Auto Contouring Research Articles

Training and validation of a commercial deep learning contouring platform

We present our experience with training and validation of a commercially available deep learning algorithm for organs at risk(OAR) auto-contouring. Computed tomography(CTs) with OARs from a cohort of 213 head and neck(H&N) patients were used for training the deep learning model. A separate cohort of 85 CTs and structure sets was used for validation. All OARs (13) were contoured by a single physician. Metrics such as the DICE similarity coefficient (DSC), Jaccard similarity coefficient (JSC), and volumetric difference (VD) were used to analyze contouring variation. Mean DSC and JSC values ranged 0.48-0.89 and 0.32-0.8, respectively, depending on OAR. A DSC value ≥0.7 indicated low inter-observer variability. In our study, all but one of the contours were above this threshold. DSC for the middle pharyngeal constrictor had the lowest value of all the contours. This may be due to the small volume of this structure. Qualitative assessment of auto-segmented structure samples confirmed the reliability of DSC by demonstrating the compatibility between the expert’s evaluation and DSC values. Overall, we found that deep learning auto contouring is a useful tool to speed up the process of contouring in radiotherapy treatment planning.

Geometric and Dosimetric Evaluation of a Commercially Available Auto-segmentation Tool for Gross Tumour Volume Delineation in Locally Advanced Non-small Cell Lung Cancer: a Feasibility Study

AimsTo quantify the reliability of a commercially available auto-segmentation tool in locally advanced non-small cell lung cancer using serial four-dimensional computed tomography (4DCT) scans during conventionally fractionated radiotherapy. Materials and methodsEight patients with serial 4DCT scans (n = 44) acquired over the course of radiotherapy were assessed. Each 4DCT had a physician-defined primary tumour manual contour (MC). An auto-contour (AC) and a user-adjusted auto-contour (UA-AC) were created for each scan. Geometric agreement of the AC and the UA-AC to the MC was assessed using the dice similarity coefficient (DSC), the centre of mass (COM) shift from the MC and the structure volume difference from the MC. Bland Altman analysis was carried out to assess agreement between contouring methods. Dosimetric reliability was assessed by comparison of planning target volume dose coverage on the MC and UA-AC. The time trend analysis of the geometric accuracy measures from the initial planning scan through to the final scan for each patient was evaluated using a Wilcoxon signed ranks test to assess the reliability of the UA-AC over the duration of radiotherapy. ResultsUser adjustment significantly improved all geometric comparison metrics over the AC alone. Improved agreement was observed in smaller tumours not abutting normal soft tissue and median values for geometric comparisons to the MC for DSC, tumour volume difference and COM offset were 0.80 (range 0.49–0.89), 0.8 cm3 (range 0.0–5.9 cm3) and 0.16 cm (range 0.09–0.69 cm), respectively. There were no significant differences in dose metrics measured from the MC and the UA-AC after Bonferroni correction. Variation in geometric agreement between the MC and the UA-AC were observed over the course of radiotherapy with both DSC (P = 0.035) and COM shift from the MC (ns) worsening. The median tumour volume difference from the MC improved at the later time point. ConclusionsThese findings suggest that the UA-AC can produce geometrically and dosimetrically acceptable contours for appropriately selected patients with non-small cell lung cancer. Larger studies are required to confirm the findings.

Every Breath You Take – Evaluating Respiratory Motion of the Bony Thorax in the Context of Bone SBRT

