Advertisement

The first-in-human implementation of adaptive 4D cone beam CT for lung cancer radiotherapy: 4DCBCT in less time with less dose

      Highlights

      • Adaptive 4DCBCT imaging has been clinically implemented for lung cancer radiotherapy patients for the first time.
      • Adaptive 4DCBCT imaging has the potential to substantially reduce 4DCBCT scan times and imaging dose.
      • Adaptive 4DCBCT imaging can be combined with motion compensated image reconstruction techniques for better results.

      Abstract

      Background and purpose

      We present the first implementation of Adaptive 4D cone beam CT (4DCBCT) that adapts the image hardware (gantry rotation speed and kV projections) in response to the patient’s real-time respiratory signal. Adaptive 4DCBCT was applied on lung cancer patients to reduce the scan time and imaging dose in the ADaptive CT Acquisition for Personalised Thoracic imaging (ADAPT) trial.

      Materials and methods

      The ADAPT technology measures the patient’s real-time respiratory signal and uses mathematical optimisation and external circuitry attached to the linear accelerator to modulate the gantry rotation speed and kV projection rate to reduce scan times and imaging dose. For each patient, ADAPT scans were acquired on two treatment fractions and reconstructed with a motion compensated reconstruction algorithm and compared to the current state-of-the-art four-minute 4DCBCT acquisition (conventional 4DCBCT). We report on the scan time, imaging dose and image quality for the first four adaptive 4DCBCT patients.

      Results

      The ADAPT imaging dose was reduced by 85% and scan times were 73 ± 12 s representing a 70% reduction compared to the 240 s conventional 4DCBCT scan. The contrast-to-noise ratio was improved from 9.2 ± 3.9 with conventional 4DCBCT to 11.7 ± 4.1 with ADAPT.

      Discussion

      The ADAPT trial represents the first time that gantry rotation speed and projection acquisition have been adapted and optimised in real-time in response to changes in the patient’s breathing. ADAPT demonstrates substantially reduced scan times and imaging dose for clinical 4DCBCT imaging that could enable more efficient and optimised lung cancer radiotherapy.

