Detailed analysis of latencies in image-based dynamic MLC tracking

Improving respiratory signal prediction with a deep neural network and simple changes to the input and output data format

Objective.To improve respiratory gating accuracy and radiation treatment throughput, we developed a generalized model based on a deep neural network (DNN) for predicting any given patient's respiratory motion.Approach.Our model uses long short-term memory (LSTM) based on a recurrent neural network (RNN), and improves upon common techniques. The first improvement is that the data input is not a one-dimensional sequence, but two-dimensional block data. This shortens the input sequence length, reducing computation time. Second, the output is not a scalar, but a sequence prediction. This increases the amount of available data, allowing improved prediction accuracy. For training and evaluation of our model, 434 sets of real-time position management data were retrospectively collected from clinical studies. The data were separated in a ratio of 4:1, with the larger set used for training models and the remaining set used for testing. We measured the accuracy of respiratory signal prediction and amplitude-based gating with prediction windows equaling 133, 333, and 533 ms. This new model was compared with the original LSTM and a non-recurrent DNN model.Main results.The mean absolute errors with the prediction window at 133, 333 and 533 ms were 0.036, 0.084, 0.119 with our model; 0.049, 0.14, 0.246 with the original LSTM-based model; and 0.041, 0.119, 0.16 with the non-recurrent DNN model, respectively. The computation time were 0.66 ms with our model; 0.63 ms the original LSTM-based model; 1.60 ms the non-recurrent DNN model, respectively. The accuracies of amplitude-based gating with the same prediction window settings and a duty cycle of approximately 50% were 98.3%, 95.8% and 92.7% with our model, 97.6%, 93.9% and 87.2% with the original LSTM-based model; and 97.9%, 94.3% and 89.5% with the non-recurrent DNN model, respectively.Significance.Our RNN algorithm for respiratory signal prediction successfully estimated tumor positions. We believe it will be useful in respiratory signal prediction technology.

Explicitly encoding the cyclic nature of breathing signal allows for accurate breathing motion prediction in radiotherapy with minimal training data

Background and purpose

Active breathing motion management in radiotherapy consists of motion monitoring, quantification and mitigation. It is impacted by associated latencies of a few 100 ms. Artificial neural networks can successfully predict breathing motion and eliminate latencies. However, they require usually a large dataset for training. The objective of this work was to demonstrate that explicitly encoding the cyclic nature of the breathing signal into the training data enables significant reduction of training datasets which can be obtained from healthy volunteers.

Material and methods

Seventy surface scanner breathing signals from 25 healthy volunteers in anterior-posterior direction were used for training and validation (ratio 4:1) of long short-term memory models. The model performance was compared to a model using decomposition into phase, amplitude and a time-dependent baseline. Testing of the models was performed on 55 independent breathing signals in anterior-posterior direction from surface scanner (35 lung, 20 liver) of 30 patients with a mean breathing amplitude of (5.9 ± 6.7) mm.

Results

Using the decomposed breathing signal allowed for a reduction of the absolute root-mean square error (RMSE) from 0.34 mm to 0.12 mm during validation. Testing using patient data yielded an average absolute RMSE of the breathing signal of (0.16 ± 0.11) mm with a prediction horizon of 500 ms.

Conclusion

It was demonstrated that a motion prediction model can be trained with less than 100 datasets of healthy volunteers if breathing cycle parameters are considered. Applied to 55 patients, the model predicted breathing motion with a high accuracy.

Neural signals-based respiratory motion tracking: a proof-of-concept study

Objective.Respiratory motion tracking techniques can provide optimal treatment accuracy for thoracoabdominal radiotherapy and robotic surgery. However, conventional imaging-based respiratory motion tracking techniques are time-lagged owing to the system latency of medical linear accelerators and surgical robots. This study aims to investigate the precursor time of respiratory-related neural signals and analyze the potential of neural signals-based respiratory motion tracking.Approach.The neural signals and respiratory motion from eighteen healthy volunteers were acquired simultaneously using a 256-channel scalp electroencephalography (EEG) system. The neural signals were preprocessed using the MNE python package to extract respiratory-related EEG neural signals. Cross-correlation analysis was performed to assess the precursor time and cross-correlation coefficient between respiratory-related EEG neural signals and respiratory motion.Main results.Respiratory-related neural signals that precede the emergence of respiratory motion are detectable via non-invasive EEG. On average, the precursor time of respiratory-related EEG neural signals was 0.68 s. The representative cross-correlation coefficients between EEG neural signals and respiratory motion of the eighteen healthy subjects varied from 0.22 to 0.87.Significance.Our findings suggest that neural signals have the potential to compensate for the system latency of medical linear accelerators and surgical robots. This indicates that neural signals-based respiratory motion tracking is a potential promising solution to respiratory motion and could be useful in thoracoabdominal radiotherapy and robotic surgery.

Real-Time Motion Analysis With 4D Deep Learning for Ultrasound-Guided Radiotherapy

Motion compensation in radiation therapy is a challenging scenario that requires estimating and forecasting motion of tissue structures to deliver the target dose. Ultrasound offers direct imaging of tissue in real-time and is considered for image guidance in radiation therapy. Recently, fast volumetric ultrasound has gained traction, but motion analysis with such high-dimensional data remains difficult. While deep learning could bring many advantages, such as fast data processing and high performance, it remains unclear how to process sequences of hundreds of image volumes efficiently and effectively. We present a 4D deep learning approach for real-time motion estimation and forecasting using long-term 4D ultrasound data. Using motion traces acquired during radiation therapy combined with various tissue types, our results demonstrate that long-term motion estimation can be performed markerless with a tracking error of 0.35±0.2 mm and with an inference time of less than 5 ms. Also, we demonstrate forecasting directly from the image data up to 900 ms into the future. Overall, our findings highlight that 4D deep learning is a promising approach for motion analysis during radiotherapy.

Evaluation of real-time tumor contour prediction using LSTM networks for MR-guided radiotherapy

BACKGROUND AND PURPOSE

Magnetic resonance imaging guided radiotherapy (MRgRT) with deformable multileaf collimator (MLC) tracking would allow to tackle both rigid displacement and tumor deformation without prolonging treatment. However, the system latency must be accounted for by predicting future tumor contours in real-time. We compared the performance of three artificial intelligence (AI) algorithms based on long short-term memory (LSTM) modules for the prediction of 2D-contours 500ms into the future.

MATERIALS AND METHODS

Models were trained (52 patients, 3.1h of motion), validated (18 patients, 0.6h) and tested (18 patients, 1.1h) with cine MRs from patients treated at one institution. Additionally, we used three patients (2.9h) treated at another institution as second testing set. We implemented 1) a classical LSTM network (LSTM-shift) predicting tumor centroid positions in superior-inferior and anterior-posterior direction which are used to shift the last observed tumor contour. The LSTM-shift model was optimized both in an offline and online fashion. We also implemented 2) a convolutional LSTM model (ConvLSTM) to directly predict future tumor contours and 3) a convolutional LSTM combined with spatial transformer layers (ConvLSTM-STL) to predict displacement fields used to warp the last tumor contour.

RESULTS

The online LSTM-shift model was found to perform slightly better than the offline LSTM-shift and significantly better than the ConvLSTM and ConvLSTM-STL. It achieved a 50% Hausdorff distance of 1.2mm and 1.0mm for the two testing sets, respectively. Larger motion ranges were found to lead to more substantial performance differences across the models.

CONCLUSION

LSTM networks predicting future centroids and shifting the last tumor contour are the most suitable for tumor contour prediction. The obtained accuracy would allow to reduce residual tracking errors during MRgRT with deformable MLC-tracking.

MRI-LINAC: A transformative technology in radiation oncology

Advances in radiotherapy technologies have enabled more precise target guidance, improved treatment verification, and greater control and versatility in radiation delivery. Amongst the recent novel technologies, Magnetic Resonance Imaging (MRI) guided radiotherapy (MRgRT) may hold the greatest potential to improve the therapeutic gains of image-guided delivery of radiation dose. The ability of the MRI linear accelerator (LINAC) to image tumors and organs with on-table MRI, to manage organ motion and dose delivery in real-time, and to adapt the radiotherapy plan on the day of treatment while the patient is on the table are major advances relative to current conventional radiation treatments. These advanced techniques demand efficient coordination and communication between members of the treatment team. MRgRT could fundamentally transform the radiotherapy delivery process within radiation oncology centers through the reorganization of the patient and treatment team workflow process. However, the MRgRT technology currently is limited by accessibility due to the cost of capital investment and the time and personnel allocation needed for each fractional treatment and the unclear clinical benefit compared to conventional radiotherapy platforms. As the technology evolves and becomes more widely available, we present the case that MRgRT has the potential to become a widely utilized treatment platform and transform the radiation oncology treatment process just as earlier disruptive radiation therapy technologies have done.

