K Malinowski

Incidence of Changes in Respiration-Induced Tumor Motion and Its Relationship With Respiratory Surrogates During Individual Treatment Fractions

PURPOSE To determine how frequently (1) tumor motion and (2) the spatial relationship between tumor and respiratory surrogate markers change during a treatment fraction in lung and pancreas cancer patients. METHODS AND MATERIALS A Cyberknife Synchrony system radiographically localized the tumor and simultaneously tracked three respiratory surrogate markers fixed to a form-fitting vest. Data in 55 lung and 29 pancreas fractions were divided into successive 10-min blocks. Mean tumor positions and tumor position distributions were compared across 10-min blocks of data. Treatment margins were calculated from both 10 and 30 min of data. Partial least squares (PLS) regression models of tumor positions as a function of external surrogate marker positions were created from the first 10 min of data in each fraction; the incidence of significant PLS model degradation was used to assess changes in the spatial relationship between tumors and surrogate markers. RESULTS The absolute change in mean tumor position from first to third 10-min blocks was >5 mm in 13% and 7% of lung and pancreas cases, respectively. Superior-inferior and medial-lateral differences in mean tumor position were significantly associated with the lobe of lung. In 61% and 54% of lung and pancreas fractions, respectively, margins calculated from 30 min of data were larger than margins calculated from 10 min of data. The change in treatment margin magnitude for superior-inferior motion was >1 mm in 42% of lung and 45% of pancreas fractions. Significantly increasing tumor position prediction model error (mean ± standard deviation rates of change of 1.6 ± 2.5 mm per 10 min) over 30 min indicated tumor-surrogate relationship changes in 63% of fractions. CONCLUSIONS Both tumor motion and the relationship between tumor and respiratory surrogate displacements change in most treatment fractions for patient in-room time of 30 min.

Mitigating Errors in External Respiratory Surrogate-Based Models of Tumor Position

PURPOSE To investigate the effect of tumor site, measurement precision, tumor-surrogate correlation, training data selection, model design, and interpatient and interfraction variations on the accuracy of external marker-based models of tumor position. METHODS AND MATERIALS Cyberknife Synchrony system log files comprising synchronously acquired positions of external markers and the tumor from 167 treatment fractions were analyzed. The accuracy of Synchrony, ordinary-least-squares regression, and partial-least-squares regression models for predicting the tumor position from the external markers was evaluated. The quantity and timing of the data used to build the predictive model were varied. The effects of tumor-surrogate correlation and the precision in both the tumor and the external surrogate position measurements were explored by adding noise to the data. RESULTS The tumor position prediction errors increased during the duration of a fraction. Increasing the training data quantities did not always lead to more accurate models. Adding uncorrelated noise to the external marker-based inputs degraded the tumor-surrogate correlation models by 16% for partial-least-squares and 57% for ordinary-least-squares. External marker and tumor position measurement errors led to tumor position prediction changes 0.3-3.6 times the magnitude of the measurement errors, varying widely with model algorithm. The tumor position prediction errors were significantly associated with the patient index but not with the fraction index or tumor site. Partial-least-squares was as accurate as Synchrony and more accurate than ordinary-least-squares. CONCLUSIONS The accuracy of surrogate-based inferential models of tumor position was affected by all the investigated factors, except for the tumor site and fraction index.

TU‐G‐BRA‐08: Understanding the Performance of Control Limit‐Based Monitoring of Respiratory Surrogate Tumor Motion Models

