Median Dice Similarity Coefficient Research Articles

Deep Learning Algorithm for Auto-Delineation of High-Risk Oropharyngeal Clinical Target Volumes With Built-In Dice Similarity Coefficient Parameter Optimization Function

Automating and standardizing the contouring of clinical target volumes (CTVs) can reduce interphysician variability, which is one of the largest sources of uncertainty in head and neck radiation therapy. In addition to using uniform margin expansions to auto-delineate high-risk CTVs, very little work has been performed to provide patient- and disease-specific high-risk CTVs. The aim of the present study was to develop a deep neural network for the auto-delineation of high-risk CTVs. Fifty-two oropharyngeal cancer patients were selected for the present study. All patients were treated at The University of Texas MD Anderson Cancer Center from January 2006 to August 2010 and had previously contoured gross tumor volumes and CTVs. We developed a deep learning algorithm using deep auto-encoders to identify physician contouring patterns at our institution. These models use distance map information from surrounding anatomic structures and the gross tumor volume as input parameters and conduct voxel-based classification to identify voxels that are part of the high-risk CTV. In addition, we developed a novel probability threshold selection function, based on the Dice similarity coefficient (DSC), to improve the generalization of the predicted volumes. The DSC-based function is implemented during an inner cross-validation loop, and probability thresholds are selected a priori during model parameter optimization. We performed a volumetric comparison between the predicted and manually contoured volumes to assess our model. The predicted volumes had a median DSC value of 0.81 (range 0.62-0.90), median mean surface distance of 2.8mm (range 1.6-5.5), and median 95th Hausdorff distance of 7.5mm (range 4.7-17.9) when comparing our predicted high-risk CTVs with the physician manual contours. These predicted high-risk CTVs provided close agreement to the ground-truth compared with current interobserver variability. The predicted contours could be implemented clinically, with only minor or no changes.

A heterogeneous tissue model for treatment planning for magnetic resonance-guided laser interstitial thermal therapy

We evaluated a physics-based model for planning for magnetic resonance-guided laser interstitial thermal therapy for focal brain lesions. Linear superposition of analytical point source solutions to the steady-state Pennes bioheat transfer equation simulates laser-induced heating in brain tissue. The line integral of the photon attenuation from the laser source enables computation of the laser interaction with heterogeneous tissue. Magnetic resonance thermometry data sets (n = 31) were used to calibrate and retrospectively validate the model’s thermal ablation prediction accuracy, which was quantified by the Dice similarity coefficient (DSC) between model-predicted and measured ablation regions (T > 57 °C). A Gaussian mixture model was used to identify independent tissue labels on pre-treatment anatomical magnetic resonance images. The tissue-dependent optical attenuation coefficients within these labels were calibrated using an interior point method that maximises DSC agreement with thermometry. The distribution of calibrated tissue properties formed a population model for our patient cohort. Model prediction accuracy was cross-validated using the population mean of the calibrated tissue properties. A homogeneous tissue model was used as a reference control. The median DSC values in cross-validation were 0.829 for the homogeneous model and 0.840 for the heterogeneous model. In cross-validation, the heterogeneous model produced a DSC higher than that produced by the homogeneous model in 23 of the 31 brain lesion ablations. Results of a paired, two-tailed Wilcoxon signed-rank test indicated that the performance improvement of the heterogeneous model over that of the homogeneous model was statistically significant (p < 0.01).

Automatic segmentation method of pelvic floor levator hiatus in ultrasound using a self-normalizing neural network.

.Segmentation of the levator hiatus in ultrasound allows the extraction of biometrics, which are of importance for pelvic floor disorder assessment. We present a fully automatic method using a convolutional neural network (CNN) to outline the levator hiatus in a two-dimensional image extracted from a three-dimensional ultrasound volume. In particular, our method uses a recently developed scaled exponential linear unit (SELU) as a nonlinear self-normalizing activation function, which for the first time has been applied in medical imaging with CNN. SELU has important advantages such as being parameter-free and mini-batch independent, which may help to overcome memory constraints during training. A dataset with 91 images from 35 patients during Valsalva, contraction, and rest, all labeled by three operators, is used for training and evaluation in a leave-one-patient-out cross validation. Results show a median Dice similarity coefficient of 0.90 with an interquartile range of 0.08, with equivalent performance to the three operators (with a Williams’ index of 1.03), and outperforming a U-Net architecture without the need for batch normalization. We conclude that the proposed fully automatic method achieved equivalent accuracy in segmenting the pelvic floor levator hiatus compared to a previous semiautomatic approach.