Stereotactic Body Radiation Therapy (SBRT) delivers highly conformal treatment in a single or few fractions of high-dose radiation yielding excellent local control rates with minimal toxicity. There is growing interest in using SBRT to treat oligometastases, which potentially includes any anatomical site where metastases can occur. Bone is one of the most common sites of metastatic disease and although methodology to treat SBRT spine is well established, very little is known about the technical considerations of bone SBRT, particularly regarding respiratory motion in the bony thorax. The purpose of this study was to evaluate respiratory motion in thoracic bone metastases and to determine if the motion differs based on anatomic location. The first 70 patients treated with thoracic bone SBRT planned with four-dimensional computed tomography (4DCT) were identified from a prospective institutional database. To quantify motion, a region of interest representative of the treatment target was contoured by a single user in Pinnacle on the 0% (inhale) and 50% (exhale) datasets using fixed autocontour threshold values to eliminate subjectivity. A point of interest was automatically placed in the centroid of each volume and the relative positional difference of these points was calculated to measure the lateral (LR), anterior/posterior (AP) and superior/inferior (SI) motion. A total linear distance (LD) vector was calculated from these values. Rib lesions were further categorized as anterior, lateral or posterior based on axial position as well as superior (1-4), middle (5-7) and inferior (8-12) depending on rib location. Rib motion was analyzed using analysis of variance (ANOVA), if it was significant (p<0.05) a post hoc analysis was done to evaluate pairwise comparisons amongst the rib location groups. There were 15/16 sternal lesions assessed as 1 patient had significant bone destruction and could not be reliably contoured. Mean (range) LR, AP and SI sternum motion was 0.2 mm (0-0.7), 1.2 mm (0-2.6) and 0.1 mm (0-0.3) respectively and the LD was 1.2 mm (0-2.6). Forty-seven rib lesions were assessed for motion; 7 were excluded due to bone destruction preventing use of auto contour tool. The mean (range) LR, AP and SI motion for all rib lesions was 1.1 mm (0-4.0), 2.4 mm (0-8.4) and 0.1 mm (0-0.8) respectively and LD was 2.8 mm (0.1-8.5). When categorized by rib location, there were 9 superiorly, 22 middle and 16 inferior located rib lesions. The mean LD was 3.2 mm, 3.5 mm and 1.6 mm for superior, middle and inferior lesions respectively. There was a significant difference between the groups, with the inferiorly located moving less in terms of AP motion (p=0.02) and LD (p=0.04). In terms of axial position, there were 6 anterior lesions, 27 lateral lesions and 14 posterior lesions. The mean LD was 3.6 mm, 3.3 mm and 1.3 mm for anterior, lateral and posterior lesions respectively. There was a significantly less motion for posteriorly located lesions in terms of AP motion (p=0.005) and LD (p=0.02). This data suggests that the respiratory motion of the bony thorax is variable but not trivial in the context of high precision radiation therapy techniques with small margins. It appears that lesions located in the sternum and posterior and/or inferior aspect of the ribs move less; however, contouring guidelines, margins and image guidance technique need to be strongly considered before eliminating motion assessment when planning thoracic bone SBRT.

Clinical Evaluation of Commercial Atlas-Based Auto-Segmentation in the Head and Neck Region.

Background: While atlas segmentation (AS) has proven to be a time-saving and promising method for radiation therapy contouring, optimal methods for its use have not been well-established. Therefore, we investigated the relationship between the size of the atlas patient population and the atlas segmentation auto contouring (AC) performance.Methods: A total of 110 patients' head planning CT images were selected. The mandible and thyroid were selected for this study. The mandibles and thyroids of the patient population were carefully segmented by two skilled clinicians. Of the 110 patients, 100 random patients were registered to 5 different atlas libraries as atlas patients, in groups of 20 to 100, with increments of 20. AS was conducted for each of the remaining 10 patients, either by simultaneous atlas segmentation (SAS) or independent atlas segmentation (IAS). The AS duration of each target patient was recorded. To validate the accuracy of the generated contours, auto contours were compared to manually generated contours (MC) using a volume-overlap-dependent metric, Dice Similarity Coefficient (DSC), and a distance-dependent metric, Hausdorff Distance (HD).Results: In both organs, as the population increased from n = 20 to n = 60, the results showed better convergence. Generally, independent cases produced better performance than simultaneous cases. For the mandible, the best performance was achieved by n = 60 [DSC = 0.92 (0.01) and HD = 6.73 (1.31) mm] and the worst by n = 100 [DSC = 0.90 (0.03) and HD = 10.10 (6.52) mm] atlas libraries. Similar results were achieved with the thyroid; the best performance was achieved by n = 60 [DSC = 0.79 (0.06) and HD = 10.17 (2.89) mm] and the worst by n = 100 [DSC = 0.72 (0.13) and HD = 12.88 (3.94) mm] atlas libraries. Both IAS and SAS showed similar results. Manual contouring of the mandible and thyroid required an average of 1,044 (±170.15) seconds, while AS required an average of 46.4 (±2.8) seconds.Conclusions: The performance of AS AC generally increased as the population of the atlas library increased. However, the performance does not drastically vary in the larger atlas libraries in contrast to the logic that bigger atlas library should lead to better results. In fact, the results do not vary significantly toward the larger atlas library. It is necessary for the institutions to independently research the optimal number of subjects.