      Keywords

      Four-dimensional cone beam computed tomography (4DCBCT) has had increased adoption as an image guidance strategy for lung cancer radiotherapy particularly for hypofractionated treatments [
      • Sonke J.J.
      • et al.
      Respiratory correlated cone beam CT.
      ,
      • Chang J.
      • et al.
      Integrating respiratory gating into a megavoltage cone-beam CT system.
      ,
      • Dietrich L.
      • et al.
      Linac-integrated 4D cone beam CT: first experimental results.
      ,
      • Kriminski S.
      • et al.
      Respiratory correlated cone-beam computed tomography on an isocentric C-arm.
      ,
      • Li T.
      • et al.
      Four-dimensional cone-beam computed tomography using an on-board imager.
      ]. However, 4DCBCT suffers from inconsistent image quality from one patient to the next, long scan times (typically 240 s), streaking artefacts and higher imaging doses than are necessary [
      • Gao H.
      • et al.
      4D cone beam CT via spatiotemporal tensor framelet.
      ,
      • Li T.
      • Koong A.
      • Xing L.
      Enhanced 4D cone-beam CT with inter-phase motion model.
      ,
      • Li T.
      • Xing L.
      Optimizing 4D cone-beam CT acquisition protocol for external beam radiotherapy.
      ,
      • Lu J.
      • et al.
      Four-dimensional cone beam CT with adaptive gantry rotation and adaptive data sampling.
      ].
      A range of methods have been proposed in the literature to reduce scan times, imaging dose or improve image quality. These range from new reconstruction techniques [
      • Gao H.
      • et al.
      4D cone beam CT via spatiotemporal tensor framelet.
      ,
      • Li T.
      • Koong A.
      • Xing L.
      Enhanced 4D cone-beam CT with inter-phase motion model.
      ,
      • Bian J.
      • et al.
      Evaluation of sparse-view reconstruction from flat-panel-detector cone-beam CT.
      ,
      • Bergner F.
      • et al.
      An investigation of 4D cone-beam CT algorithms for slowly rotating scanners.
      ,
      • Bergner F.
      • et al.
      Autoadaptive phase-correlated (AAPC) reconstruction for 4D CBCT.
      ,
      • Leng S.
      • et al.
      High temporal resolution and streak-free four-dimensional cone-beam computed tomography.
      ,
      • Brehm M.
      • et al.
      Artifact-resistant motion estimation with a patient-specific artifact model for motion-compensated cone-beam CT.
      ,
      • Brehm M.
      • et al.
      Self-adapting cyclic registration for motion-compensated cone-beam CT in image-guided radiation therapy.
      ,
      • O'Brien R.T.
      • et al.
      Optimizing 4DCBCT projection allocation to respiratory bins.
      ] optimising the gantry rotation speed [
      • Li T.
      • Xing L.
      Optimizing 4D cone-beam CT acquisition protocol for external beam radiotherapy.
      ,
      • Lu J.
      • et al.
      Four-dimensional cone beam CT with adaptive gantry rotation and adaptive data sampling.
      ] or adapting the image acquisition on-the-fly to changes in the patient’s breathing rate [
      • O'Brien R.
      • Sonke J.
      • Keall P.
      Respiratory motion guided 4DCBCT: an XCAT study to analyse image quality in fast scans (1–2min).
      ,
      • O'Brien R.
      • et al.
      Respiratory motion guided 4DCBCT-A step towards controlling 4DCBCT image quality.
      ,
      • O'Brien R.T.
      • Cooper B.J.
      • Keall P.J.
      Optimizing 4D cone beam computed tomography acquisition by varying the gantry velocity and projection time interval.
      ,
      • O'Brien R.T.
      • et al.
      Respiratory motion guided four dimensional cone beam computed tomography: encompassing irregular breathing.
      ,
      • O'Brien R.T.
      • et al.
      The first implementation of respiratory triggered 4DCBCT on a linear accelerator.
      ,
      • O'Brien R.T.
      • et al.
      Reducing 4DCBCT imaging time and dose: the first implementation of variable gantry speed 4DCBCT on a linear accelerator.
      ,
      • Cooper B.
      • O'Brien R.
      • Keall P.
      Respiratory signal triggered 4D cone-beam computed tomography on a linear accelerator.
      ,
      • Cooper B.J.
      • et al.
      Respiratory triggered 4D cone-beam computed tomography: A novel method to reduce imaging dose.
      ,
      • Cooper B.J.
      • et al.
      Quantifying the image quality and dose reduction of respiratory triggered 4D cone-beam computed tomography with patient-measured breathing.
      ,
      • Dillon O.
      • et al.
      Evaluating reconstruction algorithms for respiratory motion guided acquisition.
      ,
      • Fast M.F.
      • et al.
      Actively triggered 4d cone-beam CT acquisition.
      ]. Recent work has suggested that by combining motion compensated reconstruction methods [
      • Rit S.
      • et al.
      On-the-fly motion-compensated cone-beam CT using an a priori model of the respiratory motion.
      ,
      • Rit S.
      • et al.
      Comparative study of respiratory motion correction techniques in cone-beam computed tomography.
      ] with adaptive 4DCBCT imaging techniques [
      • O'Brien R.T.
      • et al.
      Reducing 4DCBCT imaging time and dose: the first implementation of variable gantry speed 4DCBCT on a linear accelerator.
      ] the scan times can be reduced by up to 75% and imaging doses can be reduced by 85% while maintaining similar or better image quality than is currently achieved [
      • Dillon O.
      • et al.
      Evaluating reconstruction algorithms for respiratory motion guided acquisition.
      ].
      In the present work, we describe the first-in-human clinical implementation of adaptive 4DCBCT where the image acquisition (gantry rotation speed and kV acquisition rate) were adapted and synchronised with the patient’s real-time respiratory signal. As a simplified example, if the patient breathes faster, the gantry rotation speed and kV acquisition rate increase. In practice, mathematical optimisation is used to determine and update the gantry rotation speed and kV acquisition rate in real-time that minimise image dose, scan times and maximise image quality [
      • O'Brien R.T.
      • Cooper B.J.
      • Keall P.J.
      Optimizing 4D cone beam computed tomography acquisition by varying the gantry velocity and projection time interval.
      ]. We report on the image quality, scan time and imaging dose for the first four lung cancer patients that were acquired with adaptive 4DCBCT imaging at Liverpool Hospital (Liverpool, New South Wales, Australia).

      Materials and methods

      All legal, ethical and regulatory requirements for the ADaptive CT acquisition for Personalised Thoracic imaging (ADAPT) trial were met (NCT04070586, ethics approval 2019/ETH09968).

