Can Magnetic Resonance Imaging (MRI) Only Replace MRI–Computed Tomography Planning With a Titanium Applicator for Cervical Brachytherapy?

Volume 96  Number 2S  Supplement 2016

ePoster Sessions S225

volume. Even though the outcome difference is very small, it allowed preliminary optimization of the form of RF. Future work should validate this finding in larger cohorts and derive the required LN dose level. We conclude that it is feasible to derive RF from 3D dose data and outcome only. This method will be useful to retrospectively study existing and new RF in very large patient cohorts. Author Disclosure: M.B. van Herk: None. A. McWilliam: None. B. Sanderson: None. J. Kennedy: None. L. Kershaw: None. C.M. West: None. A. Choudhury: None.

patients’ healthy tissue exposure to radiation dose. The work shows also that the treatment response for prostate is similar either with MRI + CT or MRI-only RTP workflow without any differences in early toxicity. The clinic will continue conducting MRI-only RTP. Author Disclosure: J. Korhonen: None. H. Visapa¨a¨: None. T. Seppa¨la¨: None. M. Kapanen: Chief Physicist; Tampere Hospital. K. Saarilahti: Chief Doctor; Helsinki University Central Hospital. M. Tenhunen: Chief Physicist; Helsinki University Central Hospital.

1145

1144 Clinical Experiences of Treating Prostate Cancer Patients With Magnetic Resonance ImagingeOnly Based Radiation Therapy Treatment Planning Workflow J. Korhonen,1 H. Visapa¨a¨,2 T. Seppa¨la¨,2 M. Kapanen,2 K. Saarilahti,2 and M. Tenhunen2; 1Helsinki University Central Hospital, Helsinki FIN00029, Finland, 2Helsinki University Central Hospital, Helsinki, Finland Purpose/Objective(s): Recent studies have demonstrated a possibility to omit CT imaging from the external photon radiotherapy treatment planning (RTP) workflow by developing methods for the entire process by relying on MRI only. The current study evaluated the success and robustness of the MRI-only workflow for prostate cancer patients in a clinic. The work determined the proportion of patients successfully undergone the MRIonly RTP workflow, and in case when also CT was needed, addressed the reasons for this additional patient imaging. Furthermore, the study aimed to calculate the potential cost effects and target position differences with the CT-avoiding MRI-only workflow compared to the standard process with both MRI and CT. Moreover, the study evaluated treatment responses. Materials/Methods: The MRI-only RTP was integrated into a routine clinical workflow for prostate cancer patients. Essential demands to commence MRI-only workflow were met by enabling patient positioning in MRI by following RT requirements (e.g., flat table and coil frames), applying MRI with high geometric accuracy (<2 mm within 50 cm FOV), developing a technique to transform an MR image into a CT-like image (substitute CT generation by auto-segmentation of bone followed with algorithms converting MR intensity values into HUs), and quantifying accuracy of the workflow (dosimetry and IGRT: 1% and 1 mm compared to CT). The current study included the first 150 prostate cancer patients treated in our clinic with the MRI-only RTP workflow during 2012-2015. The experiences with the MRI-only workflow compared to the conventionally applied MRI + CT process were examined for such measures as hospital resources (V), target contour accuracy (avoiding MRI to CT registration error), and treatment response (PSA). Results: For 8% of patients who were undergoing MRI-only RTP workflow also a CT was applied. The reasons included obesity, hip prosthesis, and gold marker identification uncertainties. The substitute CT generation and verification process employed a physicist or a radiographer for roughly 30 minutes, instead of applying staff for CT (2 radiographers, <30 min) as in standard MRI + CT process. Cost savings for the clinic achieved by the MRI-only workflow were w150V/patient. The MRI-only workflow avoided the potential systematic target position error of 1-4 mm caused by the MRI to CT registration uncertainty. The PSA values were reduced (pre- vs post-treatment, antiandrogen therapy included in addition to RT) 83% in the MRI-only and 77% in the MRI + CT patient groups. Conclusion: This study indicates that for over 90% of prostate cancer patients the RTP workflow can be conducted with MRI only, thus reducing treatment costs, increasing reliability of target positioning, and minimizing

