•A comprehensive commissioning protocol is designed to assess the MRI performance of integrated MR-linac systems.•The imaging performance of four recently installed 1.5T MR-linac systems is assessed ...in a multi-institutional study.•A benchmark data set is provided by making the results as well as the corresponding measurement protocols publically available.
Magnetic Resonance linear accelerator (MR-linac) systems represent a new type of technology that allows for online MR-guidance for high precision radiotherapy (RT). Currently, the first MR-linac installations are being introduced clinically. Since the imaging performance of these integrated MR-linac systems is critical for their application, a thorough commissioning of the MRI performance is essential. However, guidelines on the commissioning of MR-guided RT systems are not yet defined and data on the performance of MR-linacs are not yet available.
Here we describe a comprehensive commissioning protocol, which contains standard MRI performance measurements as well as dedicated hybrid tests that specifically assess the interactions between the Linac and the MRI system. The commissioning results of four MR-linac systems are presented in a multi-center study.
Although the four systems showed similar performance in all the standard MRI performance tests, some differences were observed relating to the hybrid character of the systems. Field homogeneity measurements identified differences in the gantry shim configuration, which was later confirmed by the vendor.
Our results highlight the importance of dedicated hybrid commissioning tests and the ability to compare the machines between institutes at this very early stage of clinical introduction. Until formal guidelines and tolerances are defined the tests described in this study may be used as a practical guideline. Moreover, the multi-center results provide initial bench mark data for future MR-linac installations.
This work aimed to quantify any principal magnetic field (B
) inhomogeneity and changes in MR image geometric distortion with continuous linac gantry rotation on an Elekta Unity MR-linac. This ...situation occurs for around a second between treatment beams during current image guided radiotherapy treatment and would occur frequently in foreseeable real-time adaptive radiotherapy treatment. Pixel by pixel maps of B
inhomogeneity were obtained via repeated high temporal resolution pulse sequences with the linac gantry static at 36 gantry angles spaced ten degrees apart, and in continuous rotation at both 1 and 2 rpm. Individual B
maps were subtracted from average maps across all data and the residual peak to peak inhomogeneity was calculated for each. The bulk geometric shift and change in physical extent of a 10 cm diameter spherical flood phantom during continuous linac gantry rotation at 1 and 2 rpm was compared to the static gantry case for two pulse sequences: the real-time clinical monitoring bFFE sequence and a non-clinical EPI sequence, chosen for its susceptibility to geometric distortion. The peak to peak inhomogeneity in the deviation-from-average ppm maps, plotted against gantry angle with the gantry in continuous rotation at 1 and 2 rpm were negligibly different from equivalent data obtained with the gantry static. The real-time clinical monitoring pulse sequence was shown to give negligible geometric distortion during continuous gantry motion, whilst a non-clinical EPI sequence showed bulk shifts of the order of one pixel and gantry angle dependent changes in extent, demonstrating the sensitivity of the chosen method. MR imaging on the Elekta Unity MR-Linac with the gantry in continuous motion is negligibly different from the static gantry case with current clinical pulse sequences. Real-time tracking and treatment plan adaptation using MR images obtained with the linac gantry in motion is possible.
•5 patients with pelvic lymph node metastases received SBRT using a 1.5 T MR-linac.•Session time was <60 min for all 25 treatment fractions.•All quality assurance tests were passed (dose calculations ...& film measurements).
Online adaptive radiotherapy using the 1.5 Tesla MR-linac is feasible for SBRT (5 × 7 Gy) of pelvic lymph node oligometastases. The workflow allows full online planning based on daily anatomy. Session duration is less than 60 min. Quality assurance tests, including independent 3D dose calculations and film measurements were passed.
For quality assurance and adaptive radiotherapy, validation of the actual delivered dose is crucial.Intrafractional anatomy changes cannot be captured satisfactorily during treatment with hitherto ...available imaging modalitites. Consequently, dose calculations are based on the assumption of static anatomy throughout the treatment. However, intra- and interfraction anatomy is dynamic and changes can be significant.In this paper, we investigate the use of an MR-linac as a dose tracking modality for the validation of treatments in abdominal targets where both respiratory and long-term peristaltic and drift motion occur.The on-line MR imaging capability of the modality provides the means to perform respiratory gating of both delivery and acquisition yielding a model-free respiratory motion management under free breathing conditions.In parallel to the treatment, the volumetric patient anatomy was captured and used to calculate the applied dose. Subsequently, the individual doses were warped back to the planning grid to obtain the actual dose accumulated over the entire treatment duration. Ultimately, the planned dose was validated by comparison with the accumulated dose.Representative for a site subject to breathing modulation, two kidney cases (25 Gy target dose) demonstrated the working principle on volunteer data and simulated delivery. The proposed workflow successfully showed its ability to track local dosimetric changes. Integration of the on-line anatomy information could reveal local dose variations -2.3-1.5 Gy in the target volume of a volunteer dataset. In the adjacent organs at risk, high local dose errors ranging from -2.5 to 1.9 Gy could be traced back.
Image processing such as deformable image registration finds its way into radiotherapy as a means to track non-rigid anatomy. With the advent of magnetic resonance imaging (MRI) guided radiotherapy, ...intrafraction anatomy snapshots become technically feasible. MRI provides the needed tissue signal for high-fidelity image registration. However, acquisitions, especially in 3D, take a considerable amount of time. Pushing towards real-time adaptive radiotherapy, MRI needs to be accelerated without degrading the quality of information. In this paper, we investigate the impact of image resolution on the quality of motion estimations. Potentially, spatially undersampled images yield comparable motion estimations. At the same time, their acquisition times would reduce greatly due to the sparser sampling. In order to substantiate this hypothesis, exemplary 4D datasets of the abdomen were downsampled gradually. Subsequently, spatiotemporal deformations are extracted consistently using the same motion estimation for each downsampled dataset. Errors between the original and the respectively downsampled version of the dataset are then evaluated. Compared to ground-truth, results show high similarity of deformations estimated from downsampled image data. Using a dataset with (2.5 mm)3 voxel size, deformation fields could be recovered well up to a downsampling factor of 2, i.e. (5 mm)3. In a therapy guidance scenario MRI, imaging speed could accordingly increase approximately fourfold, with acceptable loss of estimated motion quality.