Evaluation of diffusion weighted imaging for tumor delineation in head-and-neck radiotherapy by comparison with automatically segmented 18F-fluorodeoxyglucose positron emission tomography.

Background and purposeDiffusion weighted (DW) MRI may facilitate target volume delineation for head-and-neck (HN) radiation treatment planning. In this study we assessed the use of a dedicated, geometrically accurate, DW-MRI sequence for target volume delineation. The delineations were compared with semi-automatic segmentations on 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) images and evaluated for interobserver variation.Methods and materialsFifteen HN cancer patients underwent both DW-MRI and FDG-PET for RT treatment planning. Target delineation on DW-MRI was performed by three observers, while for PET a semi-automatic segmentation was performed using a Gaussian mixture model. For interobserver variation and intermodality variation, volumes, overlap metrics and Hausdorff distances were calculated from the delineations.ResultsThe median volumes delineated by the three observers on DW-MRI were 10.8, 10.5 and 9.0 cm3 respectively, and was larger than the median PET volume (8.0 cm3). The median conformity index of DW-MRI for interobserver variation was 0.73 (range 0.38–0.80). Compared to PET, the delineations on DW-MRI by the three observers showed a median dice similarity coefficient of 0.71, 0.69 and 0.72 respectively. The mean Hausdorff distance was small with median (range) distances between PET and DW-MRI of 2.3 (1.5–6.8), 2.5 (1.6–6.9) and 2.0 (1.35–7.6) mm respectively. Over all patients, the median 95th percentile distances were 6.0 (3.0–13.4), 6.6 (4.0–24.0) and 5.3 (3.4–26.0) mm.ConclusionUsing a dedicated DW-MRI sequence, target volumes could be defined with good interobserver agreement and a good overlap with PET. Target volume delineation using DW-MRI is promising in head-and-neck radiotherapy, combined with other modalities, it can lead to more precise target volume delineation.

Fully automated segmentation of prostate whole gland and transition zone in diffusion-weighted MRI using convolutional neural networks.

Prostate cancer is a leading cause of cancer-related death among men. Multiparametric magnetic resonance imaging has become an essential part of the diagnostic evaluation of prostate cancer. The internationally accepted interpretation scheme (Pi-Rads v2) has different algorithms for scoring of the transition zone (TZ) and peripheral zone (PZ) of the prostate as tumors can appear different in these zones. Computer-aided detection tools have shown different performances in TZ and PZ and separating these zones for training and detection is essential. The TZ-PZ segmentation which requires the segmentation of prostate whole gland and TZ is typically done manually. We present a fully automatic algorithm for delineation of the prostate gland and TZ in diffusion-weighted imaging (DWI) via a stack of fully convolutional neural networks. The proposed algorithm first detects the slices that contain a portion of prostate gland within the three-dimensional DWI volume and then it segments the prostate gland and TZ automatically. The segmentation stage of the algorithm was applied to DWI images of 104 patients and median Dice similarity coefficients of 0.93 and 0.88 were achieved for the prostate gland and TZ, respectively. The detection of image slices with and without prostate gland had an average accuracy of 0.97.

Automatic delineation of functional lung volumes with 68Ga-ventilation/perfusion PET/CT

BackgroundFunctional volumes computed from 68Ga-ventilation/perfusion (V/Q) PET/CT, which we have shown to correlate with pulmonary function test parameters (PFTs), have potential diagnostic utility in a variety of clinical applications, including radiotherapy planning. An automatic segmentation method would facilitate delineation of such volumes. The aim of this study was to develop an automated threshold-based approach to delineate functional volumes that best correlates with manual delineation.Thirty lung cancer patients undergoing both V/Q PET/CT and PFTs were analyzed. Images were acquired following inhalation of Galligas and, subsequently, intravenous administration of 68Ga-macroaggreted-albumin (MAA). Using visually defined manual contours as the reference standard, various cutoff values, expressed as a percentage of the maximal pixel value, were applied. The average volume difference and Dice similarity coefficient (DSC) were calculated, measuring the similarity of the automatic segmentation and the reference standard. Pearson’s correlation was also calculated to compare automated volumes with manual volumes, and automated volumes optimized to PFT indices.ResultsFor ventilation volumes, mean volume difference was lowest (− 0.4%) using a 15%max threshold with Pearson’s coefficient of 0.71. Applying this cutoff, median DSC was 0.93 (0.87–0.95). Nevertheless, limits of agreement in volume differences were large (− 31.0 and 30.2%) with differences ranging from − 40.4 to + 33.0%.For perfusion volumes, mean volume difference was lowest and Pearson’s coefficient was highest using a 15%max threshold (3.3% and 0.81, respectively). Applying this cutoff, median DSC was 0.93 (0.88–0.93). Nevertheless, limits of agreement were again large (− 21.1 and 27.8%) with volume differences ranging from − 18.6 to + 35.5%.Using the 15%max threshold, moderate correlation was demonstrated with FEV1/FVC (r = 0.48 and r = 0.46 for ventilation and perfusion images, respectively). No correlation was found between other PFT indices.ConclusionsTo automatically delineate functional volumes with 68Ga-V/Q PET/CT, the most appropriate cutoff was 15%max for both ventilation and perfusion images. However, using this unique threshold systematically provided unacceptable variability compared to the reference volume and relatively poor correlation with PFT parameters. Accordingly, a visually adapted semi-automatic method is favored, enabling rapid and quantitative delineation of lung functional volumes with 68Ga-V/Q PET/CT.

