subsequent fractions or implantations, respectively. This paper applies a clinically feasible method to combine the MRI based target volume contour obtained for the first fraction with CT datasets of subsequent fractions, using an automatic, applicator-based co-registration technique. The method is based on the following hypotheses, which are tested within a retrospective treatment planning study.

The MRI based HR CTV contour for the first fraction can be used as an appropriate surrogate for the HR CTV to be used when subsequent fractions are scanned with CT for treatment planning.

The applicator and the HR CTV are linked together in a reproducible way. This implies that after re-insertion of an identical tandem/ring applicator the geometrical relation between HR CTV and applicator will be the same.

The basis for the method introduced here is the use of an applicator-based automatic registration between MRI and CT datasets with a tandem/ring applicator in place. By now most planning systems can only perform applicator based fusion by manual placement of landmarks on the applicator. An automatic procedure, based on the known outer surface of the applicator seems to be clinically feasible and practical. Followed by transfer of the MRI-based contour to the CT dataset and OAR contouring directly on CT, a fully contoured 3D dataset for further adaptive treatment planning is achieved.

This hypothesis is tested by simulating its use and comparing the dosimetric results against the gold standard, which is defined as an individual MRI for each single implantation/fraction.

Material and methods

Patients and application

Our standard treatment schedule consists of four high dose rate (HDR) BT fractions which are given in two applications in consecutive weeks [9,12]. For each fraction a D90 ⩾ 7 Gy is prescribed to the HR CTV, which corresponds to a total treatment D90 ⩾ 85 Gy in EQD2 (α/β = 10 Gy), including EBRT. The first 20 patients meeting the following inclusion criteria were selected for this study: (i) use of the same applicator dimensions for both applications and (ii) two MRI and one CT available for each application. The distribution of FIGO stage classification for local tumour stage was IB1 = 1, IB2 = 1, IIB = 15, and IIIB = 3. The distribution of MRI-based HR CTV volumes at the time of the first BT treatment was 10–20 cm³ (3), 20–40 cm³ (12), 40–60 cm³ (2), and 60–90 cm³ (3). All patients received 35–45 Gy external beam radiotherapy (EBRT) to the whole pelvis, prior to the first BT fraction. Except for one patient who received 6 HDR BT fractions with a D90 ⩾ 5 Gy prescribed to the HR CTV, all patients were treated to our clinical standard. For the first fraction of each application an MRI-based treatment plan was generated. On the second day of each application another MRI series was acquired to verify positions of applicator and organs at risk. If the organ motion between first and second fractions were not expected to lead to a substantial difference to the planning aims for dose parameters, the plan from the first fraction was reused for treatment of the second fraction without any further optimization.

For small tumours the MRI/CT compatible Vienna intracavitary tandem/ring applicators (Nucletron, Veenendaal, The Netherlands) with intrauterine length of 40 or 60 mm and diameters of 26, 30 and 34 mm were used. For larger tumours with parametrial extensions that could not be covered with the intracavitary applicators alone, or for situations with difficult OAR topography, additional interstitial needles, parallel to the tandem, were used in the ring [13,14]. For two cases with more locally advanced disease (V_{HR CTV} > 60 cm³) a custom made applicator with additional interstitial needles deviating from the tandem axis was used (ViennaII) [15].

Imaging and contouring

All patients underwent MR imaging with a 0.2 Tesla low-field system (Siemens Magnetom Open-Viva; Siemens AG, Erlangen, Germany) at the time of BT, with applicators in place. The details about the sequences used are given in Dimopoulos et al. [16]. In addition to the MRI all patients underwent CT imaging (Siemens Somatom Plus 4 Volume Zoom, Erlangen, Germany). The average time between the start of MRI and CT acquisition was 54 min.

While for all MR images the slice thickness was 5 mm, CT scans were taken with 4 mm slice thickness for 11 patients and 2 mm for 9 patients. The use of intravenous contrast was not performed systematically in this cohort of patients. OAR was contoured on the CT images by physicians without knowledge of the corresponding MRI or with at least one month time difference after they contoured on MRI.