      Study objectives

      The ADAPT trial was designed to reduce 4DCBCT scan times while obtaining image quality equivalent to current implementations with a conventional 4-minute acquisition. The primary objective of the ADAPT trial is therefore to determine the feasibility of adaptive 4DCBCT imaging for patients undergoing lung cancer radiotherapy by comparing ADAPT scans with conventional 4DCBCT scans. Secondary objectives including assessing image quality for a range of breathing rates, breathing amplitudes, patient sizes and tumour locations as well as assessing image quality between different treatment fractions for the same patient (30 patients are to be recruited).

      Patient eligibility and exclusions

      Lung cancer patients over 18 years who could give informed consent were eligible provided that they were to receive at least two treatment fractions and a 4DCBCT scan was to be acquired for at least one fraction. Patients were excluded if they were pregnant or in the opinion of the treating physician could not tolerate the extra treatment time because of the two additional ADAPT scans on top of their treatment. For this study, patients were excluded if their breathing period was less than two seconds or greater than 8.5 seconds due to issues with the ADAPT system being able to calculate the respiratory phase in real-time. Table 1 gives a summary of the patient characteristics (age, height and weight) as well as the tumour location for the first four patients recruited.
      Table 1Patient characteristics for the first four patients in the ADAPT trial.
      Patient numberSexAge

      (years)
      Height (cm)Weight (kg)Tumour LocationTumour stage
      1Male7115594.4MediastinalT2N2M0
      2Male6516360.4MediastinalT1N3M0
      3Female5916263.8MediastinalT3N2M0
      4Female4916355.0MediastinalT4N2M0

      ADAPT implementation: Hardware

      The ADAPT system requires three separate components (these are described in more detail in reference O'Brien, [
      • O'Brien R.T.
      • et al.
      Reducing 4DCBCT imaging time and dose: the first implementation of variable gantry speed 4DCBCT on a linear accelerator.
      ]; these components are:
      A respiratory sensor to record the patient’s real-time respiratory signal. For this trial an intel real sense depth camera attached to a camera tripod was used Fig. 3A (supplement). From the depth camera image, a region on the patient’s chest with a clear respiratory signal was selected and the average depth over this region was used as the respiratory signal.
      External circuitry that was electrically isolated from the linear accelerator was attached to change the gantry rotation speed and modulate kV projections in real-time in response to changes in the patient’s real-time respiratory signal. The gantry angle was measured with an inclinometer attached behind the linac, Fig. 3B (supplement). The gantry rotation speed was adjusted by mechanically turning a potentiometer attached to the gantry speed control signal while the kV pulses were suppressed using a solid-state relay.
      The ADAPT microcontrollers and control computers (Fig. 3C (supplement)) which processed the real-time respiratory signal and calculated the desired gantry rotation speed and kV acquisition schedule at a frequency of 30 Hz. The ADAPT microcontroller receives input from the respiratory sensor and sends out signals to the external circuitry controlling the gantry speed and kV acquisition.

      ADAPT implementation: software control

      The ADAPT system ensures that a specified number of projections, N, are acquired in each of the pre-determined number of respiratory phases. Ten respiratory phases were used to allow direct comparison to conventional 4DCBCT. In ADAPT, the gantry rotation speed and kV projection acquisition are modulated in order to acquire these projections as equally spaced as possible to ensure consistent image quality from one patient to the next [
      • O'Brien R.T.
      • Cooper B.J.
      • Keall P.J.
      Optimizing 4D cone beam computed tomography acquisition by varying the gantry velocity and projection time interval.
      ]. In this study, the ADAPT scan was defined as N = 20 projections in each of ten respiratory phases (20 × 10 = 200 projections in total).
      To ensure that exactly 20 uniformly distributed projections are acquired in each respiratory phase bin, the ADAPT control algorithm optimises the gantry trajectory and gantry velocity and then adapts the gantry speed in real-time due to changes in the patient’s breathing rate while monitoring the projection acquisition.
      The ADAPT hardware was installed on an Elekta Versa HD linear accelerator. Within Elekta’s X-ray volume imaging software (XVI), and XVI preset was set up to rotate the gantry 220° at 6°/s. The ADAPT system then scaled down the gantry rotation speed by mechanically turning the potentiometer to the desired speed. kV acquisition was suppressed for the first 10° to allow the gantry to get up to speed. After 10°, the system modulates the gantry rotation speed and kV acquisition for 200° while acquiring the ADAPT projections. During the final 10°, the kV acquisition and gantry rotation are stopped which ends the scan within XVI.

