J Rottmann

A multi-region algorithm for markerless beam's-eye view lung tumor tracking

Methods that allow online lung tumor tracking during radiotherapy are desirable for a variety of applications that have the potential to vastly improve treatment accuracy, dose conformity and sparing of healthy tissue. Several publications have proposed the use of an on-board kV x-ray imager to assess the tumor location during treatment. However, there is some concern that this strategy may expose the patient to a significant amount of additional dose over the course of a typical radiotherapy treatment. In this paper we present an algorithm that utilizes the on-board portal imager of the treatment machine to track lung tumors. This does not expose the patient to additional dose, but is somewhat more challenging as the quality of portal images is inferior when compared to kV x-ray images. To quantify the performance of the proposed algorithm we retrospectively applied it to portal image sequences retrieved from a dynamic chest phantom study and an SBRT treatment performed at our institution. The results were compared to manual tracking by an expert. For the phantom data the tracking error was found to be smaller than 1 mm and for the patient data smaller than 2 mm, which was in the same range as the uncertainty of the gold standard.

EPID-guided 3D dose verification of lung SBRT

PURPOSE To investigate the feasibility of utilizing tumor tracks from electronic portal imaging device (EPID) images taken during treatment to verify the delivered dose. METHODS The proposed method is based on a computation of the delivered fluence by utilizing the planned fluence and the tumor motion track for each field. A phantom study was designed to assess the feasibility of the method. The CIRS dynamic thorax phantom was utilized with a realistic soft resin tumor, modeled after a real patient tumor. The dose calculated with the proposed method was compared to direct measurements taken with 15 metal oxide semiconductor field effect transistors (MOSFETs) inserted in small fissures made in the tumor model. The phantom was irradiated with the tumor static and moved with different range of motions and setup errors. EPID images were recorded throughout all deliveries and the tumor model was tracked post-treatment with in-house developed software. The planned fluence for each field was convolved with the tumor motion tracks to obtain the delivered fluence. Utilizing the delivered fluence from each field, the delivered dose was calculated. The estimated delivered dose was compared to the dose directly measured with the MOSFETs. The feasibility of the proposed method was also demonstrated on a real lung cancer patient, treated with stereotactic body radiotherapy. RESULTS The calculation of delivered dose with the delivered fluence method was in good agreement with the MOSFET measurements, with average differences ranging from 0.8% to 8.3% depending on the proximity of a dose gradient. For the patient treatment, the planned and delivered dose volume histograms were compared and verified the overall good coverage of the target volume. CONCLUSIONS The delivered fluence method was applied successfully on phantom and clinical data and its accuracy was evaluated. Verifying each treatment fraction may enable correction strategies that can be applied during the course of treatment to ensure the desired dose coverage.

Multi-Region Tracking for Lung Tumor Motion Assessment

There is a need for a method of tracking lung tumors in beam’ s-eye-view MV image sequences without implanted radiopaque fiducials. We present a multi-region tracking algorithm to follow lung tumors on CT projections and intreatment portal image movies before and during external beam radiotherapy, respectively. Finding suitable landmarks for tracking is challenging due to low contrast in the images. We begin by defining a large set of landmark candidates and a sequence of training images representing the range of tumor motion. Each landmark is found automatically by seeking regions of maximum variance in the image gray values. Small, square templates are centered around each landmark to be used for tracking in sequential MV images. An iterative learning algorithm is employed to select the most suitable templates among the large collection of candidates for the training data set. This subset of templates is then applied to a similar data set for testing. The results of the automatic multi-template selection and tracking compare well to those of manually selected single template tracking. The algorithm shows great promise as a technique for automatically tracking lung tumors in beam’ s-eye-view in-treatment images without the need for implanted radiopaque fiducials.

