Atlas-based Segmentation Method Research Articles

Single annotator versus multi-annotator: Challenge of segmenting two neighboring hippocampus head and body with high precision

Background and objectiveThis work investigates multi-annotator segmentation framework for semantic image segmentation using deep learning approaches. MethodsThe framework consists of various deep learning models which perform independently, and then their decisions are fused by a combiner network. The underlying idea of the multi-annotator (ensemble) segmentation framework is to take advantage of the synergistic effect created by the complementary information provided by the different annotators to reach a peak performance superior to each of the annotators when performing alone. The performance of the multi-annotator framework was assessed for the challenging task of the hippocampus head and body segmentation from MR images. Atlas-based segmentation method, deep learning-based multi-view segmentation approach, as well as different decision fusion algorithms, such as STAPLE and shape-based averaging, were assessed versus the proposed multi-annotator segmentation framework. ResultsAmong the different deep learning architectures and loss functions, ResNet model with cross-entropy loss function (ResNet-CE) exhibited the best performance (best performing single model) with a Dice of 88.6 ± 1.8 % and 88.6 ± 1.8 % for the hippocampus head and body, respectively. However, the multi-annotator segmentation framework (which also involved Resnet-CE as one of the annotators) exhibited superior performance with Dice of 91.1 ± 1.3 % (head) and 91.0 ± 1.3 % (body) compared to the atlas-based method with Dice of 88.5 ± 1.5 % (head) and 88.4 ± 1.5 % (body) and multi-view approach with Dice of 89.0 ± 1.4 % (head) and 88.9 ± 1.5 % (body). ConclusionThe multi-annotator segmentation framework exhibits superior performance over each annotator alone which could be employed almost for all machine learning-related problems to enhance the overall effectiveness of the deep/machine learning algorithms.

Optimized Atlas-Based Auto-Segmentation of Bony Structures from Whole-Body Computed Tomography

To develop and test a method for fully automated segmentation of bony structures from whole-body CT and evaluate its performance compared to manual segmentation METHODS AND MATERIALS;: We developed a workflow for automatic whole-body bone segmentation using atlas-based segmentation method with a post-processing module (ABSPP) in MIM MAESTRO software. 52 CT scans comprised the training set to build the atlas library and 29 CT scans comprised the test set. To validate the workflow, we compared Dice similarity coefficient (DSC), mean distance to agreement (MDA), and relative volume errors (ΔV) between ABSPP and atlas-based segmentation with no post-processing (ABSNPP), with manual segmentation as the reference (gold standard). RESULTS;: The ABSPP method resulted in significantly improved segmentation accuracy (DSC range: 0.85 to 0.98) compared to the ABSNPP method (DSC range: 0.55 to 0.87) (p<0.001). MDA results also indicated high agreement between ABSPP and manual reference delineations (range: 0.11-1.56 mm), which was significantly improved compared to ABSNPP (range: 1.00-2.34 mm) for for the majority of tested bony structures. Relative volume errors were also significantly lower for ABSPP compared to ABSNPP for most bony structures. We developed a fully automated MIM workflow for bony structure segmentation from whole-body CT, which exhibited high accuracy compared to manual delineation. The integrated post-processing module significantly improved workflow performance.

Segmentation of intrinsically very low contrast magnetic resonance brain images using tensor-based DTI registration