      ADAPT implementation: image reconstruction

      Projections were acquired at 120 kV with 20 mA for 25 ms. The detector size was 512 by 512 pixels and reconstruction was performed over 27 × 27 × 27 cm3 using 1 mm3 voxels. Standard filtered back projection via the Feldkamp-Davis-Kress (FDK) algorithm [
      • Feldkamp L.A.
      • Davis L.C.
      • Kress J.W.
      Practical cone-beam algorithm.
      ] was used for the conventional 4DCBCT reconstruction by binning the projection data into their separate respiratory phases based on the external respiratory signal and performing ten separate reconstructions.
      A motion compensated reconstruction method, that used all the projection data to compute the reconstructed 4D image set, was used for the ADAPT scans. Motion compensated reconstruction, which was one of the best performing reconstruction methods in the SPARE challenge for low projection count 4DCBCT [
      • Shieh C.C.
      • et al.
      SPARE: Sparse-view reconstruction challenge for 4D cone-beam CT from a 1-min scan.
      ] was performed by computing a deformation vector fields between all phases and a reference phase and then performing an FDK reconstruction along the path of motion represented by the deformation vector field to generate a 3D volume at the reference phase. The DVF can be generated either from the planning 4DCT before treatment [
      • Rit S.
      • et al.
      On-the-fly motion-compensated cone-beam CT using an a priori model of the respiratory motion.
      ,
      • Rit S.
      • et al.
      Comparative study of respiratory motion correction techniques in cone-beam computed tomography.
      ] so that the reconstruction time is the same as current clinical practice (i.e. available as soon as the gantry finishes rotating) or from a McKinnon-Bates reconstruction [
      • Mc Kinnon G.C.
      • Bates R.H.
      Towards imaging the beating heart usefully with a conventional CT scanner.
      ] using the 4DCBCT data on the day of treatment. The trade off with the second approach is that reconstruction times are a little longer (20 min in our current unoptimized implementation). In this study, the second approach will be used because the results were marginally better in our simulation studies [
      • Dillon O.
      • et al.
      Evaluating reconstruction algorithms for respiratory motion guided acquisition.
      ]. More details on the implementation of the motion compensated reconstruction method and analysis of computation times can be found in Dillon, [
      • Dillon O.
      • et al.
      Evaluating reconstruction algorithms for respiratory motion guided acquisition.
      ].

      ADAPT implementation: clinical workflow

      Patients recruited to the trial were treated as per Liverpool Hospital’s normal treatment protocol with the ADAPT scans acquired immediately after treatment. The Liverpool imaging protocol for lung cancer patients treated with curative conventional radiotherapy involves a 4DCBCT for the first treatment fraction and daily 3DCBCT for the remaining fractions. For the ADAPT study, we need a conventional 4DCBCT on the two fractions where the ADAPT scans are acquired. Therefore, the 3DCBCT on the second fraction was replaced with a conventional 4DCBCT to ensure a conventional 4DCBCT was available for comparison with the ADAPT scans. All patients treated with stereotactic radiotherapy undergo a 4DCBCT prior to each fraction so there was always a conventional 4DCBCT acquired for SBRT treatments.

      Performance analysis: scan times and imaging dose

      For the ADAPT method, one projection is acquired per respiratory phase per breathing cycle (i.e. 10 projections per breathing cycle), so the total scan time is the time required for the patient to complete 20 breathing cycles (e.g. 60 seconds for a 3 second breathing period). The ADAPT system also pauses acquisition if breathing becomes too irregular which happened whenever the ADAPT system was unable to compute the respiratory phase in real-time using the dynamical systems method detailed in Ruan, [
      • Ruan D.
      • et al.
      Real-time profiling of respiratory motion: baseline drift, frequency variation and fundamental pattern change.
      ]. When the system is unable to compute the respiratory phase because of irregular breathing, kV acquisition stops and the gantry is reduced to the minimum speed that did not result in an aborted scan (this was measured at 0.4 degrees per second) until the patient’s breathing returns to normal and the phase could be computed again.
      To leading order, we approximate the imaging dose by the number of projections acquired. For the conventional 4DCBCT scan, 1320 projections are acquired. The ADAPT scan acquired 200 projections in total representing an imaging dose reduction of 85% over conventional 4DCBCT.