TH‐EF‐BRB‐01: JUNIOR INVESTIGATOR WINNER — Novel EPID for Enhanced Contrast and Detective Quantum Efficiency

Purpose: Beams-eye-view imaging applications such as real-time soft-tissue motion estimation are hindered by the inherently low image contrast of electronic portal imaging devices (EPID) currently in clinical use. We introduce and characterize a novel EPID that provides substantially increased detective quantum efficiency (DQE), contrast-to-noise ratio (CNR) and dynamic range without degradation in spatial resolution. Methods: The prototype design features a stack of four conventional EPID layers combined with low noise integrated readout electronics. Each layer consists of a copper plate, a scintillator (GdO2S2:Tb) and a photodiode/TFT-switch (aSi:H). We characterize the prototype in terms of contrast-to-noise ratio (CNR), modulation transfer function (MTF) and DQE. CNR is estimated using a Las Vegas contrast phantom, presampled MTF is estimated using a slanted edge technique and the DQE is calculated from measured normalized noise power spectra (NPS) and the MTF. The prototype has been designed and built to be interchangeable with the current clinical EPID in terms of size and data output specifications. Results: DQE(0) can be nearly quadrupled to about 4.5% by using the four-layered design instead of only a single layer device. No substantial differences are observed between each layer’s individual MTF and the one for all four layers operating combined indicating that defocusing is negligible. Also, using four layers instead of one nearly doubles (factor x1.9) the CNR showing that the device is mostly quantum noise limited. Conclusion: A layered EPID design allows improving the radiation sensitivity while maintaining spatial resolution and saturation level of a single layer conventional EPID. Experimental characterization of this first 4-layered prototype demonstrates substantially improved DQE and CNR while maintaining high resolution and remaining quantum noise limited. Besides overall improved image quality and dosimetric sensitivity we anticipate this novel detector design to enable more accurate soft-tissue motion estimations during radiation therapy procedures, particularly of the lung. The project was partially supported by a grant from Varian Medical Systems, Inc. and grant No. R01CA188446-01 from the National Cancer Institute. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Cancer Institute or the National Institutes of Health.

TU‐F‐17A‐07: Real‐Time Personalized Margins

PURPOSE To maximize normal tissue sparing for treatments requiring motion encompassing margins. Motion mitigation techniques including DMLC or couch tracking can freeze tumor motion within the treatment aperture potentially allowing for smaller treatment margins and thus better sparing of normal tissue. To enable for a safe application of this concept in the clinic we propose adapting margins dynamically in real-time during radiotherapy delivery based on personalized tumor localization confidence. To demonstrate technical feasibility we present a phantom study. METHODS We utilize a realistic anthropomorphic dynamic thorax phantom with a lung tumor model embedded close to the spine. The tumor, a 3D-printout of a patients GTV, is moved 15mm peak-to-peak by diaphragm compression and monitored by continuous EPID imaging in real-time. Two treatment apertures are created for each beam, one representing ITV -based and the other GTV-based margin expansion. A soft tissue localization (STiL) algorithm utilizing the continuous EPID images is employed to freeze tumor motion within the treatment aperture by means of DMLC tracking. Depending on a tracking confidence measure (TCM), the treatment aperture is adjusted between the ITV and the GTV leaf. RESULTS We successfully demonstrate real-time personalized margin adjustment in a phantom study. We measured a system latency of about 250 ms which we compensated by utilizing a respiratory motion prediction algorithm (ridge regression). With prediction in place we observe tracking accuracies better than 1mm. For TCM=0 (as during startup) an ITV-based treatment aperture is chosen, for TCM=1 a GTV-based aperture and for 0<TCM<1 a linear interpolation between the two apertures is chosen. CONCLUSIONS We have proposed and demonstrated a concept to safely shrink treatment margins during radiotherapy and adapt to tracking confidence in real-time. Normal tissue sparing is maximized. The worst case scenario results in delivering a plan with standard margins used in the clinic today.

SU-E-J-112: The Impact of Cine EPID Image Acquisition Frame Rate On Markerless Soft-Tissue Tracking