BackgroundIn patients with hypomyelinating leukodystrophies, contrast of T1-weighted brain MRI is very low due to the lack of myelin, preventing a reliable segmentation. In diffusion tensor images the contrast is higher, thanks to anisotropy and orientation of white matter (WM) tracts. We aimed to develop and assess a tensor-guided atlas-based segmentation method suitable for segmentation of very low contrast images. Methods17 control subjects (mean age 8.0 yrs (SD 8.0)) and 27 subjects with hypomyelinating leukodystrophies (mean age 10.7 yrs (SD 10.2)) were included. DTI and 3D T1 images were segmented using a DTI-TK tensor-guided IIT-atlas-based segmentation method. For the control subjects, these segmentations were compared with a conventional segmentation of their 3D T1-weighted images. A qualitative visual assessment and a quantitative assessment using DTI metrics was performed to assess the patient segmentations. ResultsIn control subjects, the tensor-based method performed as can be expected for atlas-based segmentation methods, with Dice coefficients of 0.65, 0.72, 0.81 and 0.86 for cortical grey matter (GM), WM, deep grey matter (DGM), and thalamus, respectively. In patients with hypomyelination the visual assessment showed anatomically adequate segmentations. All tissue-specific DTI metrics differed between patients and controls. Patients with hypomyelination had reduced FA and increased mean, axial and radial diffusivities, not only in total WM, but also in the corticospinal tracts, optic radiations and thalamus. ConclusionEven in the absence of normal myelin, the presence and direction of axons allowed tensor-based registration and thereby atlas-based segmentation. We showed the applicability of the segmentation method in the context of quantitative MRI, allowing for whole-brain or regional tissue-specific and tract-specific analyses of very low contrast images.

Automated 3D Fetal Brain Segmentation Using an Optimized Deep Learning Approach.

MR imaging provides critical information about fetal brain growth and development. Currently, morphologic analysis primarily relies on manual segmentation, which is time-intensive and has limited repeatability. This work aimed to develop a deep learning-based automatic fetal brain segmentation method that provides improved accuracy and robustness compared with atlas-based methods. A total of 106 fetal MR imaging studies were acquired prospectively from fetuses between 23 and 39 weeks of gestation. We trained a deep learning model on the MR imaging scans of 65 healthy fetuses and compared its performance with a 4D atlas-based segmentation method using the Wilcoxon signed-rank test. The trained model was also evaluated on data from 41 fetuses diagnosed with congenital heart disease. The proposed method showed high consistency with the manual segmentation, with an average Dice score of 0.897. It also demonstrated significantly improved performance (P < .001) based on the Dice score and 95% Hausdorff distance in all brain regions compared with the atlas-based method. The performance of the proposed method was consistent across gestational ages. The segmentations of the brains of fetuses with high-risk congenital heart disease were also highly consistent with the manual segmentation, though the Dice score was 7% lower than that of healthy fetuses. The proposed deep learning method provides an efficient and reliable approach for fetal brain segmentation, which outperformed segmentation based on a 4D atlas and has been used in clinical and research settings.

Abstract 3071: A method for infratentorial structures segmentation for tumor treating fields treatment planning

Abstract Introduction: Tumor treating fields (TTFields) is an FDA approved treatment for the management of glioblastoma multiform (GBM) and mesothelioma (MPM) and is associated with a significant extension to patients' survival. We have developed software to optimize the location of the TTFields transducer arrays (TAs) on GBM patients' head to maximize the field in the tumor region. To that end, we first segment the patients' head MRIs into five normal and three abnormal tissues. However, the current method for head segmentation was found to be less accurate at the infratentorial regions, mark some sinuses with cerebrospinal fluid (CSF), and does not identify the cerebellum and brain-stem. Therefore, it is assuming the infratentorial structures have the same electrical properties as in the cerebrum. Methods: We have developed a new method for segmentation of the head MRI that overcomes these limitations. As in the previous method, we have incorporated an atlas that is composed of an MRI image and tissue probability maps (TPM). The TPM assign each voxel in the atlas MRI with a value in the range of 0-1. This value reflects the probability that the specific tissue resides in that voxel. We have generated new customized TPMs for the cerebellum, brain stem and sinuses. Moreover, we have carefully revised the other TPMs to ensure best results. Last, we have incorporated these TPMs in a new atlas-based segmentation method. We have validated the method on 10 GBM patients T1w +gad head MRIs. Results: The average Dice coefficient between gold-standard and algorithms' segmentation was 68.4% (SD=11.2%) that is similar to the value observed with the current method implemented in the treatment-planning software (67.8%; SD=10.6%). However, the new method presented here extends the current method with the cerebellum (average Dice = 84%) and brainstem (average Dice = 67%). Moreover, the sinuses are marked with air, and erroneous CSF segments in the sinuses were removed. Conclusions: We have presented a novel method for the segmentation of the head that facilitates accurate infratentorial and supratentorial structures' segmentation. Moreover, the method eliminates erroneous CSF segmentation in the sinuses. These improvements pave the way for TTFields planning for infratentorial tumors. Citation Format: Yana Glozman, Reuben R. Shamir, Zeev Bomzon. A method for infratentorial structures segmentation for tumor treating fields treatment planning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 3071.