      Performance analysis: image quality

      Fig. 1 displays the region selection for the metrics discussed in this section. On a phase by phase basis, the structural similarity index (SSI) was used to quantify the similarity between the conventional and the ADAPT scans with a value closer to 1 indicating a higher level of similarity between two images. The SSI was computed as
      SSI=2μCμr+c12σr,C+c2μC2+μr2+c2σC2+σr2+c2


      where μ and σ2 represent voxel value means and variances of the region of interest, c1=0.01L2 and c2=0.03L2 where L represent the dynamic range of the volumes. SSI values were averaged across all 10 bins.
      Figure thumbnail gr1
      Fig. 1The regions used for image quality metrics. The region for the SSIM (red). For CNR, a region in the lung was selected with limited anatomical structure (dark blue) and a region in the diaphragm (green). The yellow and light blue regions represent the regions used for the TIS_T (yellow) and TIS_D (light blue). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
      The contrast-to-noise ratio (CNR) was used to compare image contrast between scans where the diaphragm and lung were used as the high and low contrast regions respectively. In this study, we define the CNR as
      CNR=|μDiaphragm-μLung|σDiaphragm


      where μ and σ represent mean and standard deviation of region of interest voxels in the diaphragm and lung (see regions indicated in Fig. 1).
      We quantify image sharpness with the tissue interface sharpness (TIS) as in Riblett [
      • Riblett M.J.
      • et al.
      Data-driven respiratory motion compensation for four-dimensional cone-beam computed tomography (4D-CBCT) using groupwise deformable registration.
      ]. The TIS is computed by fitting a sigmoid function to a lung-tissue boundary. A 5 × 5 × l voxel sized region of interest was positioned at the top of the boundary of interest. Along each of 5 × 5 runs of length l20, a sigmoid function is fit in the superior-to-inferior direction. The value of TIS is the gradient of the sigmoid function with a higher value meaning a faster intensity change or sharper boundary. To reduce the dependence of the TIS for a single run length, the TIS is computed on a 5 × 5 voxel region (i.e., across 25 values) and then averaged. Where a boundary is well defined in the reconstruction field of view, we compute TIS-T at the top of the lung-tumour boundary and TIS-D at the top of the lung-diaphragm boundary.
      To quantify the tumours trajectory, we have created a cube region encapsulating the tumour. The ADAPT scan was rigidly registered to the conventional scan to ensure the same region was used between the conventional and ADAPT scans. For each respiratory phase, the centre of mass of the pixel values within the cube were computed to calculate an estimate of the trajectory of the tumour. The difference in the tumour location between the ADAPT and the conventional scan was then averaged across the ten respiratory phases to compute the average tumour position difference.

      Results

      For the first time, 4DCBCT scans on a linear accelerator were adapted in real-time to changes in the patient’s respiratory signal. Images from the conventional and ADAPT scans are given in Fig. 2. Visually, anatomical structures are reconstructed with suitable quality for image guidance using ADAPT despite a substantial reduction in the number of projections.
      Figure thumbnail gr2
      Fig. 2Images from the first four patients using conventional acquisition as well as the ADAPT system. The yellow tick marks in the coronal plane indicate the location of the axial and sagittal slices shown. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
      The SSI for each patient and each fraction is given in Table 2. For the ADAPT scans, the values are very close to 1 indicating a high level of structural similarity between the ADAPT and conventional 4DCBCT scans.
      Table 2Patient breathing rate (mean ± SD), gantry velocity and scan times (imaging time + interrupt time) for the first four patients in the ADAPT trial. The structural similarity index (SSI) is computed relative to the conventional 4DCBCT scan. The interrupt time is the time spent with acquisition paused during irregular breathing events; efforts are ongoing to determine if imaging during these irregular breathing events reduces image quality. The gantry velocity ± SD are also given for the ADAPT scans.
      Patient/FractionConventional