First experimental demonstration of VMAT combined with MLC tracking for single and multi fraction lung SBRT on an MR-linac

BACKGROUND AND PURPOSE

VMAT is not currently available on MR-linacs but could maximize plan conformality. To mitigate respiration without compromising delivery efficiency, MRI-guided MLC tumour tracking was recently developed for the 1.5 T Unity MR-linac (Elekta AB, Stockholm, Sweden) in combination with IMRT. Here, we provide a first experimental demonstration of VMAT+MLC tracking for several lung SBRT indications.

MATERIALS AND METHODS

We created central patient and phantom VMAT plans (8×7.5 Gy, 2 arcs) and we created peripheral phantom plans (3×18 & 1×34 Gy, 4 arcs). A motion phantom mimicked subject-recorded respiratory motion (A‾=11 mm, f‾=0.33 Hz, drift‾=0.3 mm/min). This was monitored using 2D-cine MRI at 4 Hz to continuously realign the beam with the target. VMAT+MLC tracking performance was evaluated using 2D film dosimetry and a novel motion-encoded and time-resolved pseudo-3D dosimetry approach.

RESULTS

We found an MLC leaf and jaw end-to-end latency of 328.05(±3.78) ms and 317.33(±4.64) ms, which was mitigated by a predictor. The VMAT plans required maximum MLC speeds of 12.1 cm/s and MLC tracking superimposes an additional 1.48 cm/s. A local 2%/1 mm gamma analysis with a static measurement as reference, revealed pass-rates of 28-46% without MLC tracking and 88-100% with MLC tracking for the 2D film analysis. Similarly the pseudo-3D gamma passing-rates increased from 22-77% to 92-100%. The dose area histograms show that MLC tracking increased the GTV D_98% by 5-20% and the PTV D_95% by 7-24%, giving similar target coverage as their respective static reference.

CONCLUSION

MRI-guided VMAT+MLC tracking is technically feasible on the MR-linac and results in highly conformal dose distribution.

A hybrid 2D/4D-MRI methodology using simultaneous multislice imaging for radiotherapy guidance

Purpose

Respiratory motion management is important in abdominothoracic radiotherapy. Fast imaging of the tumor can facilitate multileaf collimator (MLC) tracking that allows for smaller treatment margins, while repeatedly imaging the full field‐of‐view is necessary for 4D dose accumulation. This study introduces a hybrid 2D/4D‐MRI methodology that can be used for simultaneous MLC tracking and dose accumulation on a 1.5 T Unity MR‐linac (Elekta AB, Stockholm, Sweden).

Methods

We developed a hybrid 2D/4D‐MRI methodology that uses a simultaneous multislice (SMS) accelerated MRI sequence, which acquires two coronal slices simultaneously and repeatedly cycles through slice positions over the image volume. As a result, the fast 2D imaging can be used prospectively for MLC tracking and the SMS slices can be sorted retrospectively into respiratory‐correlated 4D‐MRIs for dose accumulation. Data were acquired in five healthy volunteers with an SMS‐bTFE and SMS‐TSE MRI sequence. For each sequence, a prebeam dataset and a beam‐on dataset were acquired simulating the two phases of MR‐linac treatments. Prebeam data were used to generate a 4D‐based motion model and a reference mid‐position volume, while beam‐on data were used for real‐time motion extraction and reconstruction of beam‐on 4D‐MRIs. In addition, an in‐silico computational phantom was used for validation of the hybrid 2D/4D‐MRI methodology. MLC tracking experiments were performed with the developed methodology, for which real‐time SMS data reconstruction was enabled on the scanner. A 15‐beam 8× 7.5 Gy intensity‐modulated radiotherapy plan for lung stereotactic body radiotherapy with isotropic 3 mm GTV‐to‐PTV margins was created. Dosimetry experiments were performed using a 4D motion phantom. The latency between target motion and updating the radiation beam was determined and compensated. Local gamma analyses were performed to quantify dose differences compared to a static reference delivery, and dose area histograms (DAHs) were used to quantify the GTV and PTV coverage.

Results

In‐vivo data acquisition and MLC tracking experiments were successfully performed with the developed hybrid 2D/4D‐MRI methodology. Real‐time liver–lung interface motion estimation had a Pearson's correlation of 0.996 (in‐vivo) and 0.998 (in‐silico). A median (5th–95th percentile) error of 0.0 (−0.9 to 0.7) mm and 0.0 (−0.2 to 0.2) mm was found for real‐time motion estimation for in‐vivo and in‐silico, respectively. Target motion prediction beyond the liver–lung interface had a median root mean square error of 1.6 mm (in‐vivo) and 0.5 mm (in‐silico). Beam‐on 4D MRI reconstruction required a median amount of data equal to an acquisition time of 2:21–3:17 min, which was 20% less data compared to the prebeam‐derived 4D‐MRI. System latency was reduced from 501 ± 12 ms to −1 ± 3 ms (SMS‐TSE) and from 398 ± 10 ms to −10 ± 4 ms (SMS‐bTFE) by a linear regression prediction filter. The local gamma analysis agreed within −3.8% to 3.3% (SMS‐bTFE) and −5.3% to 10% (SMS‐TSE) with a reference MRI sequence. The DAHs revealed a relative D98% GTV coverage between 97% and 100% (SMS‐bTFE) and 100% and 101% (SMS‐TSE) compared to the static reference.

Conclusions

The presented 2D/4D‐MRI methodology demonstrated the potential for accurately extracting real‐time motion for MLC tracking in abdominothoracic radiotherapy, while simultaneously reconstructing contiguous respiratory‐correlated 4D‐MRIs for dose accumulation.

Offline and online LSTM networks for respiratory motion prediction in MR-guided radiotherapy

Objective. Gated beam delivery is the current clinical practice for respiratory motion compensation in MR-guided radiotherapy, and further research is ongoing to implement tracking. To manage intra-fractional motion using multileaf collimator tracking the total system latency needs to be accounted for in real-time. In this study, long short-term memory (LSTM) networks were optimized for the prediction of superior–inferior tumor centroid positions extracted from clinically acquired 2D cine MRIs. Approach. We used 88 patients treated at the University Hospital of the LMU Munich for training and validation (70 patients, 13.1 h), and for testing (18 patients, 3.0 h). Three patients treated at Fondazione Policlinico Universitario Agostino Gemelli were used as a second testing set (1.5 h). The performance of the LSTMs in terms of root mean square error (RMSE) was compared to baseline linear regression (LR) models for forecasted time spans of 250 ms, 500 ms and 750 ms. Both the LSTM and the LR were trained with offline (offline LSTM and offline LR) and online schemes (offline+online LSTM and online LR), the latter to allow for continuous adaptation to recent respiratory patterns. Main results. We found the offline+online LSTM to perform best for all investigated forecasts. Specifically, when predicting 500 ms ahead it achieved a mean RMSE of 1.20 mm and 1.00 mm, while the best performing LR model achieved a mean RMSE of 1.42 mm and 1.22 mm for the LMU and Gemelli testing set, respectively. Significance. This indicates that LSTM networks have potential as respiratory motion predictors and that continuous online re-optimization can enhance their performance.