Knowledge-Based Auto Contouring for Radiation Therapy: Challenges in Standardizing Object Definitions, Ground Truth Delineations, Object Quality, and Image Quality

Manual contouring in RT planning of organs at risk (OARs) has been prone to significant variability due to a lack of standardized object definitions, leading to imprecision. Although clinical guidelines have been made available, ambiguities in object definitions still exist. Suboptimal object quality and image quality can also significantly impact the quality of object contouring. These issues are particularly problematic when creating robust object models for purposes of auto-contouring. As such, we present more precise definitions of selected OARs in the neck and thorax as an extension of existing guidelines. A prevailing issue is evaluation of auto-contouring methods as a function of input image quality. We propose a new approach to assess the impact of object and image quality upon auto-contouring results. We hypothesize that OAR manual contours still have significant variability from existing guidelines, and object and image quality impact auto-contouring. We first developed precise and computable definitions of the neck and thorax body regions based on anatomical considerations. We then developed precise standardized definitions of a set of key OARs in these body regions by extending object definitions available from recent guidelines. Next, we retrospectively created a database of CT images and manually drawn contours obtained from RT planning in 216 head and neck cancer patients and 200 thoracic cancer patients. A reader examined each image and assigned a quality grade to each 3D object in the image based on a set of pre-specified criteria which included: deviation in body posture and object intensity; presence of image noise, streak artifacts, object shape distortion, and pathology; and lack of object contrast. Using logical predicates, we designed an algorithm to map these quality grades to a numeric quality score for each object in each CT scan, as well as for the scan. Precise object definitions for 11 objects in the neck and 11 objects in the thorax were created. Per our standardized definitions, among all objects in the neck studies, 33.8% were acceptable without modification, 46.9% required minor changes, and 19.3% needed major editing. For the thorax, these rates were 10.5%, 56.17%, and 33.33%, respectively. ∼1% of the scans, considering all objects in each scan, were of high quality as per the above criteria. Results of analysis of an auto-contouring software as a function of the image quality numeric score will be presented at the conference. Precise standardized object definitions are essential to ensure high quality of OAR contours for RT planning. Yet, currently available guidelines are still ambiguous as borne out from our analysis. Suboptimal image quality may also significantly impact the quality of object contouring. We have arrived at more precise definitions of a set of objects in the neck and thorax, and developed a new approach to assess the impact of object and image quality upon auto-contouring results.

E-Rekha: A High-Performance Software System for Auto Contouring Head and Neck Anatomy in Adaptive Radiation Therapy

e-Rekha: A High-Performance Software System for Auto Contouring Head and Neck Anatomy in Adaptive Radiation Therapy

An MRI-compatible patient rotation system - design, construction, and first organ deformation results.