Respiratory-gated (4D) contrast-enhanced FDG PET-CT for radiotherapy planning of lower oesophageal carcinoma: feasibility and impact on planning target volume

BackgroundTo assess the feasibility and potential impact on target delineation of respiratory-gated (4D) contrast-enhanced 18Fluorine fluorodeoxyglucose (FDG) positron emission tomography - computed tomography (PET-CT), in the treatment planning position, for a prospective cohort of patients with lower third oesophageal cancer.MethodsFifteen patients were recruited into the study. Imaging included 4D PET-CT, 3D PET-CT, endoscopic ultrasound and planning 4D CT. Target volume delineation was performed on 4D CT, 4D CT with co-registered 3D PET and 4D PET-CT. Planning target volumes (PTV) generated with 4D CT (PTV4DCT), 4D CT co-registered with 3D PET-CT (PTV3DPET4DCT) and 4D PET-CT (PTV4DPETCT) were compared with multiple positional metrics.ResultsMean PTV4DCT, PTV3DPET4DCT and PTV4DPETCT were 582.4 ± 275.1 cm3, 472.5 ± 193.1 cm3 and 480.6 ± 236.9 cm3 respectively (no significant difference). Median DICE similarity coefficients comparing PTV4DCT with PTV3DPET4DCT, PTV4DCT with PTV4DPETCT and PTV3DPET4DCT with PTV4DPETCT were 0.85 (range 0.65–0.9), 0.85 (range 0.69–0.9) and 0.88 (range 0.79–0.9) respectively. The median sensitivity index for overlap comparing PTV4DCT with PTV3DPET4DCT, PTV4DCT with PTV4DPETCT and PTV3DPET4DCT with PTV4DPETCT were 0.78 (range 0.65–0.9), 0.79 (range 0.65–0.9) and 0.89 (range 0.68–0.94) respectively.ConclusionsPlanning 4D PET-CT is feasible with careful patient selection. PTV generated using 4D CT, 3D PET-CT and 4D PET-CT were of similar volume, however, overlap analysis demonstrated that approximately 20% of PTV3DPETCT and PTV4DPETCT are not included in PTV4DCT, leading to under-coverage of target volume and a potential geometric miss. Additionally, differences between PTV3DPET4DCT and PTV4DPETCT suggest a potential benefit for 4D PET-CT.Trial registrationClinicalTrials.gov Identifier – NCT02285660 (Registered 21/10/2014).

Target volume delineation of anal cancer based on magnetic resonance imaging or positron emission tomography

PurposeTo compare target volume delineation of anal cancer using positron emission tomography (PET) and magnetic resonance imaging (MRI) with respect to inter-observer and inter-modality variability.MethodsNineteen patients with anal cancer undergoing chemoradiotherapy were prospectively included. Planning computed tomography (CT) images were co-registered with 18F–fluorodexocyglucose (FDG) PET/CT images and T2 and diffusion weighted (DW) MR images. Three oncologists delineated the Gross Tumor Volume (GTV) according to national guidelines and the visible tumor tissue (GTVT). MRI and PET based delineations were evaluated by absolute volumes and Dice similarity coefficients.ResultsThe median volume of the GTVs was 27 and 31 cm3 for PET and MRI, respectively, while it was 6 and 11 cm3 for GTVT. Both GTV and GTVT volumes were highly correlated between delineators (r = 0.90 and r = 0.96, respectively). The median Dice similarity coefficient was 0.75 when comparing the GTVs based on PET/CT (GTVPET) with the GTVs based on MRI and CT (GTVMRI). The median Dice coefficient was 0.56 when comparing the visible tumor volume evaluated by PET (GTVT_PET) with the same volume evaluated by MRI (GTVT_MRI). Margins of 1–2 mm in the axial plane and 7–8 mm in superoinferior direction were required for coverage of the individual observer’s GTVs.ConclusionsThe rather good agreement between PET- and MRI-based GTVs indicates that either modality may be used for standard target delineation of anal cancer. However, larger deviations were found for GTVT, which may impact future tumor boost strategies.