Prediction of the motion of chest internal points using a recurrent neural network trained with real-time recurrent learning for latency compensation in lung cancer radiotherapy

During the radiotherapy treatment of patients with lung cancer, the radiation delivered to healthy tissue around the tumor needs to be minimized, which is difficult because of respiratory motion and the latency of linear accelerator (LINAC) systems. In the proposed study, we first use the Lucas-Kanade pyramidal optical flow algorithm to perform deformable image registration (DIR) of chest computed tomography (CT) scan images of four patients with lung cancer. We then track three internal points close to the lung tumor based on the previously computed deformation field and predict their position with a recurrent neural network (RNN) trained using real-time recurrent learning (RTRL) and gradient clipping. The breathing data is quite regular, sampled at approximately 2.5 Hz, and includes artificially added drift in the spine direction. The amplitude of the motion of the tracked points ranged from 12.0 mm to 22.7 mm. Finally, we propose a simple method for recovering and predicting three-dimensional (3D) tumor images from the tracked points and the initial tumor image, based on a linear correspondence model and the Nadaraya-Watson non-linear regression. The root-mean-square (RMS) error, maximum error and jitter corresponding to the RNN prediction on the test set were smaller than the same performance measures obtained with linear prediction and least mean squares (LMS). In particular, the maximum prediction error associated with the RNN, equal to 1.51 mm, is respectively 16.1% and 5.0% lower than the error given by a linear predictor and LMS. The average prediction time per time step with RTRL is equal to 119 ms, which is less than the 400 ms marker position sampling time. The tumor position in the predicted images appears visually correct, which is confirmed by the high mean cross-correlation between the original and predicted images, equal to 0.955. The standard deviation of the Gaussian kernel and the number of layers in the optical flow algorithm were the parameters having the most significant impact on registration performance. Their optimization led respectively to a 31.3% and 36.2% decrease in the registration error. Using only a single layer proved to be detrimental to the registration quality because tissue motion in the lower part of the lung has a high amplitude relative to the resolution of the CT scan images. The random initialization of the hidden units and the number of these hidden units were found to be the most important factors affecting the performance of the RNN. Increasing the number of hidden units from 15 to 250 led to a 56.3% decrease in the prediction error on the cross-validation data. Similarly, optimizing the standard deviation of the initial Gaussian distribution of the synaptic weights σ_init^RNN led to a 28.4% decrease in the prediction error on the cross-validation data, with the error minimized for σ_init^RNN=0.02 with the four patients.

Dosimetric evaluation of MRI-guided multi-leaf collimator tracking and trailing for lung stereotactic body radiation therapy

PURPOSE

The treatment margins for lung stereotactic body radiotherapy (SBRT) are often large to cover the tumor excursions resulting from respiration, such that underdosage of the tumor can be avoided. Magnetic resonance imaging (MRI)-guided multi-leaf collimator (MLC) tracking can potentially reduce the influence of respiration to allow for smaller treatment margins. However, tracking is accompanied by system latency that may induce residual tracking errors. Alternatively, a simpler mid-position delivery combined with trailing can be used. Trailing reduces influences of respiration by compensating for baseline motion, to potentially improve target coverage. In this study, we aim to show the feasibility of MRI-guided tracking and trailing to reduce influences of respiration during lung SBRT.

METHODS

We implemented MRI-guided tracking on the MR-linac using an Elekta research tracking interface to track tumor motion during intensity modulated radiotherapy (IMRT). A Quasar MRI 4 D phantom was used to generate Lujan motion ( cos 4 , 4 s period, 20 mm peak-to-peak amplitude) with and without 1.0 mm/min cranial drift. Phantom tumor positions were estimated from sagittal 2D cine-MRI (4 or 8 Hz) using cross-correlation-based template matching. To compensate the anticipated system latency, a linear ridge regression predictor was optimized for online MRI by comparing two predictor training approaches: training on multiple traces and training on a single trace. We created 15-beam clinical-grade lung SBRT plans for central targets (8 × 7.5 Gy) and peripheral targets (3 × 18 Gy) with different PTV margins for mid-position based motion management (3-5 mm) and for MLC tracking (3 mm). We used a film insert with a 3 cm spherical target to measure the spatial distribution and quantity of the delivered dose. A 1%/1 mm local gamma-analysis quantified dose differences between motion management strategies and reference cases. Additionally, a dose area histogram (DAH) revealed the target coverage relative to the reference scenario.

RESULTS

The prediction filter gain was on average 25% when trained on multiple traces and 44% when trained on a single trace. The filter reduced system latency from 313 ± 2 ms to 0 ± 5 ms for 4 Hz imaging and from 215 ± 3 ms to 3 ± 3 ms for 8 Hz. The local gamma analysis for the central delivery showed that tracking improved the gamma pass-rate from 23% to 96% for periodic motion and from 14% to 93% when baseline drift was applied. For the peripheral delivery during periodic motion, delivery pass-rates improved from 22% to 93%. Comparing mid-position delivery to trailing for periodic+drift motion increased the local gamma pass rate from 15% to 98% for a central delivery and from 8% to 98% for a peripheral delivery. Furthermore, the DAHs revealed a relative D 98 % GTV coverage of 101% and 97% compared to the reference scenario for, respectively, central and peripheral tracking of periodic+drift motion. For trailing, a relative D 98 % of 99% for central and 98% for peripheral trailing was found.

CONCLUSIONS

We provided a first experimental demonstration of the technical feasibility of MRI-guided MLC tracking and trailing for central and peripheral lung SBRT. Tracking maximizes the sparing of healthy tissue, while trailing is highly effective in mitigating baseline motion.

Target tracking accuracy and latency with different 4D ultrasound systems – a robotic phantom study

Ultrasound (US) imaging, in contrast to other image guidance techniques, offers the distinct advantage of providing volumetric image data in real-time (4D) without using ionizing radiation. The goal of this study was to perform the first quantitative comparison of three different 4D US systems with fast matrix array probes and real-time data streaming regarding their target tracking accuracy and system latency. Sinusoidal motion of varying amplitudes and frequencies was used to simulate breathing motion with a robotic arm and a static US phantom. US volumes and robot positions were acquired online and stored for retrospective analysis. A template matching approach was used for target localization in the US data. Target motion measured in US was compared to the reference trajectory performed by the robot to determine localization accuracy and system latency. Using the robotic setup, all investigated 4D US systems could detect a moving target with sub-millimeter accuracy. However, especially high system latency increased tracking errors substantially and should be compensated with prediction algorithms for respiratory motion compensation.

ReconSocket: a low-latency raw data streaming interface for real-time MRI-guided radiotherapy

With the recent advent of hybrid MRI-guided radiotherapy systems, continuous intra-fraction MR imaging for motion monitoring has become feasible. The ability to perform real-time custom image reconstructions is however often lacking. In this work we present a low-latency streaming solution, ReconSocket, which provides a real-time stream of k-space data from the magnetic resonance imaging (MRI) to custom reconstruction servers. We determined the performance of the data streaming by measuring the streaming latency (i.e. non-zero time delay due to data transfer and processing) and jitter (i.e. deviations from periodicity) using an ultra-fast 1D MRI acquisition of a moving phantom. Simultaneously, its position was recorded with near-zero time delay. The feasibility of low-latency custom reconstructions was tested by measuring the imaging latency (i.e. time delay between physical change and appearance of that change on the image) for several non-Cartesian 2D and 3D acquisitions using an in-house implemented reconstruction server. The measured streaming latency of the ReconSocket interface was [Formula: see text] ms. 98% of the incoming data packets arrived within a jitter range of 367 [Formula: see text]s. This shows that the ReconSocket interface can provide reliable real-time access to MRI data, acquired during the course of a MRI-guided radiotherapy fraction. The total imaging latency was measured to be 221 ms (2D) and 3889 ms (3D) for exemplary acquisitions, using the custom image reconstruction server. These imaging latencies are approximately equal to half of the temporal footprint (T _acq /2) of the respective 2D and 3D golden-angle radial sequences. For radial sequences, it was previously showed that T _acq /2 is the expected contribution of only the data acquisition to the total imaging latency. Indeed, the contribution of the non-Cartesian reconstruction to the total imaging latency was minor (<10%): 21 ms for 2D, 300 ms for 3D, indicating that the acquisition, i.e. the physical encoding of the image itself is the major contributor to the total imaging latency.

Evaluation of 2D and 3D ultrasound tracking algorithms and impact on ultrasound-guided liver radiotherapy margins

PURPOSE

Compensation for respiratory motion is important during abdominal cancer treatments. In this work we report the results of the 2015 MICCAI Challenge on Liver Ultrasound Tracking and extend the 2D results to relate them to clinical relevance in form of reducing treatment margins and hence sparing healthy tissues, while maintaining full duty cycle.

METHODS

We describe methodologies for estimating and temporally predicting respiratory liver motion from continuous ultrasound imaging, used during ultrasound-guided radiation therapy. Furthermore, we investigated the trade-off between tracking accuracy and runtime in combination with temporal prediction strategies and their impact on treatment margins.

RESULTS

Based on 2D ultrasound sequences from 39 volunteers, a mean tracking accuracy of 0.9 mm was achieved when combining the results from the 4 challenge submissions (1.2 to 3.3 mm). The two submissions for the 3D sequences from 14 volunteers provided mean accuracies of 1.7 and 1.8 mm. In combination with temporal prediction, using the faster (41 vs 228 ms) but less accurate (1.4 vs 0.9 mm) tracking method resulted in substantially reduced treatment margins (70% vs 39%) in contrast to mid-ventilation margins, as it avoided non-linear temporal prediction by keeping the treatment system latency low (150 vs 400 ms). Acceleration of the best tracking method would improve the margin reduction to 75%.