PURPOSE Although reduction of the cine EPID acquisition frame rate through multiple frame averaging may reduce hardware memory burden and decrease image noise, it can hinder the continuity of soft-tissue motion leading to poor auto-tracking results. The impact of motion blurring and image noise on the tracking performance was investigated. METHODS Phantom and patient images were acquired at a frame rate of 12.87Hz on an AS1000 portal imager. Low frame rate images were obtained by continuous frame averaging. A previously validated tracking algorithm was employed for auto-tracking. The difference between the programmed and auto-tracked positions of a Las Vegas phantom moving in the superior-inferior direction defined the tracking error (δ). Motion blurring was assessed by measuring the area change of the circle with the greatest depth. Additionally, lung tumors on 1747 frames acquired at eleven field angles from four radiotherapy patients are manually and automatically tracked with varying frame averaging. δ was defined by the position difference of the two tracking methods. Image noise was defined as the standard deviation of the background intensity. Motion blurring and image noise were correlated with δ using Pearson correlation coefficient (R). RESULTS For both phantom and patient studies, the auto-tracking errors increased at frame rates lower than 4.29Hz. Above 4.29Hz, changes in errors were negligible with δ<1.60mm. Motion blurring and image noise were observed to increase and decrease with frame averaging, respectively. Motion blurring and tracking errors were significantly correlated for the phantom (R=0.94) and patient studies (R=0.72). Moderate to poor correlation was found between image noise and tracking error with R -0.58 and -0.19 for both studies, respectively. CONCLUSION An image acquisition frame rate of at least 4.29Hz is recommended for cine EPID tracking. Motion blurring in images with frame rates below 4.39Hz can substantially reduce the accuracy of auto-tracking. This work is supported in part by the Varian Medical Systems, Inc.

TU‐E‐204B‐03: A Multiregion Algorithm for Robust Lung Tumor Tracking with Portal Image Sequences

Purpose: Methods for online lungtumor tracking during radio‐therapy are desirable for a variety of applications that have the potential to vastly improve treatment accuracy, dose conformity and sparing of healthy tissue. Using the MV beam and portal imager for this purpose is preferable as there is no additional imaging dose and the output is the location of the target in the beams‐eye‐view. Method and Materials: We present a multiregion algorithm that utilizes the on‐board portal imager of the treatment machine to track lungtumors without implanted fiducial markers. Multiple regions with unique image structures are automatically identified and optimized based on an iterative procedure. Each region is tracked independently throughout the treatment, making the algorithm robust to target deformations, rotations and partial occlusions. To quantify the performance, the proposed algorithm was retrospectively applied to portal imagesequences from a dynamic chest phantom study as well as an SBRT treatment performed at our institution. The results were compared to manual tracking by an expert. Results: For the phantom data the tracking error was found to be smaller than 1mm and for the patient data smaller than 2mm, which was in the same range as the uncertainty of the reference. All optimized regions were able to be tracked on all images in both phantom and clinical sequences. The algorithm allows tracking of 15 regions at 2fps, the rate at which the images are clinically acquired. Conclusion: We have developed and tested a robust algorithm to track lungtumors on portal imagesequences without the use of fiducial markers or user intervention. The algorithm has the potential to be implemented in real time and is not dependent on prior information of the exact tumor motion range during treatment (unlike classification algorithms which generally require this). Conflict of Interest: Varian Medical Systems Inc.

WE-DE-BRA-04: A Cost-Effective Pixelated EPID Scintillator for Enhanced Contrast and DQE.

PURPOSE Beams-eye-view imaging applications such as real-time soft-tissue motion estimation and MV-CBCT are hindered by the inherently low image contrast of electronic portal imaging devices (EPID) currently in clinical use. We investigate a cost effective scintillating glass that provides substantially increased detective quantum efficiency (DQE) and contrast to noise ratio (CNR). METHODS A pixelated scintillator prototype was built from LKH-5 glass. The array is 12mm thick; 42.4×42.4cm2 wide and features 1.51mm pixel pitch with 20µm separation (glue+septa). The LKH-5 array was mounted on the active matrix flat panel imager (AMPFI) of an AS-1200 (Varian) with the GdO2S2:Tb removed. A second AS-1200 was utilized as reference detector. The prototype EPID was characterized in terms of CNR, modulation transfer function (MTF) and DQE. Additionally, the visibility of various fiducial markers typically used in the clinic as well as a realistic 3D-printed lung tumor model was assessed. All items were placed in a 12cm thick solid water phantom. CNR is estimated using a Las Vegas contrast phantom, presampled MTF is estimated using a slanted slit technique and the DQE is calculated from measured normalized noise power spectra (NPS) and the MTF. RESULTS DQE(0) for the LKH-5 prototype increased by a factor of 8× to about 10%, compared to the AS-1200 equipped with its standard GdO2S2:Tb scintillator. CNR increased by a factor of 5.3×. Due to the pixel size the MTF50 decreased by about 55% to 0.23lp/mm. The visibility of all fiducial markers as well as the tumor model were however markedly improved in comparison to an acquisition with the same parameters using the GdO2S2:Tb scintillator. CONCLUSION LKH-5 scintillating glasses allow the cost effective construction of thick pixelated scintillators for portal imaging which can yield a substantial increase in DQE and CNR. Soft tissue and fiducial marker visibility was found to be markedly improved. The project was supported in part by NIH grant R01CA188446-01 and a grant from Varian Medical Systems.