Purpose: To develop an understanding of mechanisms underlying the performance of a control‐limit‐based monitoring technique for detecting errors in respiratory surrogate tumor displacement models.Methods: Five lungcancer patients underwent 13 dynamic magnetic resonance imaging sessions on a 1.5 T scanner using a TrueFISP sequence (200 images, 5 sagittal slices, 8mm slice thickness, interleaved acquisition, TE 1.29 msec, TR 2.57 ms, 60° flip angle, matrix 176×256 matrix, in‐plane spatial resolution 1.6–2.2 mm each direction, 1028 bandwidth). Tumors were localized in the images at 0.4 Hz for 500 sec. Five respiratory surrogates affixed to the abdomen were optically tracked during imaging. Surrogate‐based tumor motion models were created by applying partial‐least‐squares regression to the first 30 sec of data. Hotellings statistic and the input variable squared‐prediction‐error for each subsequent sample were compared to training data‐based control limits to predict errors >3mm. The experiment was repeated in tumor motion and respiratory surrogate signal simulations that isolated measurement precision, period variations, amplitude variations, end‐exhale variations, tumor drift, and gross patient motion. Sampling rates of 0.1–30 Hz (0.1–0.4 Hz for patient data) were evaluated. Results: For patient data sampled at 0.4 Hz and 0.1 Hz and 95% sensitivity, specificities were 8% and 17%. In comprehensive simulations tuned to 95% sensitivity, specificity increased from 33% at 0.1 Hz to 69% at 30 Hz. With measurement noise or tumor drifts alone, specificities were 99–100% and sensitivities were 0–1%. Gross patient motion was detected with sensitivity of 100% and specificity of 97%. For end‐exhale variations, sensitivity was 97%, and specificity was 57%. Respiratory cycle amplitude and period variations had no effect on monitoring performance. Conclusions: Contributors to control‐limit‐based monitoring performance included sampling rate, end‐exhale variations, measurement noise, and tumor drifts but not patient motion, period variations, or amplitude variations. Simulation results were qualitatively in agreement with patient results. Supported in part by grant CA124766 from the NIH/NCI and the Achievement Rewards for College Scientists scholarship.

TU‐G‐BRC‐08: Maintaining Tumor Targeting Accuracy in Gating and Tracking Systems for Respiration‐Induced Tumor Motion

Purpose: To maintain accuracy, respiratory surrogate‐based intra‐fraction tumor motion models must be updated periodically. The purpose of this study was to determine how best to time respiratory surrogate‐based tumor motion model updates by comparing a novel statistical process control (SPC) method, based on external measurements alone, to three direct measurement methods in clinical use.Methods: Position datasets recorded during 121 treatment fractions from 61 lungcancer patients were analyzed. Datasets included 26 Hz localizations of three surrogate markers affixed to the torso as well as tumor localizations from intermittent (approximately once per minute) stereoscopic radiographs. Partial‐least‐squares regression models of tumor position from marker motion were created from six concurrent tumor localizations. At each radiographic localization, model accuracy was assessed and model rebuilding with the six most recent tumor localizations was considered. Model updates were timed according to four methods: (1) never, (2) when an SPC metric (either Hotellings T,2 or the input‐variable‐squared‐prediction‐error) based on surrogate measurements alone exceeded 70,th percentile confidence limits, (3) when model error >3mm, and (4) at each radiographic localization. Results: Radial tumor displacement prediction errors (mean +/− standard deviation) for the four schema described above were 2.4 +/− 1.2 mm, 1.9 +/− 0.9 mm, 1.9 +/− 0.9 mm, and 1.7 +/− 0.9 mm, respectively. The no‐update error was significantly larger than errors of the other methods, which did not differ significantly. Mean update counts were 0, 3.3, 9.2, and 18.5, respectively, over 20 minutes. SPC‐timed updates were 36% and 18% as frequent as error‐based and each‐localization updates. Conclusions: Tumor localization accuracy improved significantly with model updates. Despite comparable tumor localization accuracy amongst the update methods, there were significantly fewer SPC‐timed model updates than error‐timed updates. This study proves the feasibility of timing model updates through analysis of external measurements alone, without direct tumor localization. This work was supported in part by grant # CA124766 from the NIH/NCI and by the Achievement Rewards for College Scientists (ARCS) scholarship.

SU‐E‐T‐512: Targeting Mobile Tumors Precisely with an Integrated Robotic Target Tracking and Dynamic Couch‐Based Motion Compensation System for a Conventional Linear Accelerator