Conventionally in radiotherapy, a very heavy beam forming apparatus (gantry) is rotated around a patient. From a mechanical perspective, a more elegant approach is to rotate the patient within a stationary beam. Key obstacles to this approach are patient tolerance and anatomical deformation. Very little information on either aspect is available in the literature. The purpose of this work was therefore to design and test an MRI-compatible patient rotation system such that the feasibility of a patient rotation workflow could be tested. A patient rotation system (PRS) was designed to fit inside the bore of a 3T MRI scanner (Skyra, Siemens) such that 3D images could be acquired at different rotation angles. Once constructed, a pelvic imaging study was carried out on a healthy volunteer. T2-weighted MRI images were taken every 45° between 0° and 360°, (with 0° equivalent to supine). The prostate, bladder, and rectum were segmented using atlas-based auto contouring. The images from each angle were registered back to the 0° image in three steps: (a) Rigid registration was based on MRI visible markers on the couch. (b) Rigid registration based on the prostate contour (equivalent to a rigid shift to the prostate). (c) Nonrigid registration. The Dice similarity coefficient (DSC) and mean average surface distance (MASD) were calculated for each organ at each step. The PRS metall design constraints and was successfully integrated with the MRI scanner. Phantom images showed minimal difference in signal or noise with or without the PRS in the MRI scanner. For the MRI images, the DSC (mean±standard deviation) over all angles in the prostate, rectum, and bladder was 0.60±0.11, 0.56±0.15, and 0.76±0.06 after rigid couch registration, 0.88±0.03, 0.81±0.08, and 0.86±0.03 after rigid prostate guided registration, and 0.85±0.03, 0.88±0.02, 0.87±0.02 after nonrigid registration. An MRI-compatible patient rotation system has been designed, constructed, and tested. A pelvic study was carried out on a healthy volunteer. Rigid registration based on the prostate contour yielded DSC overlap statistics in the prostate superior to interobserver contouring variability reported in the literature.

Auto Contour Initialization of Breast Masses in Contrast Enhanced Breast CT

Dedicated breast CT (bCT) produces high-resolution 3D tomographic images of the breast, fully resolving fibroglandular tissue structures within the breast and allowing for breast lesion detection and assessment in 3D.Various techniques have been used for detecting cancer in women. Previous studies have worked on providing a manual seed point for evaluating the area of tumor. In this study, we present a method for auto initialization of seed point enhancing the quality of detection. First, image enhancement techniques are used which is then followed by detection of circular areas in image and choosing the best out of them. In next step, centroid of the finest circle is used as the seed point. Further, 3D radial-gradient index segmentation is used to obtain a crude initial contour, which is then refined by a 3D level set-based active contour algorithm. The algorithm is run for a number of iterations to get the enhanced results.

O20. Full automation of radiation therapy treatment planning

Introduction We are using the Eclipse Application Programming Interface (API) and Mobius 3D QA system, together with in-house code, to develop an automated treatment planning system for cervix, breast (and chest wall), and head/neck cancers. The goal is to develop a system that includes multiple layers of internal QA that can be used by junior staff. Materials and methods Currently, the following preparatory steps have been fully automated: isocenter detection, couch removal, and body segmentation. For head and neck treatments, the normal tissue segmentation has been automated using an in-house algorithm, and the results scored by a radiation oncologist on 118 patients. For 4-field box cervical cancer treatments, two independent approaches have been developed to create the field apertures (jaw and MLC positions): (1) Atlas-based segmentation of bony anatomy on the patient’s CT, then calculation of apertures based on the projection of these bones to each beam’s-eye-view; (2) Deformably register atlas DRRs to the patient’s DRR for each beam, then use the deformation matrix to deform atlas blocks (MLC positions) to the patient’s DRR. These methods were tested on 39 patients, and the results scored by a radiation oncologist. Finished treatment plans (all sites) are automatically sent to Mobius to verify the dose calculation, and all parameters (MU, etc.) compared with population ranges. Results Head/neck auto contouring takes ∼15 minutes. For cord, brainstem, brain, parotid and mandible, auto-contours were scored acceptable, except for situations when the head position was non-standard. Esophagus and cochlea were generally close, but will require margins before use as a planning avoidance structures. For cervical cancer 4-field box treatments, method 1 was scored as acceptable for 96% of beams, with all failures caused by a slight error in the position of the superior border (10–15 mins per patient). Method 2 was scored as acceptable for 79% of beams, so improvements are needed here before it can be used to verify method 1. Completion of a complete cervical cancer plan, including dose calculation in Eclipse and secondary dose calculation in Mobius takes Conclusions We report on preliminary results of an automated treatment planning approach. Normal tissue segmentation for head/neck patients is very successful and has been introduced into use at our clinic. Determination of the jaw/MLC positions for 4-field box treatments of cervical cancer is also successful and has been introduced into use in our clinic.