Application of an automatic segmentation method for evaluating cardiac structure doses received by breast radiotherapy patients.

Quantifying radiation dose to cardiac substructures is important for research on the etiology and prevention of complications following radiotherapy; however, segmentation of substructures is challenging. In this study we demonstrate the application of our atlas-based automatic segmentation method to breast cancer radiotherapy plans for generating radiation doses in support of late effects research. We applied our segmentation method to contour heart substructures on the computed tomography (CT) images of 70 breast cancer patients who received external photon radiotherapy. Two cardiologists provided manual segmentation of the whole heart (WH), left/right atria, left/right ventricles, and left anterior descending artery (LAD). The automatically contours were compared with manual delineations to evaluate similarity in terms of geometry and dose. The mean Dice similarity coefficient between manual and automatic segmentations was 0.96 for the WH, 0.65 to 0.82 for the atria and ventricles, and 0.06 for the LAD. The mean average surface distance was 1.2mm for the WH, 3.4 to 4.1mm for the atria and ventricles, and 6.4mm for the LAD. We found the dose to the cardiac substructures based on our automatic segmentation agrees with manual segmentation within expected observer variability. For left breast patients, the mean absolute difference in mean dose was 0.1Gy for the WH, 0.2 to 0.7Gy for the atria and ventricles, and 1.8Gy for the LAD. For right breast patients, these values were 0.0Gy, 0.1 to 0.4Gy, and 0.4Gy, respectively. Our automatic segmentation method will facilitate the development of radiotherapy prescriptive criteria for mitigating cardiovascular complications.

Multi-Class Probabilistic Atlas-Based Whole Heart Segmentation Method in Cardiac CT and MRI

Accurate and robust whole heart substructure segmentation is crucial in developing clinical applications, such as computer-aided diagnosis and computer-aided surgery. However, the segmentation of different heart substructures is challenging because of inadequate edge or boundary information, the complexity of the background and texture, and the diversity in different substructures’ sizes and shapes. This article proposes a framework for multi-class whole heart segmentation employing non-rigid registration-based probabilistic atlas incorporating the Bayesian framework. We also propose a non-rigid registration pipeline utilizing a multi-resolution strategy for obtaining the highest attainable mutual information between the moving and fixed images. We further incorporate non-rigid registration into the expectation-maximization algorithm and implement different deep convolutional neural network-based encoder-decoder networks for ablation studies. All the extensive experiments are conducted utilizing the publicly available dataset for the whole heart segmentation containing 20 MRI and 20 CT cardiac images. The proposed approach exhibits an encouraging achievement, yielding a mean volume overlapping error of 14.5% for CT scans exceeding the state-of-the-art results by a margin of 1.3% in terms of the same metric. As the proposed approach provides better results to delineate the different substructures of the heart, it can be a medical diagnostic aiding tool for helping experts with quicker and more accurate results.

Automated multiclass tissue segmentation of clinical brain MRIs with lesions

Delineation and quantification of normal and abnormal brain tissues on Magnetic Resonance Images is fundamental to the diagnosis and longitudinal assessment of neurological diseases. Here we sought to develop a convolutional neural network for automated multiclass tissue segmentation of brain MRIs that was robust at typical clinical resolutions and in the presence of a variety of lesions. We trained a 3D U-Net for full brain multiclass tissue segmentation from a prior atlas-based segmentation method on an internal dataset that consisted of 558 clinical T1-weighted brain MRIs (453/52/53; training/validation/test) of patients with one of 50 different diagnostic entities (n = 362) or with a normal brain MRI (n = 196). We then used transfer learning to refine our model on an external dataset that consisted of 7 patients with hand-labeled tissue types. We evaluated the tissue-wise and intra-lesion performance with different loss functions and spatial prior information in the validation set and applied the best performing model to the internal and external test sets. The network achieved an average overall Dice score of 0.87 and volume similarity of 0.97 in the internal test set. Further, the network achieved a median intra-lesion tissue segmentation accuracy of 0.85 inside lesions within white matter and 0.61 inside lesions within gray matter. After transfer learning, the network achieved an average overall Dice score of 0.77 and volume similarity of 0.96 in the external dataset compared to human raters. The network had equivalent or better performance than the original atlas-based method on which it was trained across all metrics and produced segmentations in a hundredth of the time. We anticipate that this pipeline will be a useful tool for clinical decision support and quantitative analysis of clinical brain MRIs in the presence of lesions.