Variability in axillary lymph node delineation for breast cancer radiotherapy in presence of guidelines on a multi-institutional platform

Aim: To quantify the variability between radiation oncologists (ROs) when outlining axillary nodes in breast cancer.Material and methods: For each participating center, three ROs with different levels of expertise, i.e., junior (J), senior (S) and expert (E), contoured axillary nodal levels (L1, L2, L3 and L4) on the CT images of three different patients (P) of an increasing degree of anatomical complexity (from P1 to P2 to P3), according to contouring guidelines. Consensus contours were generated using the simultaneous truth and performance level estimation (STAPLE) method.Results: Fifteen centers and 42 ROs participated. Overall, the median Dice similarity coefficient was 0.66. Statistically significant differences were observed according to the level of expertise (better agreement for J and E, worse for S); the axillary level (better agreement for L1 and L4, worse for L3); the patient (better agreement for P1, worse for P3). Statistically significant differences in contouring were found in 18% of the inter-center comparison. Less than a half of the centers could claim to have a good agreement between the internal ROs.Conclusions: The overall intra-institute and inter-institute agreement was moderate. Central lymph-node levels were the most critical and variability increased as the complexity of the patient’s anatomy increased. These findings might have an effect on the interpretation of results from multicenter and even mono-institute studies.

Superpixel-based segmentation of muscle fibers in multi-channel microscopy.

BackgroundConfetti fluorescence and other multi-color genetic labelling strategies are useful for observing stem cell regeneration and for other problems of cell lineage tracing. One difficulty of such strategies is segmenting the cell boundaries, which is a very different problem from segmenting color images from the real world. This paper addresses the difficulties and presents a superpixel-based framework for segmentation of regenerated muscle fibers in mice.ResultsWe propose to integrate an edge detector into a superpixel algorithm and customize the method for multi-channel images. The enhanced superpixel method outperforms the original and another advanced superpixel algorithm in terms of both boundary recall and under-segmentation error. Our framework was applied to cross-section and lateral section images of regenerated muscle fibers from confetti-fluorescent mice. Compared with “ground-truth” segmentations, our framework yielded median Dice similarity coefficients of 0.92 and higher.ConclusionOur segmentation framework is flexible and provides very good segmentations of multi-color muscle fibers. We anticipate our methods will be useful for segmenting a variety of tissues in confetti fluorecent mice and in mice with similar multi-color labels.Electronic supplementary materialThe online version of this article (doi:10.1186/s12918-016-0372-2) contains supplementary material, which is available to authorized users.

Improving the efficiency of breast radiotherapy treatment planning using a semi-automated approach.

ObjectivesTo reduce treatment planning times while maintaining plan quality through the introduction of semi‐automated planning techniques for breast radiotherapy.MethodsAutomatic critical structure delineation was examined using the Smart Probabilistic Image Contouring Engine (SPICE) commercial autosegmentation software (Philips Radiation Oncology Systems, Fitchburg, WI) for a cohort of ten patients. Semiautomated planning was investigated by employing scripting in the treatment planning system to automate segment creation for breast step‐and‐shoot planning and create objectives for segment weight optimization; considerations were made for three different multileaf collimator (MLC) configurations. Forty patients were retrospectively planned using the script and a planning time comparison performed.ResultsThe SPICE heart and lung outlines agreed closely with clinician‐defined outlines (median Dice Similarity Coefficient > 0.9); median difference in mean heart dose was 0.0 cGy (range −10.8 to 5.4 cGy). Scripted treatment plans demonstrated equivalence with their clinical counterparts. No statistically significant differences were found for target parameters. Minimal ipsilateral lung dose increases were also observed. Statistically significant (P < 0.01) time reductions were achievable for MLCi and Agility MLC (Elekta Ltd, Crawley, UK) plans (median 4.9 and 5.9 min, respectively).ConclusionsThe use of commercial autosegmentation software enables breast plan adjustment based on doses to organs at risk. Semi‐automated techniques for breast radiotherapy planning offer modest reductions in planning times. However, in the context of a typical department's breast radiotherapy workload, minor savings per plan translate into greater efficiencies overall.