Can Magnetic Resonance Imaging (MRI) Only Replace MRIeComputed Tomography Planning With a Titanium Applicator for Cervical Brachytherapy? B. Chinsky,1 A.M. Diak,1 W. Small, Jr,2 J.C. Roeske,2 and M.M. Harkenrider2; 1Loyola University Medical Center, Maywood, IL, 2 Stritch School of Medicine, Loyola University Chicago, Maywood, IL Purpose/Objective(s): Volume-based treatment planning on computed tomography (CT) or magnetic resonance (MR) images, prescribing to the high risk clinical target volume (HR CTV), is the current standard for cervical cancer brachytherapy (BT). MR-based treatment planning affords superior soft tissue contrast compared to CT. With titanium applicators though, the MR is known to have spatial distortions which may make accurate delineation of the applicators inferior to CT. However, CT adds additional time, expense, and radiation exposure to the patient. The goal of this study was to evaluate the dosimetric differences between MR only vs. MR-CT planning with a titanium tandem and ovoid (T&O) applicator to determine if the CT can be eliminated from the treatment workflow. Materials/Methods: Ten patients were identified who received cervical BT with a titanium Fletcher-Suit-Delclos T&O applicator. Patients underwent both an MR and CT with the applicator in place. All patients had their HR CTV and organs at risk (OARs) contoured on the axial T2-weighted oblique MR, which were then transferred to the co-registered planning CT, where the applicator was then identified. Retrospectively, 3 planners independently delineated the applicator on the axial T2 3D MR while blinded to the CT; the identical dwell position times used in the delivered plan were then loaded. The following cumulative dose-volume histogram (DVH) parameters were compared to those of the previously delivered plan: HR CTV D90 and D98, bladder D0.1 cc and D2 cc, rectum D0.1 cc and D2 cc, sigmoid D0.1 cc and D2 cc. The mean values among planners of the DVH parameters for each patient from the MR only plans were calculated and compared to the delivered MR-CT plan DVH parameters using a paired t-test. Results: The results demonstrated there were no significant differences among DVH parameters with the exception of the rectum D0.1 cc. D0.1 cc of the rectum was significantly lower with MR only planning with a 7.1% decrease from 5.3 to 5.7 Gy (P Z 0.03). All dosimetric parameters were +/- 0.1-0.4 Gy on MR only relative to MR-CT. P values, the percent difference in mean values between the MR only and MR-CT plans, and mean OAR doses are listed in the Table below. Conclusion: Although there were dosimetric differences observed between the MR only and MR-CT plans, all differences were deemed relatively small and not clinically significant. Furthermore, elimination of the CT reduces treatment time, cost, and radiation exposure, making MR only planning the favorable option. This study demonstrates that cervical BT with a titanium applicator can be planned successfully on an MR alone which is now standard in our clinic. Author Disclosure: B. Chinsky: None. A.M. Diak: None. W. Small: xx; Northwestern University. Speaker’s Bureau; Zeiss. xx; ACR, Gynecologic

Abstract 1145; Table 1 HR CTV

Mean Dose

MR-CT (Gy) MR (Gy) % Difference Mean Doses P value

Bladder

Rectum

Sigmoid

D90

D98

D0.1cc

D2cc

D0.1cc

D2cc

D0.1cc

D2cc

7.7 7.6 -1.6% 0.35

6.5 6.4 -1.8% 0.23

8.5 8.8 3.7% 0.29

6.0 6.2 2.7% 0.33

5.7 5.3 -7.1% 0.03

4.2 4.0 -5.8% 0.18

5.4 5.2 -4.5% 0.23

4.0 3.9 -3.0% 0.18

S226

International Journal of Radiation Oncology  Biology  Physics

Cancer Intergroup. J.C. Roeske: Research Grant; Varian Medical Systems. M.M. Harkenrider: Research Grant; NIH.