Atlas-based liver segmentation for nonhuman primate research.

Certain viral infectious diseases cause systemic damage and the liver is an important organ affected directly by the virus and/or the hosts' response to the virus. Medical imaging indicates that the liver damage is heterogenous, and therefore, quantification of these changes requires analysis of the entire organ. Delineating the liver in preclinical imaging studies is a time-consuming and difficult task that would benefit from automated liver segmentation. A nonhuman primate atlas-based liver segmentation method was developed to support quantitative image analysis of preclinical research. A set of 82 computed tomography (CT) scans of nonhuman primates with associated manual contours delineating the liver was generated from normal and abnormal livers. The proposed technique uses rigid and deformable registrations, a majority vote algorithm, and image post-processing operations to automate the liver segmentation process. This technique was evaluated using Dice similarity, Hausdorff distance measures, and Bland-Altman plots. Automated segmentation results compare favorably with manual contouring, achieving a median Dice score of 0.91. Limits of agreement from Bland-Altman plots indicate that liver changes of 3 Hounsfield units (CT) and 0.4 SUVmean (positron emission tomography) are detectable using our automated method of segmentation, which are substantially less than changes observed in the host response to these viral infectious diseases. The proposed atlas-based liver segmentation technique is generalizable to various sizes and species of nonhuman primates and facilitates preclinical infectious disease research studies. While the image analysis software used is commercially available and facilities with funding can access the software to perform similar nonhuman primate liver quantitative analyses, the approach can be implemented in open-source frameworks as there is nothing proprietary about these methods.

A Multi-atlas Approach for Active Bone Marrow Sparing Radiation Therapy: Implementation in the NRG-GY006 Trial

Sparing active bone marrow (ABM) can reduce acute hematologic toxicity in patients undergoing chemoradiotherapy for cervical cancer, but ABM segmentation based on positron emission tomography/computed tomography (PET/CT) is costly. We sought to develop an atlas-based ABM segmentation method for implementation in a prospective clinical trial. A multiatlas was built on a training set of 144 patients and validated in 32 patients from the NRG-GY006 clinical trial. ABM for individual patients was defined as the subvolume of pelvic bone greater than the individual mean standardized uptake value on registered 18F-fluorodeoxyglucose PET/CT images. Atlas-based and custom ABM segmentations were compared using the Dice similarity coefficient and mean distance to agreement and used to generate ABM-sparing intensity modulated radiation therapy plans. Dose-volume metrics and normal tissue complication probabilities of the two approaches were compared using linear regression. Atlas-based ABM volumes (mean [standard deviation], 548.4 [88.3] cm3) were slightly larger than custom ABM volumes (535.1 [93.2] cm3), with a Dice similarity coefficient of 0.73. Total pelvic bone marrow V20 and Dmean were systematically higher and custom ABM V10 was systematically lower with custom-based plans (slope: 1.021 [95% confidence interval (CI), 1.005-1.037], 1.014 [95% CI, 1.006-1.022], and 0.98 [95% CI, 0.97-0.99], respectively). We found no significant differences between atlas-based and custom-based plans in bowel, rectum, bladder, femoral heads, or target dose-volume metrics. Atlas-based ABM segmentation can reduce pelvic bone marrow dose while achieving comparable targetandother normal tissue dosimetry. This approach may allow ABM sparing in settings where PET/CT is unavailable.