Purpose: To introduce a novel, integrated tumor tracking and dynamic robotic couch‐based target motion compensationsystem with a conventional linear accelerator and to evaluate system accuracy. Methods: We have developed an integrated real‐time tumor tracking and motion compensationsystem using the treatment couch. The system comprises an infra‐red (IR) camerasystem capable of tracking optical reflectors placed on or near a target at a rate of 60 Hz, a robotic couch that is controlled in real‐time and the embedded couch controller capable of feedback control. System accuracy was tested using, a 4D phantom, placed on the couch and programmed to simulate 20 time‐varying 3D tumor motion (derived from patient tumor trajectories) with peak‐peak displacements of up to 4 cm. The position of the optical reflector was maintained at the reference position by tracking it with the robotic couch. Camera lag, system latency and tracking errors were evaluated. Dosimetric accuracy was also studied between with and without motion compensation.Results: In the absence of the IR camera (phantom‐couch direct positive feedback control), the system latency is < 0.5 ms. The IR camera introduces a 17 ms delay in the control system. This lag induces an overall system latency of 67–150 ms. In the absences of the camera, the roboticsystem achieved with standard deviation for tracking errors accuracy of 0.01 mm, 0.03 mm and 0.04 mm in ML, AP and SI directions, respectively. The integrated system achieved standard deivation for tracking error accuracy of 0.5 mm in all three in ML, AP and SI directions, respectively. Dosimetric accuracy was improved by 11–36% Conclusions: We have developed a viable and integrated tumor tracking and real‐time robotic couch‐based target motion compensationsystem that may be integrated with a conventional linear accelerator, obviating the need for a specialized radiation delivery device. Supported by grant #CA122403 from the NCI

TU‐E‐BRC‐10: Selecting Training Data to Optimize Accuracy of Respiratory Surrogate‐Based Tumor Displacement Models

Purpose: Accuracy of respiratory surrogate models of tumor displacement is influenced by selection of appropriate training data, the simultaneous tumor localizations and respiratory surrogate measurements used to create the model. The object of this study was to explore ways in which (1) varying timing and (2) selectively discarding samples of training data can improve surrogate model accuracy. Methods: Motion from 125 lung, 10 liver, and 47 pancreas SBRT treatment fractions from 92 patients was analyzed. Each treatment fraction dataset included radiographically measured tumor positions, positions of three surrogate markers affixed to the torso, and tumor positions predicted by a commercial tumortracking system from marker measurements. Partial‐least‐squares regression models of tumor position were trained on samples from 26 Hz position data. Low‐frequency (0.05–0.10 Hz) and high‐frequency (26 Hz) training datasets 5–150 seconds in duration were created. The effect of discarding samples with the highest leverage (as measured by Hotellings T,2 statistic) was evaluated. Tumor localization errors were measured over 20 minutes. Results: The baseline tumor localization error (mean +/− standard deviation) in training data was 1.9 +/− 1.7 mm. Mean errors for high‐frequency training data were 4.7–4.9 mm and did not differ significantly with number (1–2), timing, or duration (5–30 sec) of acquisitions. Mean +/− standard deviation errors for low‐frequency training data varied from 4.4 +/− 2.7 mm (0.10 Hz, 2.5‐min acquisition) to 6.6 +/− 16.1 mm (0.05 Hz, 45‐sec acquisition). Removal of high‐leverage samples reduced tumor localization errors for high‐frequency‐derived models by 4–7% but did not reduce errors for low‐frequency data. Conclusions: Lengthening training data acquisitions improves models derived from low‐frequency but not high‐frequency training data. Conversely, removing high‐leverage training samples benefits high‐frequency data only. Training sample selection involves interplay between sampling rate, acquisition duration, and number of training samples. This work was supported in part by grant # CA124766 from the NIH/NCI and by the Achievement Rewards for College Scientists (ARCS) scholarship.

SU‐GG‐T‐06: Determining the Optimal Gating Window Size by Considering the Effect of Tumor Displacement on Dose Distributions

Purpose: To develop a population‐based multivariate linear regressionmodels for determining the optimal size of the gating window by considering the effect of the distribution of tumor displacements on target dose. This population model can then be used to prospectively guide the selection of gating window size for an individual case, based on tumor displacement distribution. Method and Materials: We considered stereotactic body radiation therapy(SBRT)treatment plans for lungtumors planned and calculated using heterogeneity corrections. Fifty tumor trajectories that varied in the distribution of tumor displacement were selected. A set of variables were defined to describe the variability in these trajectories. Four‐dimensional dose calculation was performed by sampling the distribution of tumor displacements resulting in the residual tumor displacement influenced target dose. Stepwise elimination was performed to determine only those input variables that were independently predictive of D95 and D99. A multivariate regression population‐based model was constructed for D95 and D99 as a function of these independently predictive variables (and as a result the gating window size). This model was then tested on a set of 15 new tumor trajectories and gating window prediction errors were determined. Results: Ten different models were generated for D95 (5 locations, 2 margins) and ten for D99. Average errors were calculated for all the models in terms of percentage of cases more than 5% off the prediction. For D95 the average was 3.8 with and standard deviation of 3.0, whereas for D99 it was 4.5 with a standard deviation of 3.4. Conclusion: The MLR models built show a great degree of predictability in D95 and D99 for all the 5 tumor locations and 15 breathing trajectories studied in prospective prediction. The methodology developed can be used to predict the gating window for any patient using only the tumor trajectory as input.