Applicator reconstruction

Applicator surface reconstruction on MR and CT images was performed using the 3D applicator library implemented in the treatment planning system (TPS) Oncentra GYN (Nucletron, Veenendaal, v. 1.2.1.–1.2.3). By defining the tip of the tandem and the centre of the ring, the model was placed on the 3D MRI dataset. Fine tuning was done manually by shifting and rotating the whole applicator. In the ring part of the applicator holes for the interstitial needles could be used as additional markers to define the correct applicator rotation on MRI, as they produce high intensity signals on the MRI due to their filling with blood or other fluids. For accurate representation of the actual source path in the ring, we implemented a modified circular source path in the treatment planning software, which fits best for the dwell positions which are most relevant for our clinical standard loading pattern [17]. Verification of this source path with autoradiographs is included in our clinical applicator and TPS commissioning procedures.

For reconstruction based on CT images the visible source channel as well as the needle guiding holes could be matched with the contours of the library applicator. MRI- or CT-based reconstruction of the interstitial needles was done by selecting the position of the guiding hole built into the applicator model and marking the corresponding position of the needle tip on the image. Small manual adjustments had to be made in case a needle was bent, so its source path clearly deviated from a straight line.

Target transfer between MRI and CT

For the fully MRI-based plans of the first and second BT application targets (HR CTV, GTV) and OAR (bladder, rectum, and sigmoid) contours delineated on MRI were used for dose planning according to our clinical protocol [9,12,18,19].

For the CT-based plans of the second applications of each treatment series, organs at risk were delineated on the CT images. However the target contours were copied from the MRI of the first application (MRI1) to the corresponding CT of the second application (CT2). To copy the targets a rigid image fusion option in the TPS was used. As a first step, applicators were reconstructed on both the MRI and the CT. The internal applicator coordinate systems were then used as a reference for automatic rigid image registration between the MRI and CT series. After the registration, the target structures delineated on the MRI (HR CTV_MRI1) were loaded and stored onto the CT dataset, keeping their position relative to the applicator fixed. In case of overlaps with OAR contours depicted on CT images the transferred target contours were adjusted. In the rest of this work, these transferred targets will be referred to as HR CTV_CT2.

In order to estimate the influence of the image resolution on the quality of the target transfer procedure, two subgroups of images with CT slice thickness of 4 mm (n = 11) and 2 mm (n = 9) were evaluated separately.

Plan optimization

For each patient, an MRI of the first and second applications (MRI1, MRI2) and a CT of the second application (CT2) were included in this study. All treatment plans were computed with the Oncentra GYN TPS. Manual dwell time optimization was performed, taking into account the protocol based dose constraints [9]. DVH parameters were reported for target structures (HR CTV) and OAR (bladder, rectum, sigmoid) [2].

CT-based dose optimization was performed by a physicist without knowledge of the images, target and organ contours, and dose distribution or DVH parameters of the MRI-based plan of the second application.

An overview of the whole planning workflow used in this study is given in Table 1.

Table 1.

Schematic workflow of the MRI/CT combination method.

1st Application (MRI)		2nd Application (CT)
3D applicator reconstruction		3D applicator reconstruction
Target delineation		Target transfer from 1st MRI via image fusion based on applicator
OAR delineation		OAR delineation
Dose planning and optimization		Dose planning and optimization

Dose summation and plan evaluation

In total 60 treatment plans for 20 patients were investigated. The volumes and D90 of the HR CTV as well as the D_2cc of organs at risk (bladder, rectum, sigmoid) were included in the detailed comparison of the MRI and MRI/CT techniques.

For each patient two sets of BT plans were used to compute the normalized biological equivalent dose of the total treatment consisting of EBRT + four fractions of BT, i.e., (i) two fully MRI based plans and (ii) one MRI-based and one combined MRI/CT-based plan. For each BT application of two fractions, one optimized plan was computed for the first fraction. The same plans and dose values were used for the second fractions. The total treatment doses were calculated in EQD2 using α/β = 10 Gy for target and α/β = 3 Gy for OAR.

For the fully MRI-based treatment scenario the dose of EBRT (45–50 Gy physical dose) was added to the doses from the set of two MRI-based treatment plans for the two BT applications. The result of this calculation will subsequently be referred to as “MRI-plan dose”.