Frontal aslant tracts as correlates of lexical retrieval in MS

ABSTRACT Introduction Previous studies reveal that a newly described white matter pathway, the frontal aslant tract (FAT), connecting inferior and superior frontal gyri has a role in speech and language functions. We explored the role of this tract in a phonemic and semantic fluency tasks in a cohort of multiple sclerosis patients diagnosed with cognitive impairment. Methods Thirty-five MS patients with varying degrees of cognitive impairment underwent diffusion tensor imaging and the Controlled Associated Word Test. Fractional anisotropy (FA) of FAT and arcuate fasciculus (AF) were obtained through a supervised, atlas-based tissue segmentation and parcellation method. Phonemic and semantic fluency scores were obtained from COWAT. We ran a multivariate regression model, and partial correlation analyses adjusted for age, education, and lesion load, and corrected for multiple comparisons. False discovery rate (FDR) was used for the correction of multiple comparisons. Results Bilateral FAT FA showed significant association with phonemic verbal fluency task (Left; r = 0.46, p = 0.0058 and right; r = 0.46, p = 0.0059) but not semantic fluency task and this relation remained significant after FDR correction (p = 0.02 bilaterally). Although left AF showed some significant association with phonemic fluency task, this relation was insignificant after FDR correction. Conclusion We show that bilateral FAT are correlates of phonemic verbal fluency task but not semantic in an MS cohort with cognitive impairment. This finding suggests that FAT is more specialized in lexical retrieval function as semantic fluency test encompasses all the functions except the lexical retrieval.

1052: Brain composition differences in fetuses after laser surgery for twin-twin transfusion syndrome (TTTS)

We describe the maturational pattern differences of various brain tissues between donor versus recipient twins treated with in utero laser surgery for TTTS using 3D MRI reconstruction. This was a prospective study of 20 patients with TTTS and dual survivors 4 weeks postoperatively. Fetal MRI scans were performed with a 1.5-T scanner. Multiple stacks of singleshot fast spin-echo T2-weighted slices of the fetal brain were obtained in the axial, sagittal, and coronal planes for each fetus. The images were reconstructed into 3D MRI volumes after motion correction. Total brain volume (TBV), cortical plate region (CR), developing white matter region (WMR), cerebellum, and ventricles were delineated using a semi-automated, atlas based segmentation method (Figure 1(a)). CR, WMR, cerebellum, and ventricular volumes were normalized by TBV for each subject. The CR and WMR were additionally normalized to the supratentorial structures only, called total parenchymal volume (TPV). Relationship between gestational age (GA), twin status (donor/recipient) and normalized tissue volumes were modeled using robust, multiple linear regression and tested for significance (defined as p<0.05). Eighteen of 20 twin pairs had sufficient image quality for analysis. The median gestational age (GA) at laser surgery was 19.8 (16.4-26.4) weeks, and 25.1 (22.1-30.7) weeks at MRI examination. Figure 1 (b-f) shows the comparative relationship between normalized tissue volumes and gestational age between the donor and recipient twin pairs. TBV was significantly lower in the donor twins (Figure 1b), although TPV were not different between the twin pairs (Figure 1c). Recipients showed significantly lower proportions of CR and cerebellar volumes normalized to TBV compared to donors (Figure 1d,e). Ventricular volumes normalized to TBV were significantly higher in the donor twins (Figure 1f). No significant differences were detected in WMR volumes between the two groups. Recipients showed significantly lower proportions of CR normalized to TPV compared to donors. Although the total brain volume of the donor twin was smaller than that of the recipient twin, the proportional brain tissue volumes of the donor were preserved. Rather, it was the recipient twin that had relatively reduced proportions of cortical plate region and cerebellar volumes.

Automatic liver segmentation in computed tomography using general-purpose shape modeling methods