SU‐GG‐T‐08: A Novel Treatment Couch for Real‐Time Tracking of Respiration Induced Target Motion: Evaluating Its Geometric Accuracy

Purpose: To characterize the performance of a novel treatment couch designed and developed for respiration‐induced real‐time tumor motion tracking and to investigate its behavior with real tumor trajectories. Method and Materials: A new treatment couch has been developed to track targets in real‐time. The system is capable of tracking targets with less than 0.5 mm accuracy. The max physical velocity for the couch is 13 cm/sec along SI and ML directions and 10 cm/sec along AP direction. It can reach a maximum acceleration of 100 cm/sec2 along all the three axes. To quantify the performance of the couch we used 25 tumor trajectories showing a peak‐to‐peak displacement in the range of 20 – 39 mm and sampled at 38 ms. These trajectories were fed to 3D time‐varying trajectories derived from patient data. The motion of the phantom measured in real‐time via a camera system was feed into the couch control system. Tracking errors were analyzed.Results: We found that the mean error in tracking for 20 tumor trajectories was 0.37 mm with a standard deviation of 0.29 mm. The system dead time was approximately 40ms. For individual trajectories, the median error varied from 0.15 mm to 0.70 mm whereas the individual mean values varied from 0.17 mm to 0.74 mm. Couch speed and acceleration were acceptable for real‐time target tracking.Conclusion: We have developed a new couch for use with a linear accelerator that can track and correct for intrafraction (including respiration‐induced) target motion. Mean and median tracking errors are less than 0.5 mm even for peak‐to‐peak displacements up to 4 cm. The couch shows a fairly stable and vibration free behavior over a wide range of tumor displacements studied. Conflict of Interest: NIH grant CA124766

MO‐EE‐A3‐02: Inferring Nodal Volume and Primary Tumor Positions From Multiple Anatomical Surrogates Using 4D CT in Stage III Lung Cancer

Purpose: To investigate the feasibility of modeling primary tumor and nodal volume positions from anatomical surrogates in order to reduce the contouring burden of planning from 4D CTs in Stage III lungcancer.Method and Materials: To localize their centroid positions at each respiratory phase, we contoured nodal volumes and primary tumors in 16 Stage III lungcancer planning 10‐equal‐phase 4D CTs. We also localized a series of anatomical respiratory surrogates (carina, xyphoid, nipples, midsternal external marker) in each image. To explore the feasibility of our proposed method, we 1) characterized the correlations between target and surrogate 3D motion, 2) applied Ordinary Least‐Squares (OLS) and Partial Least‐Squares (PLS) to a random subset (3–8) of images to predict the target positions in the remaining images, 3) determined the best set of three respiratory phase bins to contour, and 4) used them to create 3‐phase models using either all anatomical surrogates or carina alone. Results: The surrogate best‐correlated to target motion was most often the carina but varied widely. Depending on the number of phases used to build the models, mean errors ranged from 1.0mm to 1.4mm and from 0.8mm to 1.0mm for OLS and PLS, respectively. When the 0%, 40%, and 70% respiration phases were used, 3‐phase models had mean(±standard deviation) errors of 0.8±0.5mm and 2.4±9.0mm for models based on all surrogates and carina alone, respectively. For target coordinates with motion>5mm, the mean 3‐phase error was 1.3mm. Conclusion: Using only three contoured respiratory phases to train the models, the mean model error was on the order of CT resolution. Inferential modeling of the primary tumor and nodal volume positions may have the potential to decrease the time required to process 4D CT scans, thereby improving therapy by allowing for incorporation of patient‐specific margins in the planning process.