SU‐F‐J‐72: A Clinical Usable Integrated Contouring Quality Evaluation Software for Radiotherapy

Purpose:To introduce the Auto Contour Evaluation (ACE) software, which is the clinical usable, user friendly, efficient and all‐in‐one toolbox for automatically identify common contouring errors in radiotherapy treatment planning using supervised machine learning techniques.Methods:ACE is developed with C# using Microsoft .Net framework and Windows Presentation Foundation (WPF) for elegant GUI design and smooth GUI transition animations through the integration of graphics engines and high dots per inch (DPI) settings on modern high resolution monitors. The industrial standard software design pattern, Model‐View‐ViewModel (MVVM) pattern, is chosen to be the major architecture of ACE for neat coding structure, deep modularization, easy maintainability and seamless communication with other clinical software. ACE consists of 1) a patient data importing module integrated with clinical patient database server, 2) a 2D DICOM image and RT structure simultaneously displaying module, 3) a 3D RT structure visualization module using Visualization Toolkit or VTK library and 4) a contour evaluation module using supervised pattern recognition algorithms to detect contouring errors and display detection results. ACE relies on supervised learning algorithms to handle all image processing and data processing jobs. Implementations of related algorithms are powered by Accord.Net scientific computing library for better efficiency and effectiveness.Results:ACE can take patient's CT images and RT structures from commercial treatment planning software via direct user input or from patients’ database. All functionalities including 2D and 3D image visualization and RT contours error detection have been demonstrated with real clinical patient cases.Conclusion:ACE implements supervised learning algorithms and combines image processing and graphical visualization modules for RT contours verification. ACE has great potential for automated radiotherapy contouring quality verification. Structured with MVVM pattern, it is highly maintainable and extensible, and support smooth connections with other clinical software tools.

Performance and Effect of Automatic Contouring in Automatic Adaptive Planning for Prostate Cancer Volumetric Modulated Radiotherapy

Volumetric modulated radiotherapy (VMAT) has been widely used in treating prostate cancer. However, when large daily inter-fractional anatomic changes occur, a simple iso-center shift may not be adequate for target coverage and normal tissue sparing. The fully adaptive radiotherapy involves re-planning process based on daily repeat computed tomography images (CTs), which is impractical with the manual re-contouring and trial-and-error re-planning process. In this paper, we quantitatively evaluated the performance of automatic contouring in an automatic adaptive planning method on cone beam CTs, and demonstrated the feasibility and effectiveness of using the auto contours for online automatic adaptive planning to compensate for the target dose deficit and the critical organ overdose caused by inter-fractional anatomical changes for prostate cancer. Keywords: Adaptive radiotherapy, automatic contouring, automatic re-planning, deformable image registration, inter-fractional motion, prostate cancer, volumetric modulated radiotherapy.

EP-1150 Accurate 3D mandibular VMAT dose prediction from 18-FDG pet derived auto contours: streamlined ORN prophylaxis

EP-1150 Accurate 3D mandibular VMAT dose prediction from 18-FDG pet derived auto contours: streamlined ORN prophylaxis

Evaluation of an Atlas-Based Segmentation Method for High-Risk Prostate Cancer With RTOG-Defined Pelvic Lymph Node Levels