Comparison of internal target volumes defined on 3-dimensional, 4-dimensonal, and cone-beam CT images of non-small-cell lung cancer.

PurposeThe purpose of this study was to compare the positional and volumetric differences of internal target volumes defined on three-dimensional computed tomography (3DCT), four-dimensional CT (4DCT), and cone-beam CT (CBCT) images of non-small-cell lung cancer (NSCLC).Materials and methodsThirty-one patients with NSCLC sequentially underwent 3DCT and 4DCT simulation scans of the thorax during free breathing. The first CBCT was performed and registered to the planning CT using the bony anatomy registration during radiotherapy. The gross tumor volumes were contoured on the basis of 3DCT, maximum intensity projection (MIP) of 4DCT, and CBCT. CTV3D (clinical target volume), internal target volumes, ITVMIP and ITVCBCT, were defined with a 7 mm margin accounting for microscopic disease. ITV10 mm and ITV5 mm were defined on the basis of CTV3D: ITV10 mm with a 5 mm margin in left–right (LR), anterior–posterior (AP) directions and 10 mm in cranial–caudal (CC) direction; ITV5 mm with an isotropic internal margin (IM) of 5 mm. The differences in the position, size, Dice’s similarity coefficient (DSC) and inclusion relation of different volumes were evaluated.ResultsThe median size ratios of ITV10 mm, ITV5 mm, and ITVMIP to ITVCBCT were 2.33, 1.88, and 1.03, respectively, for tumors in the upper lobe and 2.13, 1.76, and 1.1, respectively, for tumors in the middle-lower lobe. The median DSCs of ITV10 mm, ITV5 mm, ITVMIP, and ITVCBCT were 0.6, 0.66, and 0.83 for all patients. The median percentages of ITVCBCT not included in ITV10 mm, ITV5 mm, and ITVMIP were 0.1%, 1.63%, and 15.21%, respectively, while the median percentages of ITV10 mm, ITV5 mm, and ITVMIP not included in ITVCBCT were 57.08%, 48.89%, and 20.04%, respectively.ConclusionThe use of the individual ITV derived from 4DCT merely based on bony registration in radiotherapy may result in a target miss. The ITVs derived from 3DCT with isotropic margins have a good coverage of the ITV from CBCT, but the use of those would result in a high proportion of normal tissue being irradiated unnecessarily.

Tissue segmentation of computed tomography images using a Random Forest algorithm: a feasibility study

There is a need for robust, fully automated whole body organ segmentation for diagnostic CT. This study investigates and optimizes a Random Forest algorithm for automated organ segmentation; explores the limitations of a Random Forest algorithm applied to the CT environment; and demonstrates segmentation accuracy in a feasibility study of pediatric and adult patients. To the best of our knowledge, this is the first study to investigate a trainable Weka segmentation (TWS) implementation using Random Forest machine-learning as a means to develop a fully automated tissue segmentation tool developed specifically for pediatric and adult examinations in a diagnostic CT environment.Current innovation in computed tomography (CT) is focused on radiomics, patient-specific radiation dose calculation, and image quality improvement using iterative reconstruction, all of which require specific knowledge of tissue and organ systems within a CT image. The purpose of this study was to develop a fully automated Random Forest classifier algorithm for segmentation of neck–chest–abdomen–pelvis CT examinations based on pediatric and adult CT protocols. Seven materials were classified: background, lung/internal air or gas, fat, muscle, solid organ parenchyma, blood/contrast enhanced fluid, and bone tissue using Matlab and the TWS plugin of FIJI. The following classifier feature filters of TWS were investigated: minimum, maximum, mean, and variance evaluated over a voxel radius of 2n, (n from 0 to 4), along with noise reduction and edge preserving filters: Gaussian, bilateral, Kuwahara, and anisotropic diffusion. The Random Forest algorithm used 200 trees with 2 features randomly selected per node. The optimized auto-segmentation algorithm resulted in 16 image features including features derived from maximum, mean, variance Gaussian and Kuwahara filters. Dice similarity coefficient (DSC) calculations between manually segmented and Random Forest algorithm segmented images from 21 patient image sections, were analyzed. The automated algorithm produced segmentation of seven material classes with a median DSC of 0.86 ± 0.03 for pediatric patient protocols, and 0.85 ± 0.04 for adult patient protocols. Additionally, 100 randomly selected patient examinations were segmented and analyzed, and a mean sensitivity of 0.91 (range: 0.82–0.98), specificity of 0.89 (range: 0.70–0.98), and accuracy of 0.90 (range: 0.76–0.98) were demonstrated. In this study, we demonstrate that this fully automated segmentation tool was able to produce fast and accurate segmentation of the neck and trunk of the body over a wide range of patient habitus and scan parameters.