For the MRI/CT-based scenario, the total dose was calculated from EBRT plus the dose from the MRI-based plan of the first application and the dose from the CT-based plan of the second application. The result represents the DVH values which would be reported without any knowledge of an MRI at the time of the second application. This is the dose we would expect to be delivered to the patient using MRI for the first and CT for subsequent applications. Hence, further on, this dose will be referred to as “virtual dose for MRI/CT-plan”.

For assessing the quality of the combined MRI/CT planning method, the CT-based dose distributions were also evaluated on the original MRI images and corresponding contours of the second application. The loading pattern and dwell times were transferred from the CT to the MRI dataset and DVH parameters were computed for the MRI-based target and OAR contours. The result gives the dose that would have really been delivered to the MRI-based target volume of the second BT by application of the CT plan, and is therefore referred to as “real dose for MRI/CT-plan”.

The three types of dose summation are illustrated by a schematic drawing in Fig. 1.

Fig. 1.

Schematic representation of the dose summation types for the two different treatment planning scenarios: (I) fully MRI-based planning (MRI plan dose), (II) combined MRI/CT-based planning with (IIa) calculation of expected DVH values based on MRI of first and CT of second application (MRI/CT plan – virtual dose) and (IIb) calculation of dose that would be delivered to MRI-based target contours for both applications (MRI/CT plan – real dose).

Differences between DVH parameters based on different planning scenarios were calculated and reported as mean difference ±1 standard deviation. The observed differences were evaluated using a two-sided paired t-test. Differences with p < 0.05 were considered to be statistically significant.

Results

Target volumes

The mean difference between volumes of target contours delineated on the MRI of the first and second applications, i.e., volume of HR CTV_MRI1 − volume of HR CTV_MRI2, which can be influenced by tumour shrinkage between fractions, varying image quality and contouring uncertainties, was 4.6 ± 7.4 cm³ (9.2 ± 17.9%) with a median of 3.1 cm³. According to a two-sided t-test this difference was statistically significant (p = 0.009). The volumes of HR CTV_MRI2 structures exceeded the volumes of the structures transferred from MRI to CT (HR CTV_CT2) by 1.7 ± 6.6 cm³ (2.9 ± 20.4%) with a median of 0.7 cm³ (p = 0.26).

The first MRI target volumes (HR CTV_MRI1) were found to be slightly larger than the volumes of the same structures transferred onto the CT dataset (HR CTV_CT2) by 3.1 ± 3.9 cm³ (9.7 ± 9.5%) with a median of 2.5 cm³ (p = 0.002). Volume was lost due to interpolation techniques when copying the contours between two image series. For two cases the volume of the original HR CTV_MRI1 exceeded the volume of the transferred structure, calculated on the CT images, by more than 20%. For plans based on CT datasets with 4 mm slice thickness the mean transfer-related volume decrease was 15.4 ± 9.4% (p = 0.08), and for those with 2 mm slice thickness 2.7 ± 1.7% (p = 0.04).

Comparison of physical dose per fraction

For the fully MRI based plans the mean D90_MRIplan of the HR CTV_MRI2 volumes was 7.9 ± 0.7 Gy. The average of the virtual D90_CTplan was 7.7 ± 0.8 Gy for HR CTV_CT2. When the CT-based plans were evaluated on the purely MRI-based contours, HR CTV_MRI2, the mean real D90_CTplan was 8.0 ± 1.2 Gy. The comparison of virtual and real doses from the CT plans is shown in Fig. 2.

Fig. 2.

Comparison of D90 HR CTV in physical dose for one fraction, computed with the CT-based plan for the MRI target contour (HR CTV_MRI2) of the second application and for the transferred target (HR CTV_CT2) (real dose MRI/CT plan vs. virtual dose MRI/CT plan). Open symbols correspond to intracavitary implants, filled symbols to intracavitary + interstitial applications.

The mean difference between the real and virtual D90_CTplan, i.e., D90_CTplan HR CTV_MRI2 − D90_CTplan HR CTV_CT2, was 0.28 ± 0.76 Gy (0.54 ± 10.91%) for these fractions (p = 0.22). A positive value means that the real D90 of the CT plan was higher when evaluated on MRI structures.