      Breathing period
      ADAPT
      Breathing periodScan timeInterrupt timeGantry velocitySSI
      Patient 1/Fraction 13.1 ± 0.5 s2.9 ± 0.6 s75 s15 s3.3 ± 0.6°/s0.99
      Patient 1/Fraction 22.6 ± 0.5 s2.6 ± 0.6 s72 s20 s3.4 ± 1.1°/s0.99
      Patient 2/Fraction 22.6 ± 0.4 s2.9 ± 0.5 s95 s37 s2.9 ± 0.5°/s0.98
      Patient 3/Fraction 13.4 ± 0.4 s3.6 ± 0.4 s80 s8 s2.7 ± 0.7°/s0.97
      Patient 3/Fraction 23.0 ± 0.5 s3.1 ± 0.3 s71 s9 s3.1 ± 0.6°/s0.99
      Patient 4/Fraction 12.3 ± 0.2 s2.1 ± 0.5 s59 s17 s3.2 ± 1.4°/s0.99
      Patient 4/Fraction 22.1 ± 0.4 s2.2 ± 0.8 s61 s13 s2.9 ± 1.4°/s0.99
      Average73 ± 12 s17 ± 9 s0.98 ± 0.01
      Scan times with ADAPT are primarily dependent on the patient’s breathing rate. Across the first four patients in the trial, the ADAPT scan averaged 73 s with a standard deviation of 12 s which compares favourably to the conventional scan time of 240 s. The patients breathing irregularity (standard deviation in breathing period) was marginally higher during the ADAPT scan for 5 out of 7 fractions. This could be a statistical anomaly in the first few patients, or it could have been caused due to patient fatigue (the ADAPT scans were at the end of treatment) or due to the patient modifying their breathing as a result of the ADAPT system.
      The interrupt time and degree of modulation of the gantry speed (standard deviation in gantry speed) are listed in Table 2. The interrupt time includes all pauses to the system for unexpected breathing behaviour including unexpected motion. For the first four patients, the level of gantry modulation and interrupt times do not appear to be directly linked to breathing irregularity.
      ADAPT has a higher CNR for 5 of the 7 acqusitions (lower CNR for patient 1 fraction 2 and patient 2 fraction 2), see Table 3. The average CNR is improved from 9.2 with conventional 4DCBCT to 11.8 with ADAPT. Table 3 shows that ADAPT improved TIS-D for 6 of 7 acqusitions (except for patient 3 fraction 1) with the average TIS-D improving from 0.7 with conventional 4DCBCT to 1.3 with ADAPT. TIS-T follows a similar pattern increasing from 0.3 with conventional 4DCBCT to 1.1 with ADAPT.
      Table 3The contrast-to-noise ratio (CNR) and tissue interface sharpness (TIS) for the first four patients in the ADAPT trial. Higher values of CNR indicate better image contrast and higher values of TIS indicate sharper, more defined, interfaces between boundaries. TIS-D computes the TIS across the lung-diaphragm boundary if the diaphragm boundary is clearly visible in the field of view (NA if it is not). TIS-T computes the TIS across the lung-tumour boundary.
      Patient/FractionConventional 4DCBCTADAPT Fast Scan
      CNRTIS-DTIS-TCNRTIS-DTIS-T
      1/14.30.40.415.01.10.7
      1/24.20.40.23.71.01.2
      2/212.10.6NA11.91.6NA
      3/16.11.50.411.90.91.6
      3/210.9NA0.311.6NA0.7
      4/112.80.7NA14.43.9NA
      4/214.30.6NA14.41.0NA
      Average9.2 ± 3.90.7 ± 0.40.3 ± 0.111.8 ± 3.61.3 ± 1.31.1 ± 0.4
      The tumour position difference between the ADAPT and conventional scans when averaged across the ten respiratory phases was in the range of 0 to 0.3 mm for all four patients.