BackgroundLiver segmentation in computed tomography is required in many clinical applications. The segmentation methods used can be classified according to a number of criteria. One important criterion for method selection is the shape representation of the segmented organ. The aim of the work is automatic liver segmentation using general purpose shape modeling methods.MethodsAs part of the research, methods based on shape information at various levels of advancement were used. The single atlas based segmentation method was used as the simplest shape-based method. This method is derived from a single atlas using the deformable free-form deformation of the control point curves. Subsequently, the classic and modified Active Shape Model (ASM) was used, using medium body shape models. As the most advanced and main method generalized statistical shape models, Gaussian Process Morphable Models was used, which are based on multi-dimensional Gaussian distributions of the shape deformation field.ResultsMutual information and sum os square distance were used as similarity measures. The poorest results were obtained for the single atlas method. For the ASM method in 10 analyzed cases for seven test images, the Dice coefficient was above 55%, of which for three of them the coefficient was over 70%, which placed the method in second place. The best results were obtained for the method of generalized statistical distribution of the deformation field. The DICE coefficient for this method was 88.5%ConclusionsThis value of 88.5 % Dice coefficient can be explained by the use of general-purpose shape modeling methods with a large variance of the shape of the modeled object—the liver and limitations on the size of our training data set, which was limited to 10 cases. The obtained results in presented fully automatic method are comparable with dedicated methods for liver segmentation. In addition, the deforamtion features of the model can be modeled mathematically by using various kernel functions, which allows to segment the liver on a comparable level using a smaller learning set.

Breast-region segmentation in MRI using chest region atlas and SVM

An important step for computerized analysis of breast magnetic resonance imaging (MRI) is segmentation of the breast region. Due to the similar signal intensity of fibroglandular tissue and the chest wall, the segmentation process is difficult for breasts with fibroglandular tissue connected to the chest wall. In order to overcome this challenge, a new framework is presented that relies on a chest region atlas. The proposed method first detects the approximated breast-chest wall boundary using an intensity-based operation. A support vector machine (SVM) then determines the connectivity of fibroglandular tissue to the chest wall by the extracted features from the obtained breast-chest wall boundary. Finally, the obtained breast-chest wall boundary is accurately refined using the geometric shape of the chest region, which is obtained by an atlas-based segmentation method. The proposed method is validated using a dataset of 5964 breast MRI images from 126 women. The Dice similarity coefficient (DSC), total overlap (TO), false negative (FN), and false positive (FP) values are calculated to measure the similarity between automatic and manual segmentation results. Our method achieves DSC, TO, FN, and FP values of 96.46%, 96.41%, 3.59%, and 3.51%, respectively. The results prove the effectiveness of the presented algorithm for breasts with different sizes, shapes, and density patterns.

Atlas-based segmentation technique incorporating inter-observer delineation uncertainty for whole breast

Accurate, efficient auto-segmentation methods are essential for the clinical efficacy of adaptive radiotherapy delivered with highly conformal techniques. Current atlas based auto-segmentation techniques are adequate in this respect, however fail to account for inter-observer variation. An atlas-based segmentation method that incorporates inter-observer variation is proposed. This method is validated for a whole breast radiotherapy cohort containing 28 CT datasets with CTVs delineated by eight observers. To optimise atlas accuracy, the cohort was divided into categories by mean body mass index and laterality, with atlas’ generated for each in a leave-one-out approach. Observer CTVs were merged and thresholded to generate an auto-segmentation model representing both inter-observer and inter-patient differences. For each category, the atlas was registered to the left-out dataset to enable propagation of the auto-segmentation from atlas space. Auto-segmentation time was recorded. The segmentation was compared to the gold-standard contour using the dice similarity coefficient (DSC) and mean absolute surface distance (MASD). Comparison with the smallest and largest CTV was also made. This atlas-based auto-segmentation method incorporating inter-observer variation was shown to be efficient (<4min) and accurate for whole breast radiotherapy, with good agreement (DSC>0.7, MASD <9.3mm) between the auto-segmented contours and CTV volumes.

A Local ROI-specific Atlas-based Segmentation of Prostate Gland and Transitional Zone in Diffusion MRI

Segmentation of prostate and related anatomic structure, such as transitional zone, in medical images facilitates prostate cancer detection, as well as a number of other clinical practices. In this paper, we propose a semi-automatic local ROI-specific atlas-based segmentation (LABS) method to segment prostate gland and transitional zone in diffusion magnetic resonance images. Inspired by a sequential registration-based segmentation method, the proposed method further reduces the amount of user intervention and focuses on the vicinity of prostate for atlas matching and atlas-to-target registration by specifying the bounding boxes of prostate gland on key slices of volume images. We evaluated the method on an atlas database with the 100 cases by performing a leave-one-out study. Our proposed method produced favorable outcomes with an average Dice similarity coefficient of 0.85±0.03 for prostate gland and 0.77±0.06 for transitional zone segmentations, which indicates the effectiveness of the proposed method and its potential to be used in relevant clinical applications.