For CT-based plans the DVH analysis for OAR delineated on the CT images yielded a D_2cc of 5.1 ± 0.8 Gy, 3.2 ± 0.8 Gy and 3.9 ± 0.7 Gy for bladder, rectum and sigmoid, respectively. Calculation of the D_2cc of the CT-based plans on organ contours delineated on MRI of the second application resulted in 5.0 ± 1.0 Gy, 2.9 ± 0.6 Gy and 3.5 ± 1.2 Gy for bladder, rectum and sigmoid.

Differences larger than ± 1 Gy between D_2cc calculated on CT and MRI organ contours were observed in 6 cases for bladder, in 1 case for rectum and in 5 cases for sigmoid.

Comparison of total treatment dose

For comparison of the total treatment dose delivered with the fully MRI-based and MRI/CT-combination techniques, the dose from EBRT (45–50 Gy) was added to the corresponding doses for all 4–6 fractions of BT given in two applications (Fig. 3). Total EQD2 doses are reported for target and OAR in Table 2.

Fig. 3.

Comparison of D90 for the whole course of treatment for the fully MRI-based BT (D90_MRI) and the MRI/CT method (D90_MRI/CT). Doses were calculated with corresponding plans on the same MRI-based HR CTV contours (real dose MRI/CT plan vs. dose MRI plan). Different symbols indicate the range of HR CTV_MRI volumes. Open symbols correspond to intracavitary implants, filled symbols to intracavitary + interstitial applications.

Table 2.

Summary of the DVH parameters used for optimization, for HR CTV and OAR. Total doses are given for the three scenarios: “MRI plan” (dose delivered to the patient with fully MRI-based BT), “MRI/CT plan virtual dose” (expected dose computed from 1 MRI-based and 1 MRI/CT-based application) and “MRI/CT plan real dose” (dose that would have been delivered with MRI/CT method; calculations based on MRI contours for both applications).

	D90 (HR CTV) [Gy EQD2]	D2cc bladder [Gy EQD2]	D2cc rectum [Gy EQD2]	D2cc sigmoid [Gy EQD2]
MRI-plan dose (2 MRI)
Mean	91.2	76.0	57.2	62.8
SD	5.3	8.7	5.9	7.2
MRI/CT plan virtual dose (1 MRI + 1 CT)
Mean	90.4	76.0	58.6	64.9
SD	5.0	8.1	4.8	6.2
MRI/CT plan real dose (2 MRI)
Mean	91.9	76.0	57.3	63.8
SD	6.4	8.8	4.1	8.0

The virtual D90 of the MRI/CT-based plans was found to be 1.5 ± 4.3 Gy lower than the corresponding total real D90 (p = 0.14). The mean difference of D90 HR CTV for the fully MRI-based plans and the real D90 HR CTV for the MRI/CT-combination technique was −0.7 ± 5.3 Gy. A negative sign means that on average the real D90 was lower for the fully MRI based plans than for the MRI/CT plans. The difference was not found to be statistically significant (p = 0.59).

For OAR, calculation of the DVH for the MRI/CT plans on CT- or MRI-based structures resulted in mean differences of D_2cc of 0 ± 4.9 Gy, 1.3 ± 1.2 Gy and 1.1 ± 4.2 Gy, for bladder (p = 0.99), rectum (p = 0.0001) and sigmoid (p = 0.26), respectively. A positive value means the virtual total D_2cc was larger than the real total D_2cc. When MRI and MRI/CT plans both were evaluated with the MRI anatomy, the mean differences of real total D_2cc for MRI vs. MRI/CT plans were 0 ± 3.6 Gy (bladder, p = 0.99), 0.1 ± 2.7 Gy (rectum, p = 0.90) and 1.0 ± 2.9 Gy (sigmoid, p = 0.15). A positive value means that the real total D_2cc of the MRI/CT plans was larger than the total D_2cc of the MRI plan (Fig. 4).

Fig. 4.