WE-G-207-06: 3D Fluoroscopic Image Generation From Patient-Specific 4DCBCT-Based Motion Models Derived From Physical Phantom and Clinical Patient Images

Purpose: Respiratory-correlated cone-beam CT (4DCBCT) images acquired immediately prior to treatment have the potential to represent patient motion patterns and anatomy during treatment, including both intra- and inter-fractional changes. We develop a method to generate patient-specific motion models based on 4DCBCT images acquired with existing clinical equipment and used to generate time varying volumetric images (3D fluoroscopic images) representing motion during treatment delivery. Methods: Motion models are derived by deformably registering each 4DCBCT phase to a reference phase, and performing principal component analysis (PCA) on the resulting displacement vector fields. 3D fluoroscopic images are estimated by optimizing the resulting PCA coefficients iteratively through comparison of the cone-beam projections simulating kV treatment imaging and digitally reconstructed radiographs generated from the motion model. Patient and physical phantom datasets are used to evaluate the method in terms of tumor localization error compared to manually defined ground truth positions. Results: 4DCBCT-based motion models were derived and used to generate 3D fluoroscopic images at treatment time. For the patient datasets, the average tumor localization error and the 95th percentile were 1.57 and 3.13 respectively in subsets of four patient datasets. For the physical phantom datasets, the average tumor localization error and the 95th percentile were 1.14 and 2.78 respectively in two datasets. 4DCBCT motion models are shown to perform well in the context of generating 3D fluoroscopic images due to their ability to reproduce anatomical changes at treatment time. Conclusion: This study showed the feasibility of deriving 4DCBCT-based motion models and using them to generate 3D fluoroscopic images at treatment time in real clinical settings. 4DCBCT-based motion models were found to account for the 3D non-rigid motion of the patient anatomy during treatment and have the potential to localize tumor and other patient anatomical structures at treatment time even when inter-fractional changes occur. This project was supported, in part, through a Master Research Agreement with Varian Medical Systems, Inc., Palo Alto, CA. The project was also supported, in part, by Award Number R21CA156068 from the National Cancer Institute.

SU‐E‐J‐228: Dose Accumulation Studies with a Dynamic Physical Anthropomorphic Phantom and An External Surrogate‐Based Motion Model

PURPOSE To estimate the dose delivered to a physical anthropomorphic phantom based on: 1) CT scans representing each phase of respiration; and 2) 3D images generated from a respiratory motion model and an external surrogate signal. OSLDs are used to measure doses delivered to the phantom. METHODS A commercially available physical phantom was modified, replacing the lung system with foam rubber of lung-equivalent density and placing a tumor made of bolus within the lung. A wooden diaphragm driven by a programmable motor compressed the foam with a realistic breathing pattern based on patient measurements. CT scans of the phantom were taken at several phases of a breathing cycle, and the dose delivered by a nine-field treatment at each phase was calculated with Monte Carlo. Dose distributions at each phase were mapped to a reference phase with vectors from deformable image registration performed on the original CT images. A second estimate of the delivered dose was performed replacing the CT scans and associated vectors with images and deformations generated by a motion model. Finally, with a one-field treatment plan, the estimated delivered dose was compared to measurements with OSLDs placed in the phantom. RESULTS The estimated dose delivered to the tumor using CT scans agreed with the estimate using model-generated images, and the difference in the D95 for the two approaches was less than 2%. This demonstrates that images generated by the motion model can be used for dose estimates. Dose measured with OSLDs at nine points within the tumor and foam lung of the phantom agreed with predictions within measurement uncertainties. CONCLUSION The images generated from a motion model based on an external surrogate trace can be used to estimate dose delivered during treatment. Dose estimates were validated with measurements.