CONCLUSIONS

Liver motion estimation and prediction during free-breathing from 2D ultrasound images can substantially reduce the in-plane motion uncertainty and hence treatment margins. Employing an accurate tracking method while avoiding non-linear temporal prediction would be favorable. This approach has the potential to shorten treatment time compared to breath-hold and gated approaches, and increase treatment efficiency and safety.

Tumor Trailing for Liver SBRT on the MR-Linac

PURPOSE

Tumor trailing is a treatment delivery technique that continuously adjusts the beam aperture according to the last available time-averaged position of the target. This study investigates whether tumor trailing on a magnetic resonance (MR) linear accelerator (linac) can improve target coverage in liver stereotactic body radiation therapy (SBRT) in the case of baseline motion.

METHODS AND MATERIALS

For 17 patients with oligometastatic liver disease, midposition SBRT treatment plans (3 × 20 Gy, 11-beam intensity modulated radiotherapy) were created for the Elekta Unity MR-Linac. Treatment was simulated using an in-house-developed delivery emulator. Respiratory motion was modelled as the superposition of periodic motion (patient-specific amplitude, 4-second period) and the following baseline motion scenarios: a continuous linear drift (0.5 mm/min), (2) a single shift halfway through treatment (10 mm), (3) a periodic drift (amplitude: 5 mm, period: 5 minutes), or (4) MR imaging-measured baseline drifts. Delivered dose was calculated under full consideration of the patient and machine motion interplay. In addition, trailing was experimentally validated on the MR-Linac using a programmable motion phantom.

RESULTS

The average simulated delivery and beam-on times were 15.9 and 8.7 minutes, respectively. An imaging frequency of ≥1 Hz was deemed necessary for trailing. Trailing increased the median gross tumor volume D98% dose by 1.9 Gy (linear drift), 1.2 Gy (single shift), 0.7 Gy (periodic drift), and 0.5 to 1.5 Gy (measured drifts) per fraction, compared with a conventional delivery. In the phantom experiments, the 3%/2 mm local gamma pass rate nearly doubled to 98% when using trailing.

CONCLUSION

Tumor trailing on the MR-Linac restores target dose in liver SBRT in the case of baseline motion for the presented patient cohort.

Feasibility of predicting tumor motion using online data acquired during treatment and a generalized neural network optimized with offline patient tumor trajectories

PURPOSE

The accurate prediction of intrafraction lung tumor motion is required to compensate for system latency in image-guided adaptive radiotherapy systems. The goal of this study was to identify an optimal prediction model that has a short learning period so that prediction and adaptation can commence soon after treatment begins, and requires minimal reoptimization for individual patients. Specifically, the feasibility of predicting tumor position using a combination of a generalized (i.e., averaged) neural network, optimized using historical patient data (i.e., tumor trajectories) obtained offline, coupled with the use of real-time online tumor positions (obtained during treatment delivery) was examined.

METHODS

A 3-layer perceptron neural network was implemented to predict tumor motion for a prediction horizon of 650 ms. A backpropagation algorithm and batch gradient descent approach were used to train the model. Twenty-seven 1-min lung tumor motion samples (selected from a CyberKnife patient dataset) were sampled at a rate of 7.5 Hz (0.133 s) to emulate the frame rate of an electronic portal imaging device (EPID). A sliding temporal window was used to sample the data for learning. The sliding window length was set to be equivalent to the first breathing cycle detected from each trajectory. Performing a parametric sweep, an averaged error surface of mean square errors (MSE) was obtained from the prediction responses of seven trajectories used for the training of the model (Group 1). An optimal input data size and number of hidden neurons were selected to represent the generalized model. To evaluate the prediction performance of the generalized model on unseen data, twenty tumor traces (Group 2) that were not involved in the training of the model were used for the leave-one-out cross-validation purposes.

RESULTS

An input data size of 35 samples (4.6 s) and 20 hidden neurons were selected for the generalized neural network. An average sliding window length of 28 data samples was used. The average initial learning period prior to the availability of the first predicted tumor position was 8.53 ± 1.03 s. Average mean absolute error (MAE) of 0.59 ± 0.13 mm and 0.56 ± 0.18 mm were obtained from Groups 1 and 2, respectively, giving an overall MAE of 0.57 ± 0.17 mm. Average root-mean-square-error (RMSE) of 0.67 ± 0.36 for all the traces (0.76 ± 0.34 mm, Group 1 and 0.63 ± 0.36 mm, Group 2), is comparable to previously published results. Prediction errors are mainly due to the irregular periodicities between cycles. Since the errors from Groups 1 and 2 are within the same range, it demonstrates that this model can generalize and predict on unseen data.

CONCLUSIONS

This is a first attempt to use an averaged MSE error surface (obtained from the prediction of different patients' tumor trajectories) to determine the parameters of a generalized neural network. This network could be deployed as a plug-and-play predictor for tumor trajectory during treatment delivery, eliminating the need for optimizing individual networks with pretreatment patient data.

Magnitude, Impact, and Management of Respiration-induced Target Motion in Radiotherapy Treatment: A Comprehensive Review

Tumors in thoracic and upper abdomen regions such as lungs, liver, pancreas, esophagus, and breast move due to respiration. Respiration-induced motion introduces uncertainties in radiotherapy treatments of these sites and is regarded as a significant bottleneck in achieving highly conformal dose distributions. Recent developments in radiation therapy have resulted in (i) motion-encompassing, (ii) respiratory gating, and (iii) tracking methods for adapting the radiation beam aperture to account for the respiration-induced target motion. The purpose of this review is to discuss the magnitude, impact, and management of respiration-induced tumor motion.

Predictive local receptive fields based respiratory motion tracking for motion-adaptive radiotherapy

Extracranial robotic radiotherapy employs external markers and a correlation model to trace the tumor motion caused by the respiration. The real-time tracking of tumor motion however requires a prediction model to compensate the latencies induced by the software (image data acquisition and processing) and hardware (mechanical and kinematic) limitations of the treatment system. A new prediction algorithm based on local receptive fields extreme learning machines (pLRF-ELM) is proposed for respiratory motion prediction. All the existing respiratory motion prediction methods model the non-stationary respiratory motion traces directly to predict the future values. Unlike these existing methods, the pLRF-ELM performs prediction by modeling the higher-level features obtained by mapping the raw respiratory motion into the random feature space of ELM instead of directly modeling the raw respiratory motion. The developed method is evaluated using the dataset acquired from 31 patients for two horizons in-line with the latencies of treatment systems like CyberKnife. Results showed that pLRF-ELM is superior to that of existing prediction methods. Results further highlight that the abstracted higher-level features are suitable to approximate the nonlinear and non-stationary characteristics of respiratory motion for accurate prediction.

Multimodal chest surface motion data for respiratory and cardiovascular monitoring applications

Chest surface motion is of significant importance as it contains information of respiratory and cardiac systems together with the complex coupling between these two systems. Chest surface motion is not only critical in radiotherapy, but also useful in personalized systems for continuous cardiorespiratory monitoring. In this dataset, a multimodal setup is employed to simultaneously acquire cardiorespiratory signals. These signals include high-density trunk surface motion (from 16 distinct locations) with VICON motion capture system, nasal breathing from a thermal sensor, respiratory effort from a strain belt and electrocardiogram in lead-II configuration. This dataset contains 72 trials recorded from 11 participants with a cumulative duration of approximately 215 min under various conditions such as normal breathing, breath-hold, irregular breathing and post-exercise recovery. The presented dataset is not only useful for evaluating prediction algorithms for radiotherapy applications, but can also be employed for the development of techniques to evaluate the cardio-mechanics and hemodynamic parameters of chest surface motion.

Online 4D ultrasound guidance for real-time motion compensation by MLC tracking

PURPOSE

With the trend in radiotherapy moving toward dose escalation and hypofractionation, the need for highly accurate targeting increases. While MLC tracking is already being successfully used for motion compensation of moving targets in the prostate, current real-time target localization methods rely on repeated x-ray imaging and implanted fiducial markers or electromagnetic transponders rather than direct target visualization. In contrast, ultrasound imaging can yield volumetric data in real-time (3D + time = 4D) without ionizing radiation. The authors report the first results of combining these promising techniques-online 4D ultrasound guidance and MLC tracking-in a phantom.