Intensity modulated radiation therapy (IMRT) for high risk prostate cancer requires careful contouring and segmentation of at risk pelvic lymph nodes. However, this process can be very time consuming and difficult compared to traditional prostate only fields. One possible solution to improve target volume contouring efficiency is to use atlas based segmentation software. For this pilot study, we compiled a prostate target atlas database and sought to develop an atlas based segmentation method. Institutional review board approval was obtained to retrospectively analyze patient CT simulation scans to develop an atlas database. Atlas contours included prostate, seminal vesicles, and RTOG-defined pelvic lymph node levels for treatment volumes. Lymph node levels were contoured using RTOG guidelines by following the internal/external iliac vessels and its associated lymph nodes from the L5-S1 vertebral level to obturator nodes at the pubic symphysis. A total of 23 patients with contours reviewed by three different physicians were included in our atlas database. Once this database was compiled we tested the feasibility of developing an atlas based segmentation algorithm and tested it using Dice similarity index (DICE) scores. Atlas segmentation was designed using a leave-one-out analysis (subject being tested was excluded from the atlas). Then the 5 best-matched atlas subjects were automatically chosen and deformed to the test subject. The atlas segmentation method included an additional automatic registration approximation component to influence the intensity-based deformable registration. The 5 sets of auto contours were combined using majority vote where 3 of 5 contours overlapped. Final auto contours were then compared to the manually defined contours using DICE. Using this algorithm for all 23 atlas subjects, the mean DICE score was 0.67 +/- 0.14 for the prostate target volume and 0.74 +/- 0.05 for the pelvic lymph node target volume. Our atlas segmentation method provided accurate contours for high risk prostate cancer. This segmentation method has may have potential for greater accuracy and less variance than traditional inter-individual user contours. In particular, the accuracy of our atlas based segmentation algorithm for pelvic lymph nodes was very good. This is likely due to their association with rigid structures and reduced motion compared to other structures such as the prostate and seminal vesicles. Therefore, our results demonstrate, that atlas based auto-contouring of the pelvic nodes could both help increase consistency for treatment volumes in high risk patients between different cancer centers as well as decrease time related to treatment planning for radiation oncologist.

SU-E-J-220: Evaluation of Atlas-Based Auto-Segmentation (ABAS) in Head-And-Neck Adaptive Radiotherapy

Purpose: Evaluate the accuracy of atlas-based auto segmentation of organs at risk (OARs) on both helical CT (HCT) and cone beam CT (CBCT) images in head and neck (HN) cancer adaptive radiotherapy (ART). Methods: Six HN patients treated in the ART process were included in this study. For each patient, three images were selected: pretreatment planning CT (PreTx-HCT), in treatment CT for replanning (InTx-HCT) and a CBCT acquired in the same day of the InTx-HCT. Three clinical procedures of auto segmentation and deformable registration performed in the ART process were evaluated: a) auto segmentation on PreTx-HCT using multi-subject atlases, b) intra-patient propagation of OARs from PreTx-HCT to InTx-HCT using deformable HCT-to-HCT image registration, and c) intra-patient propagation of OARs from PreTx-HCT to CBCT using deformable CBCT-to-HCT image registration. Seven OARs (brainstem, cord, L/R parotid, L/R submandibular gland and mandible) were manually contoured on PreTx-HCT and InTx-HCT for comparison. In addition, manual contours on InTx-CT were copied on the same day CBCT, and a local region rigid body registration was performed accordingly for each individual OAR. For procedures a) and b), auto contours were compared to manual contours, and for c) auto contours were compared to those rigidly transferred contours onmore » CBCT. Dice similarity coefficients (DSC) and mean surface distances of agreement (MSDA) were calculated for evaluation. Results: For procedure a), the mean DSC/MSDA of most OARs are >80%/±2mm. For intra-patient HCT-to-HCT propagation, the Resultimproved to >85%/±1.5mm. Compared to HCT-to-HCT, the mean DSC for HCT-to-CBCT propagation drops ∼2–3% and MSDA increases ∼0.2mm. This Resultindicates that the inferior imaging quality of CBCT seems only degrade auto propagation performance slightly. Conclusion: Auto segmentation and deformable propagation can generate OAR structures on HCT and CBCT images with clinically acceptable accuracy. Therefore, they can be reliably implemented in the clinical HN ART process.« less

SU‐E‐J‐151: Dosimetric Evaluation of DIR Mapped Contours for Image Guided Adaptive Radiotherapy with 4D Cone‐Beam CT