Materials/Methods: Fifteen H&N cases with CT, MR, and PET images were collected and saved in a library of plans. Experienced physicians manually annotated OARs including spinal cord, brainstem, optic nerves, chiasm, eyes, mandible, tongue, parotid glands. We also have ten super-resolution MR images of the tongue area, where the genioglossus and inferior longitudinalis tongue muscles are defined as organs of interest. We applied the concepts of random forest- and deep learning-based object classification for automated image annotation in order to investigate feasibility of using machine learning for head and neck radiotherapy planning process. In this new paradigm of segmentation, random forests were used to automate the landmark detection from the target super-resolution MR images. The detected landmarks then guided landmark-based atlasing of manual segmentations from training images to the target image. The landmarks are also used to automatically measure the morphological properties of the depicted organs. Alternatively to auto-segmentation with random forest-based landmark detection, we applied deep convolutional neural networks for voxel-wise segmentation of OARs in single and multi-modal images. Results: We presented a comprehensive study on using machine learning concepts for auto-segmentation of OARs and tongue muscles for the H&N radiotherapy planning. An accuracy of 1.87 mm was achieved in the random forest-based landmarks detection, which, taking into account imperfection of MR images, can be considered to be close-to-human landmark detection performance. Even for the most challenging structures such as the genioglossus and inferior longitudinalis tongue muscles, segmentation accuracy measured in terms of Dice coefficient was of 81.8%, which is close to interrater variability. Overall, the use of deep-learning afforded an unprecedented opportunity to improve the accuracy and robustness of OAR segmentation. Conclusion: A novel machine learning framework has been developed for image annotation and structure segmentation. Our results indicate the great potential of deep learning in radiotherapy treatment planning. Author Disclosure: B. Ibragimov: None. F. Pernus: None. P. Strojan: None. L. Xing: ; Stanford University.

1146 Quantitative and Dosimetric Evaluation of Offline Adaptive Radiation Therapy Toward Establishing a Decision Support Framework for Evaluating the Necessity for Real-Time Adaptation M. Reilly, J. Kavanaugh, O.L. Green, and S. Mutic; Washington University School of Medicine, St. Louis, MO Purpose/Objective(s): To establish decision support metrics via quantitative assessment of dose volume histograms utilizing a retrospective analysis of patients who had secondary treatment plans in our clinic. Specifically, to establish the fraction number, trigger event, and length of time in days at which patients were re-scanned / re-planned to determine the dosimetric impact on a patient’s overall composite treatment. Ultimately, this work will permit evolution of adaptive radiotherapy from qualitative / subjective review of a patient’s chart and daily imaging toward evidence based / quantitative decision metric system. Materials/Methods: Retrospective cohort of 52 lung, 35 breast, and 105 head and neck patients were included in an IRB approved study. The initial cohort was down-selected to 5 patients for each of the 3 treatment sites and utilized rescan CT for performing dosimetric outcomes of the composite plans versus if no action or intervention had been taken. Patient composite treatment plans were used for DVH comparison and the patients electronic records used to track progress through the offline adaptive radiotherapy workflow. Results: Trigger events for determining a patient’s candidacy for an ontreatment CT and re-plan included: tumor volume loss via daily CBCT, patient weight loss and/or gain >10%, localized edema, and daily setup and immobilization difficulties. The average length of time between a decision to re-scan a patient and implementation of their new treatment plan was 3.4  1.2 days. The decision to perform adaptive radiotherapy ranged the central 80% of the radiotherapy course, e.g., no patients were adapted earlier or later than 10% of their total number of fractions. Conclusion: The analyses indicate that adapting a patient’s treatment plan with >50% of the total number of fractions remaining has a marginal quantitative influence on the patients composite treatment should an adaptive plan not have been performed, e.g., <1 Gy mean dose sparing for OAR (dependent on proximity to the target). However, for those patients who were adapted earlier in their radiotherapy course an approximate 2-8% improvement in the new target volume covered by prescription dose was observed. This indicates that a new paradigm in monitoring a patient’s radiotherapy fractionation course should be considered e moving from a strictly interval or weekly review of a patient’s treatment to one that considers a non-linear observation with considerable more emphasis on a front-loaded review. Author Disclosure: M. Reilly: None. J. Kavanaugh: None. O.L. Green: None. S. Mutic: Honoraria & Travel; ViewRay, Inc., Varian Medical Systems, Siemens Healthcare. Stock; Radialogica, LLC. Partnership; TreatSafely, LLC. Responsible for the technical direction of the company; Radialogica, LLC. Responsible for general management of the organization and all aspects of the business; TreatSafely, LLC.