METHODS

A software framework for real-time target localization was installed directly on a 4D ultrasound station and used to detect a 2 mm spherical lead marker inside a water tank. The lead marker was rigidly attached to a motion stage programmed to reproduce nine characteristic tumor trajectories chosen from large databases (five prostate, four lung). The 3D marker position detected by ultrasound was transferred to a computer program for MLC tracking at a rate of 21.3 Hz and used for real-time MLC aperture adaption on a conventional linear accelerator. The tracking system latency was measured using sinusoidal trajectories and compensated for by applying a kernel density prediction algorithm for the lung traces. To measure geometric accuracy, static anterior and lateral conformal fields as well as a 358° arc with a 10 cm circular aperture were delivered for each trajectory. The two-dimensional (2D) geometric tracking error was measured as the difference between marker position and MLC aperture center in continuously acquired portal images. For dosimetric evaluation, VMAT treatment plans with high and low modulation were delivered to a biplanar diode array dosimeter using the same trajectories. Dose measurements with and without MLC tracking were compared to a static reference dose using 3%/3 mm and 2%/2 mm γ-tests.

RESULTS

The overall tracking system latency was 172 ms. The mean 2D root-mean-square tracking error was 1.03 mm (0.80 mm prostate, 1.31 mm lung). MLC tracking improved the dose delivery in all cases with an overall reduction in the γ-failure rate of 91.2% (3%/3 mm) and 89.9% (2%/2 mm) compared to no motion compensation. Low modulation VMAT plans had no (3%/3 mm) or minimal (2%/2 mm) residual γ-failures while tracking reduced the γ-failure rate from 17.4% to 2.8% (3%/3 mm) and from 33.9% to 6.5% (2%/2 mm) for plans with high modulation.

CONCLUSIONS

Real-time 4D ultrasound tracking was successfully integrated with online MLC tracking for the first time. The developed framework showed an accuracy and latency comparable with other MLC tracking methods while holding the potential to measure and adapt to target motion, including rotation and deformation, noninvasively.

First evaluation of the feasibility of MLC tracking using ultrasound motion estimation

PURPOSE

To quantify the performance of the Clarity ultrasound (US) imaging system (Elekta AB, Stockholm, Sweden) for real-time dynamic multileaf collimator (MLC) tracking.

METHODS

The Clarity calibration and quality assurance phantom was mounted on a motion platform moving with a periodic sine wave trajectory. The detected position of a 30 mm hypoechogenic sphere within the phantom was continuously reported via Clarity's real-time streaming interface to an in-house tracking and delivery software and subsequently used to adapt the MLC aperture. A portal imager measured MV treatment field/MLC apertures and motion platform positions throughout each experiment to independently quantify system latency and geometric error. Based on the measured range of latency values, a prostate stereotactic body radiation therapy (SBRT) delivery was performed with three realistic motion trajectories. The dosimetric impact of system latency on MLC tracking was directly measured using a 3D dosimeter mounted on the motion platform.

RESULTS

For 2D US imaging, the overall system latency, including all delay times from the imaging and delivery chain, ranged from 392 to 424 ms depending on the lateral sector size. For 3D US imaging, the latency ranged from 566 to 1031 ms depending on the elevational sweep. The latency-corrected geometric root-mean squared error was below 0.75 mm (2D US) and below 1.75 mm (3D US). For the prostate SBRT delivery, the impact of a range of system latencies (400-1000 ms) on the MLC tracking performance was minimal in terms of gamma failure rate.

CONCLUSIONS

Real-time MLC tracking based on a noninvasive US input is technologically feasible. Current system latencies are higher than those for x-ray imaging systems, but US can provide full volumetric image data and the impact of system latency was measured to be small for a prostate SBRT case when using a US-like motion input.

Ensemble framework based real-time respiratory motion prediction for adaptive radiotherapy applications

Successful treatment of tumors with motion-adaptive radiotherapy requires accurate prediction of respiratory motion, ideally with a prediction horizon larger than the latency in radiotherapy system. Accurate prediction of respiratory motion is however a non-trivial task due to the presence of irregularities and intra-trace variabilities, such as baseline drift and temporal changes in fundamental frequency pattern. In this paper, to enhance the accuracy of the respiratory motion prediction, we propose a stacked regression ensemble framework that integrates heterogeneous respiratory motion prediction algorithms. We further address two crucial issues for developing a successful ensemble framework: (1) selection of appropriate prediction methods to ensemble (level-0 methods) among the best existing prediction methods; and (2) finding a suitable generalization approach that can successfully exploit the relative advantages of the chosen level-0 methods. The efficacy of the developed ensemble framework is assessed with real respiratory motion traces acquired from 31 patients undergoing treatment. Results show that the developed ensemble framework improves the prediction performance significantly compared to the best existing methods.

Effect of MLC tracking latency on conformal volumetric modulated arc therapy (VMAT) plans in 4D stereotactic lung treatment

BACKGROUND AND PURPOSE

The latency of a multileaf collimator (MLC) tracking system used to overcome respiratory motion causes misalignment of the treatment beam with respect to the gross tumour volume, which may result in reduced target coverage. This study investigates the magnitude of this effect.

MATERIAL AND METHODS

Simulated superior-inferior breathing motion was used to construct histograms of isocentre offset with respect to the gross tumour volume (GTV) for a variety of tracking latencies. Dose distributions for conformal volumetric modulated arc therapy (VMAT) arcs were then calculated at a range of offsets and summed according to these displacement histograms. The results were verified by delivering the plans to a Delta(4) phantom on a motion platform.

RESULTS

In the absence of an internal target margin, a tracking latency of 150 ms reduces the GTV D95% by approximately 2%. With a margin of 2 mm, the same drop in dose occurs for a tracking latency of 450 ms. Lung V(13Gy) is unaffected by a range of latencies. These results are supported by the phantom measurements.

CONCLUSIONS

Assuming that internal motion can be modelled by a rigid translation of the patient, MLC tracking of conformal VMAT can be effectively accomplished in the absence of an internal target margin for substantial breathing motion (4 s period and 20 mm peak-peak amplitude) so long as the system latency is less than 150 ms.

Real-time prediction of respiratory motion traces for radiotherapy with ensemble learning

In this paper, we introduce a hybrid method for prediction of respiratory motion to overcome the inherent delay in robotic radiosurgery while treating lung tumors. The hybrid method adopts least squares support vector machine (LS-SVM) based ensemble learning approach to exploit the relative advantages of the individual methods local circular motion (LCM) with extended Kalman filter (EKF) and autoregressive moving average (ARMA) model with fading memory Kalman filter (FMKF). The efficiency the proposed hybrid approach was assessed with the real respiratory motion traces of 31 patients while treating with CyberKnife(TM). Results show that the proposed hybrid method improves the prediction accuracy by approximately 10% for prediction horizons of 460 ms compared to the existing methods.

Dynamic tumor tracking using the Elekta Agility MLC

PURPOSE

To evaluate the performance of the Elekta Agility multileaf collimator (MLC) for dynamic real-time tumor tracking.

METHODS

The authors have developed a new control software which interfaces to the Agility MLC to dynamically program the movement of individual leaves, the dynamic leaf guides (DLGs), and the Y collimators ("jaws") based on the actual target trajectory. A motion platform was used to perform dynamic tracking experiments with sinusoidal trajectories. The actual target positions reported by the motion platform at 20, 30, or 40 Hz were used as shift vectors for the MLC in beams-eye-view. The system latency of the MLC (i.e., the average latency comprising target device reporting latencies and MLC adjustment latency) and the geometric tracking accuracy were extracted from a sequence of MV portal images acquired during irradiation for the following treatment scenarios: leaf-only motion, jaw + leaf motion, and DLG + leaf motion.

RESULTS

The portal imager measurements indicated a clear dependence of the system latency on the target position reporting frequency. Deducting the effect of the target frequency, the leaf adjustment latency was measured to be 38 ± 3 ms for a maximum target speed v of 13 mm/s. The jaw + leaf adjustment latency was 53 ± 3 at a similar speed. The system latency at a target position frequency of 30 Hz was in the range of 56-61 ms for the leaves (v ≤ 31 mm/s), 71-78 ms for the jaw + leaf motion (v ≤ 25 mm/s), and 58-72 ms for the DLG + leaf motion (v ≤ 59 mm/s). The tracking accuracy showed a similar dependency on the target position frequency and the maximum target speed. For the leaves, the root-mean-squared error (RMSE) was between 0.6-1.5 mm depending on the maximum target speed. For the jaw + leaf (DLG + leaf) motion, the RMSE was between 0.7-1.5 mm (1.9-3.4 mm).