Purpose:To estimate dosimetric errors resulting from using contours deformably mapped from planning CT to 4D cone beam CT (CBCT) images for image‐guided adaptive radiotherapy of locally advanced non‐small cell lung cancer (NSCLC).Methods:Ten locally advanced non‐small cell lung cancer (NSCLC) patients underwent one planning 4D fan‐beam CT (4DFBCT) and weekly 4DCBCT scans. Multiple physicians delineated the gross tumor volume (GTV) and normal structures in planning CT images and only GTV in CBCT images. Manual contours were mapped from planning CT to CBCTs using small deformation, inverse consistent linear elastic (SICLE) algorithm for two scans in each patient. Two physicians reviewed and rated the DIR‐mapped (auto) and manual GTV contours as clinically acceptable (CA), clinically acceptable after minor modification (CAMM) and unacceptable (CU). Mapped normal structures were visually inspected and corrected if necessary, and used to override tissue density for dose calculation. CTV (6mm expansion of GTV) and PTV (5mm expansion of CTV) were created. VMAT plans were generated using the DIR‐mapped contours to deliver 66 Gy in 33 fractions with 95% and 100% coverage (V66) to PTV and CTV, respectively. Plan evaluation for V66 was based on manual PTV and CTV contours.Results:Mean PTV V66 was 84% (range 75% – 95%) and mean CTV V66 was 97% (range 93% – 100%) for CAMM scored plans (12 plans); and was 90% (range 80% – 95%) and 99% (range 95% – 100%) for CA scored plans (7 plans). The difference in V66 between CAMM and CA was significant for PTV (p = 0.03) and approached significance for CTV (p = 0.07).Conclusion:The quality of DIR‐mapped contours directly impacted the plan quality for 4DCBCT‐based adaptation. Larger safety margins may be needed when planning with auto contours for IGART with 4DCBCT images.Reseach was supported by NIH P01CA116602.

SU-E-P-21: To Investigate CT and KV CBCT Image Deformable Contour Registration Method Using Lung Phantom

Purpose: To analyze CT and KV CBCT image deformable contour registration method and explore calibration CBCT image quality method to achive the goal of Adaptive Radiotherapy(ART) using lung phantom Methods: Acquiring CT, CBCT and modified CBCT (mCBCT) images(calibtated by electron density and image quality) of the anthropomorphic lung phantom. The body contour, simulated targets volume (a sphere with diameter 3.0 centimeters located on left lung, mediastinum and right lung) and simulated organ at risks(OARs) on CT image were deformed to CBCT and mCBCT images with Mimivista software. We achieved the auto deformable conntours and compared with manual contours. We anlyzed the DSC(dice similarity coefficient), PPV(positive predictive value) and SENS(sensitivity) of the body, targets and OARs for three images. Results: Compared to auto contours on CBCT, auto contours on mCBCT had a higher value of 0.7% Dice index and a higher value of 2.51% PPV for body; a higher DSC of 32.50%,a higher PPV of 21.30% and a higher of SENS 30.39% for simulation PTV;a higher DSC of 17.35% and a higher SENS of 32.82% for heart. There were no difference for other parameters of simulation organs. Conclusion: KV CBCT images callibtated by electron density and image quality with the anthropomorphic lung phantom have a good deformable registration contours compared to CBCT auto contours without calibration.

SU-E-J-105: Evaluating Auto-Contouring of Prostate KV Conebeam CTs.

Auto-contouring of daily kV conebeam CTs (kVCBCT) is critical for online adaptive radiotherapy. We surveyed a set of geometric and dosimetric measures to determine which best assess the suitability of kVCBCT contouring algorithms for use in prostate region adaptive planning. Six patients with daily kVCBCT undergoing IMRT of the prostate to 78 Gy were selected. Dose was recalculated with patient density forced to water. Contours were generated on nine kVCBCT for each patient using ABAS (Elekta Ltd.) and also by a physician. The prostate mean dose, D100, D98, where determined and V70 (% and cc) and mean dose for the bladder and rectum for both physician and auto contours. The Dice's Coefficient (DC) was calculated between auto and physician contours, as well as a restricted DC (rDC) which combines geometric and dosimetric information by comparing only the volume within a high dose region. Prostate accuracy can only be discerned with D100, additionally there is no correlation (R2=0.036) between D100 and DC. For all organs, mean dose does not reflect contour suitability. There is large variation in V70 (% and cc) for rectum and bladder, implying V70 is a sensitive indicator of contour suitability. There is however no correlation of V70 with DC. A dose region (>57 Gy) to calculate rDC was chosen to provided optimal correlation (R2>0.81) to V70 (% and cc, rectum and bladder). V70 for rectum and bladder shows the most sensitivity to contour suitability, by concentrating on where accuracy is most vital. Prostate contouring is less critical due to treatment margins, D100 provides the most discerning metric. DC and mean dose are not useable. In lieu of V70, rDC could be used. The rDC can be approximated by limiting to within 2-3 cm of the prostate contour, alleviating the need to calculate dose. A research version of ABAS was provided by Elekta.