      Discussion

      Our results indicate that substantial reductions in imaging dose and scan times can be achieved with ADAPT while maintaining suitable image quality for image guidance compared to the current state of the art. These results confirm our earlier simulation work in phantom studies O'Brien, [
      • O'Brien R.T.
      • et al.
      Reducing 4DCBCT imaging time and dose: the first implementation of variable gantry speed 4DCBCT on a linear accelerator.
      ]; Dillon, [
      • Dillon O.
      • et al.
      Evaluating reconstruction algorithms for respiratory motion guided acquisition.
      ]. The shorter scan times have the potential to reduce patient discomfort, fatigue and intrafraction motion due to spending almost three minutes less on the treatment couch.
      The ADAPT approach has the potential benefit of producing more consistent image quality from one patient to the next than compared to current clinical practice. More consistent image quality is expected to provide more confidence for radiation therapists and radiation oncologists particularly for hypofractionated treatments in and around the mediastinum. For example, consistent image quality is expected to provide more predictable results and error margins for image guidance, segmentation and registration tasks that are strongly influenced by higher image gradients. More generally, the introduction of adaptive 4DCBCT gives the option of trading off image quality, imaging time and dose. If for example, a clinic was comfortable with current scan times, a higher image quality scan could be enabled with adaptive 4DCBCT. The optimum trade off of these three variables on a per-patient basis will likely depend on the treatment intent, the fractionation schedule and the patient’s respiratory pattern.
      There are several limitations of this study. Of particular importance is that this is an analysis of the first four patients who all had mediastinal tumours and were relatively fast breathers. Assessing image quality across a broader patient cohort covering a range of patients, tumour locations and breathing patterns is expected to be address after the recruitment of the full 30 patient cohort.
      The reconstruction method uses deformable registration and although deformable registration is in widespread clinical use, it can introduce errors particularly at boundaries between highly mobile and stationary anatomy. Due to a lack of a ground truth, to quantify the motion measured in our reconstructions, we have used the SSI on a phase by phase basis between the conventional and ADAPT scans. The SSI values are close to 1, indicating that the dynamics are well resolved but a more direct method would be desirable. Finally, the ADAPT system requires the use of an external respiratory signal, which is currently not required for Elekta’s implementation of 4DCBCT but is in widespread use on the Varian platform and when acquiring 4DCT scans for treatment planning.
      This is the first time that linac hardware, rotation speed and projection acquisition, has been adapted in real-time to the patient with the aim of reducing scan times and dose for lung cancer radiotherapy.

      Conflict of interest

      Ricky O’Brien and Paul Keall are inventors on a patent underpinning the adaptive imaging technology in this paper.

      Acknowledgements

      We acknowledge the patients who enrolled in the trial and donated both their time and imaging dose. This work was funded by an NHMRC Australia [1–34] project grant 1138899 and partially by the Cancer Australia PdCCRS project grant number 1123068. RO was funded by a Cancer Institute NSW career development fellowship while PK was funded by an NHMRC Senior Principal Research Fellowship . We also thank the staff in the Department of Radiation Oncology at Liverpool Hospital for their support.

      Appendix A. Supplementary data

      The following are the Supplementary data to this article:

      References

        • Sonke J.J.
        • et al.
        Respiratory correlated cone beam CT.
        Med Phys. 2005; 32: 1176-1186
        • Chang J.
        • et al.
        Integrating respiratory gating into a megavoltage cone-beam CT system.
        Med Phys. 2006; 33: 2354-2361
        • Dietrich L.
        • et al.
        Linac-integrated 4D cone beam CT: first experimental results.
        Phys Med Biol. 2006; 51: 2939-2952
        • Kriminski S.
        • et al.
        Respiratory correlated cone-beam computed tomography on an isocentric C-arm.
        Phys Med Biol. 2005; 50: 5263-5280
        • Li T.
        • et al.
        Four-dimensional cone-beam computed tomography using an on-board imager.
        Med Phys. 2006; 33: 3825-3833
        • Gao H.
        • et al.
        4D cone beam CT via spatiotemporal tensor framelet.
        Med Phys. 2012; 39: 6943-6946
        • Li T.
        • Koong A.
        • Xing L.
        Enhanced 4D cone-beam CT with inter-phase motion model.
        Med Phys. 2007; 34: 3688-3695
        • Li T.
        • Xing L.
        Optimizing 4D cone-beam CT acquisition protocol for external beam radiotherapy.
        Int J Radiat Oncol Biol Phys. 2007; 67: 1211-1219
        • Lu J.
        • et al.
        Four-dimensional cone beam CT with adaptive gantry rotation and adaptive data sampling.
        Med Phys. 2007; 34: 3520-3529
        • Bian J.
        • et al.
        Evaluation of sparse-view reconstruction from flat-panel-detector cone-beam CT.
        Phys Med Biol. 2010; 55: 6575-6599
        • Bergner F.
        • et al.
        An investigation of 4D cone-beam CT algorithms for slowly rotating scanners.
        Med Phys. 2010; 37: 5044-5053
        • Bergner F.
        • et al.
        Autoadaptive phase-correlated (AAPC) reconstruction for 4D CBCT.
        Med Phys. 2009; 36: 5695-5706
        • Leng S.
        • et al.
        High temporal resolution and streak-free four-dimensional cone-beam computed tomography.
        Phys Med Biol. 2008; 53: 5653-5673
        • Brehm M.
        • et al.
        Artifact-resistant motion estimation with a patient-specific artifact model for motion-compensated cone-beam CT.
        Med Phys. 2013; 40101913
        • Brehm M.
        • et al.
        Self-adapting cyclic registration for motion-compensated cone-beam CT in image-guided radiation therapy.
        Med Phys. 2012; 39: 7603-7618
        • O'Brien R.T.
        • et al.
        Optimizing 4DCBCT projection allocation to respiratory bins.
        Phys Med Biol. 2014; 59: 5631-5649
        • O'Brien R.
        • Sonke J.
        • Keall P.
        Respiratory motion guided 4DCBCT: an XCAT study to analyse image quality in fast scans (1–2min).
        Med Phys. 2017; 44: 3284-3285
        • O'Brien R.
        • et al.
        Respiratory motion guided 4DCBCT-A step towards controlling 4DCBCT image quality.
        Med Phys. 2016; 43 (3800–3800)
        • O'Brien R.T.
        • Cooper B.J.
        • Keall P.J.
        Optimizing 4D cone beam computed tomography acquisition by varying the gantry velocity and projection time interval.
        Phys Med Biol. 2013; 58: 1705-1723
        • O'Brien R.T.
        • et al.
        Respiratory motion guided four dimensional cone beam computed tomography: encompassing irregular breathing.
        Phys Med Biol. 2014; 59: 579-595
        • O'Brien R.T.
        • et al.
        The first implementation of respiratory triggered 4DCBCT on a linear accelerator.
        Phys Med Biol. 2016; 61: 3488-3499
        • O'Brien R.T.
        • et al.
        Reducing 4DCBCT imaging time and dose: the first implementation of variable gantry speed 4DCBCT on a linear accelerator.
        Phys Med Biol. 2017; 62: 4300-4317
        • Cooper B.
        • O'Brien R.
        • Keall P.
        Respiratory signal triggered 4D cone-beam computed tomography on a linear accelerator.
        Med Phys. 2012; 39 (3605–3605)
        • Cooper B.J.
        • et al.
        Respiratory triggered 4D cone-beam computed tomography: A novel method to reduce imaging dose.
        Med Phys. 2013; 40
        • Cooper B.J.
        • et al.
        Quantifying the image quality and dose reduction of respiratory triggered 4D cone-beam computed tomography with patient-measured breathing.
        Phys Med Biol. 2015; 60: 9493-9513
        • Dillon O.
        • et al.
        Evaluating reconstruction algorithms for respiratory motion guided acquisition.
        Phys Med Biol. 2020;
        • Fast M.F.
        • et al.
        Actively triggered 4d cone-beam CT acquisition.
        Med Phys. 2013; 40091909
        • Rit S.
        • et al.
        On-the-fly motion-compensated cone-beam CT using an a priori model of the respiratory motion.
        Med Phys. 2009; 36: 2283-2296
        • Rit S.
        • et al.
        Comparative study of respiratory motion correction techniques in cone-beam computed tomography.
        Radiother Oncol. 2011; 100: 356-359
        • Feldkamp L.A.
        • Davis L.C.
        • Kress J.W.
        Practical cone-beam algorithm.
        J Opt Soc Am A-Opt Image Sci Vision. 1984; 1: 612-619
        • Shieh C.C.
        • et al.
        SPARE: Sparse-view reconstruction challenge for 4D cone-beam CT from a 1-min scan.
        Med Phys. 2019; 46: 3799-3811
        • Mc Kinnon G.C.
        • Bates R.H.
        Towards imaging the beating heart usefully with a conventional CT scanner.
        IEEE Trans Biomed Eng. 1981; 28: 123-127
        • Ruan D.
        • et al.
        Real-time profiling of respiratory motion: baseline drift, frequency variation and fundamental pattern change.
        Phys Med Biol. 2009; 54: 4777-4792
        • Riblett M.J.
        • et al.
        Data-driven respiratory motion compensation for four-dimensional cone-beam computed tomography (4D-CBCT) using groupwise deformable registration.
        Med Phys. 2018; 45: 4471-4482