CONCLUSIONS

The authors have measured the latency and geometric accuracy of the Agility MLC, facilitating its future use for clinical tracking applications.

Automated marker tracking using noisy X-ray images degraded by the treatment beam

This study demonstrates the feasibility of automated marker tracking for the real-time detection of intrafractional target motion using noisy kilovoltage (kV) X-ray images degraded by the megavoltage (MV) treatment beam. The authors previously introduced the in-line imaging geometry, in which the flat-panel detector (FPD) is mounted directly underneath the treatment head of the linear accelerator. They found that the 121 kVp image quality was severely compromised by the 6 MV beam passing through the FPD at the same time. Specific MV-induced artefacts present a considerable challenge for automated marker detection algorithms. For this study, the authors developed a new imaging geometry by re-positioning the FPD and the X-ray tube. This improved the contrast-to-noise-ratio between 40% and 72% at the 1.2 mAs/image exposure setting. The increase in image quality clearly facilitates the quick and stable detection of motion with the aid of a template matching algorithm. The setup was tested with an anthropomorphic lung phantom (including an artificial lung tumour). In the tumour one or three Calypso beacons were embedded to achieve better contrast during MV radiation. For a single beacon, image acquisition and automated marker detection typically took around 76 ± 6 ms. The success rate was found to be highly dependent on imaging dose and gantry angle. To eliminate possible false detections, the authors implemented a training phase prior to treatment beam irradiation and also introduced speed limits for motion between subsequent images.

Collimator based tracking with an add-on multileaf collimator: Moduleaf

Radiotherapy is one of the most important methods used for the treatment of cancer. Irradiating a moving target is also one of the most challenging tasks to accomplish in modern radiotherapy. We have developed a tracking system by modifying an add-on collimator, the Siemens Moduleaf, for realtime applications in radiotherapy. As the add-on collimator works nearly completely independently of the linear accelerator (LinAc), no modifications to the latter were necessary. The adaptations to the Moduleaf were mainly software-based. In order to reduce the complexity of the system, outdated electronic parts were replaced with newer components where practical.Verification was performed by measuring the latency of the system as well as the impact on applied dose to a predefined target volume, moving in the leaf's travel direction. Latency measurements in the software were accomplished by comparing the target and current positions of the leaves. For dose measurements, a Gafchromic EBT2 film was placed beneath the target 4D phantom, in between solid water plates and moved alongside with it. Comparing the dose distribution on the film with a moving target between 'tracking disabled' towards 'tracking enabled' functions resulted in penumbra widths of 23 mm to 4 mm for 0.1 Hz sinusoidal movements with an amplitude of 32 mm, respectively. The maximum speed was therefore 20 mm s(-1). Latency was measured to be less than 50 ms for the signal runtimes. Based on the results, a tracking-capable add-on collimator seems to be a useful tool for reducing the margins for the treatment of small, slow-moving targets.

Toward the development of intrafraction tumor deformation tracking using a dynamic multi-leaf collimator

PURPOSE

Intrafraction deformation limits targeting accuracy in radiotherapy. Studies show tumor deformation of over 10 mm for both single tumor deformation and system deformation (due to differential motion between primary tumors and involved lymph nodes). Such deformation cannot be adapted to with current radiotherapy methods. The objective of this study was to develop and experimentally investigate the ability of a dynamic multi-leaf collimator (DMLC) tracking system to account for tumor deformation.

METHODS

To compensate for tumor deformation, the DMLC tracking strategy is to warp the planned beam aperture directly to conform to the new tumor shape based on real time tumor deformation input. Two deformable phantoms that correspond to a single tumor and a tumor system were developed. The planar deformations derived from the phantom images in beam's eye view were used to guide the aperture warping. An in-house deformable image registration software was developed to automatically trigger the registration once new target image was acquired and send the computed deformation to the DMLC tracking software. Because the registration speed is not fast enough to implement the experiment in real-time manner, the phantom deformation only proceeded to the next position until registration of the current deformation position was completed. The deformation tracking accuracy was evaluated by a geometric target coverage metric defined as the sum of the area incorrectly outside and inside the ideal aperture. The individual contributions from the deformable registration algorithm and the finite leaf width to the tracking uncertainty were analyzed. Clinical proof-of-principle experiment of deformation tracking using previously acquired MR images of a lung cancer patient was implemented to represent the MRI-Linac environment. Intensity-modulated radiation therapy (IMRT) treatment delivered with enabled deformation tracking was simulated and demonstrated.

RESULTS

The first experimental investigation of adapting to tumor deformation has been performed using simple deformable phantoms. For the single tumor deformation, the A(u)+A(o) was reduced over 56% when deformation was larger than 2 mm. Overall, the total improvement was 82%. For the tumor system deformation, the A(u)+A(o) reductions were all above 75% and the total A(u)+A(o) improvement was 86%. Similar coverage improvement was also found in simulating deformation tracking during IMRT delivery. The deformable image registration algorithm was identified as the dominant contributor to the tracking error rather than the finite leaf width. The discrepancy between the warped beam shape and the ideal beam shape due to the deformable registration was observed to be partially compensated during leaf fitting due to the finite leaf width. The clinical proof-of-principle experiment demonstrated the feasibility of intrafraction deformable tracking for clinical scenarios.

CONCLUSIONS

For the first time, we developed and demonstrated an experimental system that is capable of adapting the MLC aperture to account for tumor deformation. This work provides a potentially widely available management method to effectively account for intrafractional tumor deformation. This proof-of-principle study is the first experimental step toward the development of an image-guided radiotherapy system to treat deforming tumors in real-time.

Use of dMLC for implementation of dynamic respiratory-gated radiation therapy

PURPOSE

To simulate and evaluate the use of dynamic multileaf collimators (dMLC) in respiratory gating to compensate for baseline drift.

METHODS

Tumor motion tracking data from 30 lung tumors over 322 treatment fractions was analyzed with the finite state model. A dynamic respiratory gating window was established in real-time by determining the average positions during the previous two end-of-expiration breathing phases and centering the dMLC aperture on a weighted average of these positions. A simulated dMLC with physical motion constraints was used in dynamic gating treatment simulations. Fluence maps were created to provide a statistical description of radiation delivery for each fraction. Duty cycle was also calculated for each fraction.

RESULTS

The average duty cycle was 2.3% greater under dynamic gating conditions. Dynamic gating also showed higher fluences and less tumor obstruction. Additionally, dynamic gating required fewer beam toggles and each delivery period was longer on average than with static gating.

CONCLUSIONS

The use of dynamic gating showed better performance than static gating and the physical constraints of a dMLC were shown to not be an impediment to dynamic gating.

IMRT treatment planning on 4D geometries for the era of dynamic MLC tracking

The problem addressed here was to obtain optimal and deliverable dynamic multileaf collimator (MLC) leaf sequences from four-dimensional (4D) geometries for dynamic MLC tracking delivery. The envisaged scenario was where respiratory phase and position information of the target was available during treatment, from which the optimal treatment plan could be further adapted in real time. A tool for 4D treatment plan optimization was developed that integrates a commercially available treatment planning system and a general-purpose optimization system. The 4D planning method was applied to the 4D computed tomography planning scans of three lung cancer patients. The optimization variables were MLC leaf positions as a function of monitor units and respiratory phase. The objective function was the deformable dose-summed 4D treatment plan score. MLC leaf motion was constrained by the maximum leaf velocity between control points in terms of monitor units for tumor motion parallel to the leaf travel direction and between phases for tumor motion parallel to the leaf travel direction. For comparison and a starting point for the 4D optimization, three-dimensional (3D) optimization was performed on each of the phases. The output of the 4D IMRT planning process is a leaf sequence which is a function of both monitor unit and phase, which can be delivered to a patient whose breathing may vary between the imaging and treatment sessions. The 4D treatment plan score improved during 4D optimization by 34%, 4%, and 50% for Patients A, B, and C, respectively, indicating 4D optimization generated a better 4D treatment plan than the deformable sum of individually optimized phase plans. The dose-volume histograms for each phase remained similar, indicating robustness of the 4D treatment plan to respiratory variations expected during treatment delivery. In summary, 4D optimization for respiratory phase-dependent treatment planning with dynamic MLC motion tracking improved the 4D treatment plan score by 4–50% compared with 3D optimization. The 4D treatment plans had leaf sequences that varied from phase to phase to account for anatomic motion, but showed similar target dose distributions in each phase. The current method could in principle be generalized for use in offline replanning between fractions or for online 4D treatment planning based on 4D cone-beam CT images. Computation time remains a challenge.