WE‐E‐BRC‐01: Impact of Tumor Size and 18F‐FDG Tracer Uptake Heterogeneity in Non‐Small Cell Lung Cancer Tumor Automatic Delineation on PET and CT Images for Gross Tumor Volumes Determination

Purpose: The objective of this study was to investigate the impact of tumor size and tracer uptake heterogeneity on the automatic delineation of 18F‐FDG PET and CT tumor volumes (TV) in NSCLC. Methods: 26 NSCLC patients with FDG PET/CT were considered. For 18 cases, the tumor maximum diameter was measured by macroscopic examination. Two observers performed manual delineation of the tumors on the CT images and of the PET uptakes using a fixed threshold (50% of the max), an adaptive threshold and the Fuzzy Locally Adaptive Bayesian (FLAB). Diameters of the delineated volumes were compared to the histopathology reference when available. Uptake heterogeneity was assessed using the coefficient of variation (std.dev/mean) within TV. Results: High consensus was reached regarding manual delineation of the CT TV (&lt;10% variability). All diameters measured on the delineated images correlated with histopathology (r&gt;0.84, p&lt;0.0001). However, various levels of accuracy were reached depending on the method: manual delineation on CT resulted in large overevaluation (+30±40%), whereas all delineations on PET images resulted in underevaluation (from −16±17% for T50, to −5±10% for FLAB). CT‐based TV were significantly (p&lt;0.05) larger than PET‐based ones. No correlation was found between CT TV and differences between CT‐based and PET‐based TV. On the other hand, a significant correlation (r=0.5, p&lt;0.01) was found between CT TV and the PET uptake heterogeneity, as well as between this heterogeneity and differences between FLAB and threshold‐based delineations (r&gt;0.7, p&lt;0.0002). Conclusions: 18F‐FDG PET has the potential to provide specific tumor targeting for dose boosting/painting in NSCLC. However, threshold‐based techniques that have been suggested for the auto contouring of the PET uptake should not be used for tumors above 2–3cm in size, as they fail to address heterogeneous uptakes that tend to occur in these larger volumes. More advanced automatic delineation tools are required in such cases.

Pathologic Correlation of PET-CT Based Auto Contouring for Radiation Planning in Lung Cancer

Radiation therapy in lung cancer relies on CT and functional imaging (FDG-PET) to delineate tumor volumes. Semi-automatic contouring tools have been developed for PET to improve on the inter-observer bias of manual contouring and intrinsic differences in imaging equipment. A common method involves using a threshold at a given percentage of the max activity, which may be less accurate with smaller tumors and tumors with low source to background ratio. To overcome this deficiency, a gradient algorithm, which detects changes in image counts at the border of the tumor, has been developed.

Pathologic Correlation of PET-CT Based Auto Contouring for Radiation Planning in Lung Cancer

Purpose/Objective(s): Radiation therapy in lung cancer relies on CT and functional imaging (FDG-PET) to delineate tumor volumes. Semi-automatic contouring tools have been developed for PET to improve on the inter-observer bias of manual contouring and intrinsic differences in imaging equipment. A common method involves using a threshold at a given percentage of the max activity, which may be less accurate with smaller tumors and tumors with low source to background ratio. To overcome this deficiency, a gradient algorithm, which detects changes in image counts at the border of the tumor, has been developed. Few studies have correlated these methods to pathological specimens.