1147 Development of a Novel Deep Learning Algorithm for Autosegmentation of Clinical Tumor Volume and Organs at Risk in Head and Neck Radiation Therapy Planning B. Ibragimov,1 F. Pernus,2 P. Strojan,3 and L. Xing4; 1Stanford University, Stanford, CA, 2University of Ljubljana, Ljubljana, Slovenia, 3Institute of Oncology, Ljubljana, Slovenia, 4Department of Radiation Oncology, Stanford University, Stanford, CA Purpose/Objective(s): Accurate and efficient delineation of tumor target and organs-at-risks is essential for the success of radiotherapy. Despite the decades of intense research efforts, auto-segmentation has not yet become clinical practice. In this study, we present a random forest- and deep learningbased algorithm for object classification with the aim to automatically segment organs-at-risk (OAR) for head and neck (H&N) treatment planning.

1148 Computed Tomography Ventilation Image Guided Adaptive Functional Avoidance in Radiation Therapy for Locally Advanced Lung Cancer T. Yamamoto,1 S. Kabus,2 M. Bal,3 P. Keall,4 S.H. Benedict,1 C. Wright,1 and M.E. Daly1; 1University of California Davis Comprehensive Cancer Center, Sacramento, CA, 2Philips Research, Hamburg, Germany, 3Philips Healthcare, Best, Netherlands, 4University of Sydney, Sydney, Australia Purpose/Objective(s): An emerging technology, CT ventilation imaging, which enables lung functional avoidance radiotherapy (RT), has recently been translated into the clinic to reduce pulmonary toxicity. Tumor regression during a course of treatment is common in lung cancer RT, and often leads to recovery of lung function. Adaptive RT (ART) strategies that account for such changes have not been explored to date. We hypothesized that CT ventilation image-guided adaptive functional avoidance significantly reduces the dose to the functional lung compared to conventional planning schemes. Materials/Methods: Repeat CT scans were acquired before RT and during a course of RT after w20 Gy and w34 Gy for five patients with locally advanced lung cancer enrolled on an ongoing prospective clinical trial. Ventilation images were calculated by deformable image registration of 4D CT image data sets and quantitative image analysis. Spatial heterogeneity of ventilation was assessed based on the difference between the ipsilateral to contralateral ventilation ratio and unity. Four IMRT plans were created for each patient: (1) functional ART, (2) anatomic ART, (3) functional nonART, and (4) anatomic non-ART. Functional plans were designed to selectively avoid functional lung regions and meet dose-function constraints as well as standard constraints to other normal tissues, while anatomic plans were designed to meet standard constraints. Plan adaptation was performed after both 20 Gy and 34 Gy for ART plans. The following metrics were quantified for each plan: (1) the accumulated function-weighted mean lung dose (fMLD), and (2) mean accumulated dose to (0-25), (25-50), (50-75), and (75-100)% percentile ventilation regions. A paired t-test was used to compare those metrics of four plans to test the hypothesis.