Finding and implementing a suitable machine learning (ML) solution to a task at hand has several facets. The technical side of ML has widely been discussed in detail, see, e. g., (Heizmann, M., A. ...Braun, M. Hüttel, C. Klüver, E. Marquardt, M. Overdick and M. Ulrich. 2020. Artificial Intelligence with Neural Networks in Optical Measurement and Inspection Systems. at – Automatisierungstechnik 68(6): 477–487). This contribution focusses on the industrial implementation issues of ML projects, particularly for machine vision (MV) tasks. Especially in small and medium-sized enterprises (SMEs), resources cannot be activated at will in order to use a new technology like ML. We take this into account by, on the one hand, helping to realistically evaluate the opportunities and challenges involved in implementing ML projects for a given task. On the other hand, we consider not only technical aspects, but also organizational, social and customer-related ones. It is discussed which know-how a company itself has to bring into an ML project and which tasks can also be performed by service providers. Here, it becomes clear that ML techniques can be used at different levels of detail. The question of “make or buy” is therefore also an entrepreneurial one when introducing ML into one’s own products and processes, and must be answered with a view to one’s own possibilities and structures.
In radiotherapy, abdominal and thoracic sites are candidates for performing motion tracking. With real-time control it is possible to adjust the multileaf collimator (MLC) position to the target ...position. However, positions are not perfectly matched and position errors arise from system delays and complicated response of the electromechanic MLC system. Although, it is possible to compensate parts of these errors by using predictors, residual errors remain and need to be compensated to retain target coverage. This work presents a method to statistically describe tracking errors and to automatically derive a patient-specific, per-segment margin to compensate the arising underdosage on-line, i.e. during plan delivery. The statistics of the geometric error between intended and actual machine position are derived using kernel density estimators. Subsequently a margin is calculated on-line according to a selected coverage parameter, which determines the amount of accepted underdosage. The margin is then applied onto the actual segment to accommodate the positioning errors in the enlarged segment. The proof-of-concept was tested in an on-line tracking experiment and showed the ability to recover underdosages for two test cases, increasing V90% in the underdosed area about 47% and 41%, respectively. The used dose model was able to predict the loss of dose due to tracking errors and could be used to infer the necessary margins. The implementation had a running time of 23 ms which is compatible with real-time requirements of MLC tracking systems. The auto-adaptivity to machine and patient characteristics makes the technique a generic yet intuitive candidate to avoid underdosages due to MLC tracking errors.
Stereotactic body radiation therapy (SBRT) has shown great promise in increasing local control rates for renal-cell carcinoma (RCC). Characterized by steep dose gradients and high fraction doses, ...these hypo-fractionated treatments are, however, prone to dosimetric errors as a result of variations in intra-fraction respiratory-induced motion, such as drifts and amplitude alterations. This may lead to significant variations in the deposited dose. This study aims to develop a method for calculating the accumulated dose for MRI-guided SBRT of RCC in the presence of intra-fraction respiratory variations and determine the effect of such variations on the deposited dose. For this, RCC SBRT treatments were simulated while the underlying anatomy was moving, based on motion information from three motion models with increasing complexity: (1) STATIC, in which static anatomy was assumed, (2) AVG-RESP, in which 4D-MRI phase-volumes were time-weighted, and (3) PCA, a method that generates 3D volumes with sufficient spatio-temporal resolution to capture respiration and intra-fraction variations. Five RCC patients and two volunteers were included and treatments delivery was simulated, using motion derived from subject-specific MR imaging. Motion was most accurately estimated using the PCA method with root-mean-squared errors of 2.7, 2.4, 1.0 mm for STATIC, AVG-RESP and PCA, respectively. The heterogeneous patient group demonstrated relatively large dosimetric differences between the STATIC and AVG-RESP, and the PCA reconstructed dose maps, with hotspots up to Formula: see text of the D99 and an underdosed GTV in three out of the five patients. This shows the potential importance of including intra-fraction motion variations in dose calculations.
Shift Variance in Scene Text Detection Glitzner, Markus; Neudeck, Jan-Hendrik; Härtinger, Philipp
arXiv (Cornell University),
08/2022
Paper, Journal Article
Odprti dostop
Theory of convolutional neural networks suggests the property of shift equivariance, i.e., that a shifted input causes an equally shifted output. In practice, however, this is not always the case. ...This poses a great problem for scene text detection for which a consistent spatial response is crucial, irrespective of the position of the text in the scene. Using a simple synthetic experiment, we demonstrate the inherent shift variance of a state-of-the-art fully convolutional text detector. Furthermore, using the same experimental setting, we show how small architectural changes can lead to an improved shift equivariance and less variation of the detector output. We validate the synthetic results using a real-world training schedule on the text detection network. To quantify the amount of shift variability, we propose a metric based on well-established text detection benchmarks. While the proposed architectural changes are not able to fully recover shift equivariance, adding smoothing filters can substantially improve shift consistency on common text datasets. Considering the potentially large impact of small shifts, we propose to extend the commonly used text detection metrics by the metric described in this work, in order to be able to quantify the consistency of text detectors.