Comparison of D_2cc for OAR, for the whole treatment course, calculated for MRI anatomy. D_{2cc MRI} denotes the fully MRI-based plan parameters, D_{2cc MRI/CT} corresponds to the plans generated with the MRI/CT combination method (real dose MRI/CT plan vs. dose MRI plan).

The duration of the time between the start of the MRI and CT acquisition was not found to correlate with the difference between DVH parameters of the CT plans computed on CT vs. MRI contours.

Discussion

MRI is nowadays considered the gold standard for 3D image guided adaptive cervix cancer brachytherapy as it provides detailed anatomical information for application and treatment planning. Clincal outcome [1,5,8,9,20,21] and dose–effect relations for MRI-based BT [10,19,22,23] have been reported for targets and OAR by various groups. The limitations of delineating target structures on CT images have recently been investigated in comparison with MRI-based reference structures [11,24,25].

Since MRI is not available for planning of each BT fraction in all institutions where image guided cervix cancer BT is performed, other studies combining MRI and CT for different fractions have been reported in the literature.

Beriwal et al. [26] first reported the use of MRI at the time of the first BT planning with applicator in place, followed by subsequent CT-based fractions. The target structure delineated on the MRI of the first fraction was used as a guideline for contouring on CT for subsequent fractions. Clinical results of this MRI/CT hybrid method, for intracavitary cervix BT, were presented but not compared to conventional plans or fully MRI-based plans.

Wakatsuki et al. [27] and Kang et al. [28] used MRI acquired before the BT treatment, without applicator in situ, to improve contouring of the HR CTV on subsequent CT images at the time of brachytherapy. This way dose plans could be optimized to a more realistic target structure than could be contoured on CT alone, and OAR as depicted on CT. A similar MRI/CT technique was used by Fokdal et al. [29] for MRI-based BT pre-planning with a MUPIT applicator in situ.

An MRI/CT hybrid technique using MRI for pre-BT planning with tandem ovoid applicators in situ and subsequent plan adaptation based on CT contours at the time of BT treatment was reported by Dolezel et al. [30]. In their study rigid image fusion of MR and CT images based on pre-defined applicator points in combination with soft-tissue landmarks was applied and contours were transferred from MRI to CT for optimization of the pre-plan, on OAR contours depicted on CT. The technique was evaluated by comparison of fully 3D based hybrid MRI/CT plans with standard plans (HR CTV D90 vs. dose to point A). The authors confirmed substantial improvement of target coverage and OAR sparing using an adaptive treatment strategy.

The technique of combining MRI for the first BT fraction and CT for subsequent adaptive planning of the following fractions, based on automatic rigid image registration based on the tandem-ring applicators, for systematic and reproducible transfer of the target contours from MRI to CT plans, is reported here for the first time. In all other studies where combinations of MRI and CT for adaptive planning of cervix cancer BT were analyzed, the results could not be directly compared to a fully MRI-based comparison set of treatment plans, as it is done in the present study. Moreover, our study covers various types of target volumes occurring in clinical practice, which allows for comparison between purely intracavitary implants and those including interstitial needles.

In four cases the difference of D90_CTplan per fraction for HR CTV_CT2 and HR CTV _MRI2 was larger than 1 Gy (Fig. 2). Three of these outliers correspond to intermediate HR CTV volumes (25–29 cm³). All of them were treated with additional interstitial needles due to OAR topography. With small volumes and additional needles small differences in the target structures can have large impacts on the DVH as steep dose gradients occur close to the target surface.

The fourth outlier corresponds to the largest HR CTV volume in the sample (86.6 cm³), which was treated with the Vienna II applicator. An inspection of the MRI-based plans of the first and second applications for this patient revealed that the applicator position in relation to the target was different for the two applications. The applicator rotation in relation to the anterior-posterior axis as defined by the MRI was 21° while no rotation of the target structure could be observed.

Hence, after the automatic transfer of HR CTV_MRI1 to the CT of the second application the CT plan was generated for an anatomical situation, which was significantly different from the real anatomy as depicted on the MRI for that day. Therefore it was not possible to achieve the desired target coverage when planning on the CT, taking into account also the difficult sigmoid topography. The virtual D90 was smaller than 7 Gy. However, when projected onto the MRI anatomy a D90 of 7.9 Gy was reached by the MRI/CT-based plan.