SU-D-303-01: Spatial Distribution of Bone Metastases In Metastatic Castrate-Resistant Prostate Cancer

Purpose: Identification of metastatic bone lesions is critical in prostate cancer, where treatments may be more effective in patients with fewer lesions. This study aims characterize the distribution and spread of bone lesions and create a probability map of metastatic spread in bone. Methods: Fifty-five metastatic castrate-resistant prostate cancer patients received up to 3 whole-body [F-18]NaF PET/CT scans. Lesions were identified by physician on PET/CT and contoured using a threshold of SUV>15. An atlas-based segmentation method was used to create CT regions, which determined skeletal location of lesions. Patients were divided into 3 groups with low (N<40), medium (40100) numbers of lesions. A combination of articulated and deformable registrations was used to register the skeletal segments and lesions of each patient to a single skeleton. All the lesion data was then combined to make a probability map. Results: A total of 4038 metastatic lesions (mean 74, range 2–304) were identified. Skeletal regions with highest occurrence of lesions included ribs, thoracic spine, and pelvis with 21%, 19%, and 15% of the total number lesions and 8%, 18%, and 31 % of the total lesion volume, respectively. Interestingly, patients with fewer lesions were found to have a lower proportion of lesions inmore » the ribs (9% in low vs. 27% in high number of lesions). Additionally, the probability map showed specific areas in the spine and pelvis where over 75% of patients had metastases, and other areas in the skeleton with a less than 2% of metastases. Conclusion: We identified skeletal regions with higher incidence of metastases and specific sub-regions in the skeleton that had high or low probability of occurrence of metastases. Additionally, we found that metastatic lesions in the ribs and skull occur more commonly in advanced disease. These results may have future applications in computer-aided diagnosis. Funding from the Prostate Cancer Foundation.« less

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.

Evaluation of an Atlas-Based Segmentation Method for Prostate MRI

Evaluation of an Atlas-Based Segmentation Method for Prostate MRI

SU-E-J-234: Auto-Segmentation Strategies for Treatment Targets and Critical Organs in Head-And-Neck Cancer Patients

Purpose: To determine an optimum combination of model- and atlas-based segmentation methods for the automatic segmentation of targets and critical organs for head-and-neck cancer patients. Methods: Ten base-of-tongue cancer patients with clinical staging ranging from T1N2bM0 to T2N2cM0 were selected retrospectively for this study. Patient CT images were imported into RayStation TPS (V4.0) where both model-based segmentation (MBS) and atlas-based segmentation (ABS) are available. MBS includes models with adjustable shape, size and property parameters provided by RayStation for the critical organs of spinal cord, brainstem, parotids, and mandible. ABS requires user-defined atlases with images and contours. To create atlases specifically for our patient cases, “standard” patients were selected by physicians for each TMN stage. The ABS contours include all MBS contours plus the primary and nodal GTV and up to three levels of CTV. Contours for each patient were auto-delineated using MBS and ABS methods which were then modified by physicians and used as ground truth. Auto-segmented and physician corrected contours were compared and evaluated using Dice similarity coefficient. Results: The auto-segmented contours were considered acceptable if the Dice values were above 0.7. For the cord, brainstem, parotids(L+R), and mandible, the number of patients whose Dice value was acceptable were 0, 4, 5, 7, and 10, respectively for MBS method, and 4, 4, 5, 4, and 7, respectively for ABS method. For the targets, none of the auto-segmented contour had an accpetable Dice value. The best one was CTV2 with the average Dice of 0.5. Conclusion: With the exception of cord, MBS was better for delineation critical organs. Auto-segmented GTV and CTV contours using ABS method agreed poorly with physician corrected contours because of the large amount of inter-patient variation in disease location even with the exact same TMN staging. Better auto-segmentation methods should be explored to improve target delineation.