Audiovisual biofeedback improves motion prediction accuracy

PURPOSE

The accuracy of motion prediction, utilized to overcome the system latency of motion management radiotherapy systems, is hampered by irregularities present in the patients' respiratory pattern. Audiovisual (AV) biofeedback has been shown to reduce respiratory irregularities. The aim of this study was to test the hypothesis that AV biofeedback improves the accuracy of motion prediction.

METHODS

An AV biofeedback system combined with real-time respiratory data acquisition and MR images were implemented in this project. One-dimensional respiratory data from (1) the abdominal wall (30 Hz) and (2) the thoracic diaphragm (5 Hz) were obtained from 15 healthy human subjects across 30 studies. The subjects were required to breathe with and without the guidance of AV biofeedback during each study. The obtained respiratory signals were then implemented in a kernel density estimation prediction algorithm. For each of the 30 studies, five different prediction times ranging from 50 to 1400 ms were tested (150 predictions performed). Prediction error was quantified as the root mean square error (RMSE); the RMSE was calculated from the difference between the real and predicted respiratory data. The statistical significance of the prediction results was determined by the Student's t-test.

RESULTS

Prediction accuracy was considerably improved by the implementation of AV biofeedback. Of the 150 respiratory predictions performed, prediction accuracy was improved 69% (103/150) of the time for abdominal wall data, and 78% (117/150) of the time for diaphragm data. The average reduction in RMSE due to AV biofeedback over unguided respiration was 26% (p < 0.001) and 29% (p < 0.001) for abdominal wall and diaphragm respiratory motion, respectively.

CONCLUSIONS

This study was the first to demonstrate that the reduction of respiratory irregularities due to the implementation of AV biofeedback improves prediction accuracy. This would result in increased efficiency of motion management techniques affected by system latencies used in radiotherapy.

Customized prediction of respiratory motion with clustering from multiple patient interaction

Information processing of radiotherapy systems has become an important research area for sophisticated radiation treatment methodology. Geometrically precise delivery of radiotherapy in the thorax and upper abdomen is compromised by respiratory motion during treatment. Accurate prediction of the respiratory motion would be beneficial for improving tumor targeting. However, a wide variety of breathing patterns can make it difficult to predict the breathing motion with explicit models. We proposed a respiratory motion predictor, that is, customized prediction with multiple patient interactions using neural network (CNN). For the preprocedure of prediction for individual patient, we construct the clustering based on breathing patterns of multiple patients using the feature selection metrics that are composed of a variety of breathing features. In the intraprocedure, the proposed CNN used neural networks (NN) for a part of the prediction and the extended Kalman filter (EKF) for a part of the correction. The prediction accuracy of the proposed method was investigated with a variety of prediction time horizons using normalized root mean squared error (NRMSE) values in comparison with the alternate recurrent neural network (RNN). We have also evaluated the prediction accuracy using the marginal value that can be used as the reference value to judge how many signals lie outside the confidence level. The experimental results showed that the proposed CNN can outperform RNN with respect to the prediction accuracy with an improvement of 50%.

A time-varying seasonal autoregressive model-based prediction of respiratory motion for tumor following radiotherapy

To achieve a better therapeutic effect and suppress side effects for lung cancer treatments, latency involved in current radiotherapy devices is aimed to be compensated for improving accuracy of continuous (not gating) irradiation to a respiratory moving tumor. A novel prediction method of lung tumor motion is developed for compensating the latency. An essential core of the method is to extract information valuable for the prediction, that is, the periodic nature inherent in respiratory motion. A seasonal autoregressive model useful to represent periodic motion has been extended to take into account the fluctuation of periodic nature in respiratory motion. The extended model estimates the fluctuation by using a correlation-based analysis for adaptation. The prediction performance of the proposed method was evaluated by using data sets of actual tumor motion and compared with those of the state-of-the-art methods. The proposed method demonstrated a high performance within submillimeter accuracy. That is, the average error of 1.0 s ahead predictions was 0.931 ± 0.055 mm. The accuracy achieved by the proposed method was the best among those by the others. The results suggest that the method can compensate the latency with sufficient accuracy for clinical use and contribute to improve the irradiation accuracy to the moving tumor.

Markerless EPID image guided dynamic multi-leaf collimator tracking for lung tumors

Compensation of target motion during the delivery of radiotherapy has the potential to improve treatment accuracy, dose conformity and sparing of healthy tissue. We implement an online image guided therapy system based on soft tissue localization (STiL) of the target from electronic portal images and treatment aperture adaptation with a dynamic multi-leaf collimator (DMLC). The treatment aperture is moved synchronously and in real time with the tumor during the entire breathing cycle. The system is implemented and tested on a Varian TX clinical linear accelerator featuring an AS-1000 electronic portal imaging device (EPID) acquiring images at a frame rate of 12.86 Hz throughout the treatment. A position update cycle for the treatment aperture consists of four steps: in the first step at time t = t0 a frame is grabbed, in the second step the frame is processed with the STiL algorithm to get the tumor position at t = t0, in a third step the tumor position at t = ti + δt is predicted to overcome system latencies and in the fourth step, the DMLC control software calculates the required leaf motions and applies them at time t = ti + δt. The prediction model is trained before the start of the treatment with data representing the tumor motion. We analyze the system latency with a dynamic chest phantom (4D motion phantom, Washington University). We estimate the average planar position deviation between target and treatment aperture in a clinical setting by driving the phantom with several lung tumor trajectories (recorded from fiducial tracking during radiotherapy delivery to the lung). DMLC tracking for lung stereotactic body radiation therapy without fiducial markers was successfully demonstrated. The inherent system latency is found to be δt = (230 ± 11) ms for a MV portal image acquisition frame rate of 12.86 Hz. The root mean square deviation between tumor and aperture position is smaller than 1 mm. We demonstrate the feasibility of real-time markerless DMLC tracking with a standard LINAC-mounted (EPID).

Implementation and experimental results of 4D tumor tracking using robotic couch

PURPOSE

This study presents the implementation and experimental results of a novel technique for 4D tumor tracking using a commercially available and commonly used treatment couch and evaluates the tumor tracking accuracy in clinical settings.

METHODS

Commercially available couch is capable of positioning the patient accurately; however, currently there is no provision for compensating physiological movement using the treatment couch in real-time. In this paper, a real-time couch tracking control technique is presented together with experimental results in tumor motion compensation in four dimensions (superior-inferior, lateral, anterior-posterior, and time). To implement real-time couch motion for tracking, a novel control system for the treatment couch was developed. The primary functional requirements for this novel technique were: (a) the treatment couch should maintain all previous∕normal features for patient setup and positioning, (b) the new control system should be used as a parallel system when tumor tracking would be deployed, and (c) tracking could be performed in a single direction and∕or concurrently in all three directions of the couch motion (longitudinal, lateral, and vertical). To the authors' best knowledge, the implementation of such technique to a regular treatment couch for tumor tracking has not been reported so far. To evaluate the performance of the tracking couch, we investigated the mechanical characteristics of the system such as system positioning resolution, repeatability, accuracy, and tracking performance. Performance of the tracking system was evaluated using dosimetric test as an endpoint. To investigate the accuracy of real-time tracking in the clinical setting, the existing clinical treatment couch was replaced with our experimental couch and the linear accelerator was used to deliver 3D conformal radiation therapy (3D-CRT) and intensity modulated radiation therapy (IMRT) treatment plans with and without tracking. The results of radiation dose distribution from these two sets of experiments were compared and presented here.

RESULTS

The mechanical accuracies were 0.12, 0.14, and 0.18 mm in X, Y, and Z directions. The repeatability of the desired motion was within ±0.2 mm. The differences of central axis dose between the 3D-CRT stationary plan and two tracking plans with different motion trajectories were 0.21% and 1.19%. The absolute dose differences of both 3D tracking plans comparing to the stationary plan were 1.09% and 1.20%. Comparing the stationary IMRT plan with the tracking IMRT plan, it was observed that the central axis dose difference was -0.87% and the absolute difference of both IMRT plans was 0.55%.