As indicated in Fig. 2 the largest deviations of the D90 between the two methods were found for combined intracavitary plus interstitial implants. Considering the whole data sample, no clear correlation was found between the D90 of the HR CTV_CT2 or HR CTV_MRI2 and the tumour size or application technique.

The results of the analysis of total treatment dose differences between the full MRI and MRI/CT combination methods (Fig. 3) show that while planning with CT images the real D90 was underestimated on average by 1.5 ± 4.3 Gy. However, the mean difference between D90 of the fully MRI-based plans and MRI/CT plans evaluated on the actual MRI anatomy was only −0.7 ± 5.3 Gy, with the real D90 for MRI/CT plans being on average slightly larger than for the MRI plans. The observed differences were not statistically significant. Therefore, on average, the quality of the MRI/CT combination method seems comparable to the fully MRI-based planning. The main reason for the deviations found for individual plans are the differences in volume and shape of the target contours between transferred target HR CTV_CT2 and actual HR CTV_MRI2 target. These differences however are partly due to volume loss during the transfer procedure (interpolation between MRI and CT slices with different alignment and thickness) and actual tumour shrinkage between consecutive treatment weeks. However, as these cases received BT after or at the end of EBRT the tumour shrinkage was less pronounced compared to our previous findings also based on different treatment schedules [18]. The latter is reflected in the differences of the MRI1 and MRI2 target contours. As half of the patients were imaged with 2 mm slice thickness for CT the dependence of HR CTV volume loss on image resolution could be studied. The measured tumour shrinkage was found to be of the same order as the volume loss during target transfer, when using 4 mm slice thickness, i.e., 8.8 ± 20.5% and 15.4 ± 9.4%, respectively. For 2 mm CT slices the average transfer-related volume loss was reduced to 2.7%. However, for the two planning methods no correlation between the differences in total D90 with CT slice thickness could be detected in either of the two CT subgroups.

There was only one case where the minimum dose constraint for the MRI-based HR CTV was missed by the MRI/CT plan. The D90_MRI/CT of the HR CTV_MRI was 77.0 Gy. However, for this target, which was the second largest in the sample (V_{HR CTV} = 67.4 cm³) the total D90 of the fully MRI-based plan was already only 81.9 Gy.

For one plan the real total D_{2cc MRI/CT} for sigmoid failed to stay below the dose limit by 2 Gy, even though there was no violation encountered during “virtual dose” planning with the CT-based structures. Depiction of the sigmoid in this patient’s CT scan was not very clear, which introduced a higher uncertainty in the CT contouring than for other cases. Moreover, non-negligible interfraction variations, i.e., organ motion between MRI and CT acquisition, might have taken place and may have contributed to this large deviation.

For all other cases all dose constraints for OAR were fulfilled.

Special information from clinical examination was not taken into account while planning on CT. This limitation constitutes a disadvantage of any retrospective treatment planning study. Therefore, during a clinical implementation phase of this method a parallel prospective protocol should be applied.

The present analysis is limited to tandem/ring applicators, although further work should show appropriate use of the underlying method for other application techniques.

Conclusions

This study has shown that a combination of MRI for first fraction and subsequent CT based planning is feasible as well as quick and easy when automatic applicator-based image registration and target transfer are implemented in the TPS. The results show striking similarity to fully MRI-based planning in cases of small tumours and intracavitary applications, both in terms of HR CTV coverage and respecting of OAR dose limits. The MRI/CT combination method may further be improved by focussing on information from clinical examination in the CT-based planning process and better understanding of CT contouring.

However, for larger tumours and complex applications, as well as situations with unfavourable OAR topography, especially for the sigmoid, MRI based adaptive BT planning remains the superior method and is hence still to be considered as the gold standard.

Conflict of interest statement

The Department of Radiotherapy at the Medical University of Vienna receives financial and/or equipment support for research and educational purposes from Nucletron B.V. and Varian Medical Systems Inc. C. Kirisits has been a consultant to Nucletron B.V.