CONCLUSIONS

The experimental results revealed that the treatment couch could be successfully used for real-time tumor tracking with a high level of accuracy. It was demonstrated that 4D tumor tracking was feasible using existing couch with implementation of appropriate tracking methodology and with modifications in the control system.

Four dimensional CT imaging: a review of current technologies and modalities

Organ motion is a substantial concern in the treatment of thoracic tumours using radiotherapy. A number of technologies have evolved in order to address this both during computed tomography (CT) imaging and radiation delivery. This review paper investigates the various technologies which have been developed in the field of CT scanning as well as their accuracy, cost and the implications of their clinical implementation. The scanning modalities covered include: slow CT, breath hold CT, gated CT and retrospectively correlated CT (4DCT). It was found that there are advantages and drawbacks to each of the mentioned techniques relating to patient dose, scan time, extra equipment and workload. Also some scanning techniques are only compatible with certain treatment modalities which would further influence the decision as to which technologies to implement.

Image-based dynamic multileaf collimator tracking of moving targets during intensity-modulated arc therapy

PURPOSE

Intensity-modulated arc therapy (IMAT) enables efficient and highly conformal dose delivery. However, intrafraction motion may compromise the delivered target dose distribution. Dynamic multileaf collimator (DMLC) tracking can potentially mitigate the impact of target motion on the dose. The purpose of this study was to use a single kV imager for DMLC tracking during IMAT and to investigate the ability of this tracking to maintain the dose distribution.

METHODS

A motion phantom carrying a two-dimensional (2D) ion chamber array and buildup material with an embedded gold marker reproduced eight representative tumor trajectories (four lung tumors, four prostate). For each trajectory, a low and high IMAT plan were delivered with and without DMLC tracking. The three-dimensional (3D) real-time target position signal for tracking was provided by fluoroscopic kV images acquired immediately before and during treatment. For each image, the 3D position of the embedded marker was estimated from the imaged 2D position by a probability-based method. The MLC leaves were continuously refitted to the estimated 3D position. For lung, prediction was used to compensate for the tracking latency. The delivered 2D dose distributions were measured with the ion chamber array and compared with a reference dose distribution delivered without target motion using a 3%/3 mm γ-test.

RESULTS

For lung tumor motion, tracking reduced the mean γ-failure rate from 38% to 0.7% for low-modulation IMAT plans and from 44% to 2.8% for high-modulation plans. For prostate, the γ-failure rate reduction was from 19% to 0% (low modulation) and from 20% to 2.7% (high modulation). The dominant contributor to the residual γ-failures during tracking was target localization errors for most lung cases and leaf fitting errors for most prostate cases.

CONCLUSION

Image-based tracking for IMAT was demonstrated for the first time. The tracking greatly improved the dose distributions to moving targets.

Electromagnetic detection and real-time DMLC adaptation to target rotation during radiotherapy

PURPOSE

Intrafraction rotation of more than 45° and 25° has been observed for lung and prostate tumors, respectively. Such rotation is not routinely adapted to during current radiotherapy, which may compromise tumor dose coverage. The aim of the study was to investigate the geometric and dosimetric performance of an electromagnetically guided real-time dynamic multileaf collimator (DMLC) tracking system to adapt to intrafractional tumor rotation.

MATERIALS/METHODS

Target rotation was provided by changing the treatment couch angle. The target rotation was measured by a research Calypso system integrated with a real-time DMLC tracking system employed on a Varian linac. The geometric beam-target rotational alignment difference was measured using electronic portal images. The dosimetric accuracy was quantified using a two-dimensional ion chamber array. For each beam, the following five delivery modes were tested: 1) nonrotated target (reference); 2) fixed rotated target with tracking; 3) fixed rotated target without tracking; 4) actively rotating target with tracking; and 5) actively rotating target without tracking. Dosimetric performance of the latter four modes was measured and compared to the reference dose distribution using a 3 mm/3% γ-test.

RESULTS

Geometrically, the beam-target rotational alignment difference was 0.3° ± 0.6° for fixed rotation and 0.3° ± 1.3° for active rotation. Dosimetrically, the average failure rate for the γ-test for a fixed rotated target was 11% with tracking and 36% without tracking. The average failure rate for an actively rotating target was 9% with tracking and 35% without tracking.

CONCLUSIONS

For the first time, real-time target rotation has been accurately detected and adapted to during radiation delivery via DMLC tracking. The beam-target rotational alignment difference was mostly within 1°. Dose distributions to fixed and actively rotating targets with DMLC tracking were significantly superior to those without tracking.

Experimental investigation of a moving averaging algorithm for motion perpendicular to the leaf travel direction in dynamic MLC target tracking

PURPOSE

In dynamic multileaf collimator (MLC) motion tracking with complex intensity-modulated radiation therapy (IMRT) fields, target motion perpendicular to the MLC leaf travel direction can cause beam holds, which increase beam delivery time by up to a factor of 4. As a means to balance delivery efficiency and accuracy, a moving average algorithm was incorporated into a dynamic MLC motion tracking system (i.e., moving average tracking) to account for target motion perpendicular to the MLC leaf travel direction. The experimental investigation of the moving average algorithm compared with real-time tracking and no compensation beam delivery is described.

METHODS

The properties of the moving average algorithm were measured and compared with those of real-time tracking (dynamic MLC motion tracking accounting for both target motion parallel and perpendicular to the leaf travel direction) and no compensation beam delivery. The algorithm was investigated using a synthetic motion trace with a baseline drift and four patient-measured 3D tumor motion traces representing regular and irregular motions with varying baseline drifts. Each motion trace was reproduced by a moving platform. The delivery efficiency, geometric accuracy, and dosimetric accuracy were evaluated for conformal, step-and-shoot IMRT, and dynamic sliding window IMRT treatment plans using the synthetic and patient motion traces. The dosimetric accuracy was quantified via a tgamma-test with a 3%/3 mm criterion.

RESULTS

The delivery efficiency ranged from 89 to 100% for moving average tracking, 26%-100% for real-time tracking, and 100% (by definition) for no compensation. The root-mean-square geometric error ranged from 3.2 to 4.0 mm for moving average tracking, 0.7-1.1 mm for real-time tracking, and 3.7-7.2 mm for no compensation. The percentage of dosimetric points failing the gamma-test ranged from 4 to 30% for moving average tracking, 0%-23% for real-time tracking, and 10%-47% for no compensation.

CONCLUSIONS

The delivery efficiency of moving average tracking was up to four times higher than that of real-time tracking and approached the efficiency of no compensation for all cases. The geometric accuracy and dosimetric accuracy of the moving average algorithm was between real-time tracking and no compensation, approximately half the percentage of dosimetric points failing the gamma-test compared with no compensation.

Tracking latency in image-based dynamic MLC tracking with direct image access

PURPOSE

Target tracking is a promising method for motion compensation in radiotherapy. For image-based dynamic multileaf collimator (DMLC) tracking, latency has been shown to be the main contributor to geometrical errors in tracking of respiratory motion, specifically due to slow transfer of image data from the image acquisition system to the tracking system via image file storage on a hard disk. The purpose of the current study was to integrate direct image access with a DMLC tracking system and to quantify the tracking latency of the integrated system for both kV and MV image-based tracking.

METHOD

A DMLC tracking system integrated with a linear accelerator was used for tracking of a motion phantom with an embedded tungsten marker. Real-time target localization was based on x-ray images acquired either with a portal imager or a kV imager mounted orthogonal to the treatment beam. Images were processed directly without intermediate disk access. Continuous portal images and system log files were stored during treatment delivery for detailed offline analysis of the tracking latency.

RESULTS

The mean tracking system latency for kV and MV image-based tracking as function of the imaging interval ΔT(image) increased linearly with ΔT(image) as 148 ms + 0.58 * ΔT(image) (kV) and 162 ms + 1.1 * ΔT(image) (MV). The latency contribution from image acquisition and image transfer for kV image-based tracking was independent on ΔT(image) at 103 ± 14 ms. For MV-based tracking, it increased with ΔT(image) as 124 ms + 0.44 * ΔT(image). For ΔT(image) = 200 ms (5 Hz imaging), the total latency was reduced from 550 ms to 264 ms for kV image-based tracking and from 500 ms to 382 ms for MV image-based tracking as compared to the previously used indirect image transfer via image file storage on a hard disk.

CONCLUSION

kV and MV image-based DMLC tracking was successfully integrated with direct image access. It resulted in substantial tracking latency reductions compared with image-based tracking without direct image access.