SU‐F‐J‐02: Flexible Training of MR‐Guided Laser Ablation Models Via Global Optimization

Purpose:MR‐guided laser ablation is currying interest in neurosurgery as a minimally invasive procedure. The procedure is planned via an assumed lesion overlaid on various anatomic and functional images. Computational models of heat transfer may further inform planning the intervention, reducing surgical morbidity and ensuring therapeutic effect. These models require physical parameter data, via less relevant literature values. Using commercialized laser ablation equipment combined with MR temperature imaging (MRTI) feedback, recent work has demonstrated the use of inverse problems as a means to gain the necessary information via retrospective analysis. Here, performance of a global inverse problem solution is explored that allows for simultaneous optimization of multiple objective functions.Methods:A steady state bioheat model is trained on MRTI patient datasets (n=20); the MRTI regions exceeding a 57°C threshold are considered lethally damaged. Optical and blood perfusion parameters are densely sampled, allowing for global optimization of the model. There are two objective functions to optimize with respect to damage prediction: Dice similarity coefficient (DSC) and Hausdorff distance (HD). DSC is an overlap metric; HD is the distance‐to‐agreement in physical distance. Predictive performance is simulated via leave‐one‐out cross‐validation (LOOCV) and is compared to the same model using naive literature values.Results:Post‐optimization and during LOOCV, mean and median DSC are 0.824 and 0.850 respectively; mean and median HD are 4.64mm and 4.03mm respectively. These compare favorably to the model using literature values. Mean and median DSC are 0.6685 and 0.6564 respectively; mean and median HD are 5.85mm and 5.58mm respectively. A two‐tailed paired Student's t‐test for the mean DSC reported p=0.0002. For mean HD, p=0.0064.Conclusion:Clinical MRTI feedback provides useful information for designing optimized laser ablation models. Global optimization facilitates simultaneous optimization of multiple objective functions. Trained models are considerably more accurate than naive literature based models as measured by DSC and HD.

Pieces-of-parts for supervoxel segmentation with global context: Application to DCE-MRI tumour delineation.

Rectal tumour segmentation in dynamic contrast-enhanced MRI (DCE-MRI) is a challenging task, and an automated and consistent method would be highly desirable to improve the modelling and prediction of patient outcomes from tissue contrast enhancement characteristics - particularly in routine clinical practice. A framework is developed to automate DCE-MRI tumour segmentation, by introducing: perfusion-supervoxels to over-segment and classify DCE-MRI volumes using the dynamic contrast enhancement characteristics; and the pieces-of-parts graphical model, which adds global (anatomic) constraints that further refine the supervoxel components that comprise the tumour. The framework was evaluated on 23 DCE-MRI scans of patients with rectal adenocarcinomas, and achieved a voxelwise area-under the receiver operating characteristic curve (AUC) of 0.97 compared to expert delineations. Creating a binary tumour segmentation, 21 of the 23 cases were segmented correctly with a median Dice similarity coefficient (DSC) of 0.63, which is close to the inter-rater variability of this challenging task. A second study is also included to demonstrate the method's generalisability and achieved a DSC of 0.71. The framework achieves promising results for the underexplored area of rectal tumour segmentation in DCE-MRI, and the methods have potential to be applied to other DCE-MRI and supervoxel segmentation problems.

Evaluation and optimization of the parameters used in multiple-atlas-based segmentation of prostate cancers in radiation therapy.

To evaluate and optimize the parameters used in multiple-atlas-based segmentation of prostate cancers in radiation therapy. A retrospective study was conducted, and the accuracy of the multiple-atlas-based segmentation was tested on 30 patients. The effect of library size (LS), number of atlases used for contour averaging and the contour averaging strategy were also studied. The autogenerated contours were compared with the manually drawn contours. Dice similarity coefficient (DSC) and Hausdorff distance were used to evaluate the segmentation agreement. Mixed results were found between simultaneous truth and performance level estimation (STAPLE) and majority vote (MV) strategies. Multiple-atlas approaches were relatively insensitive to LS. A LS of ten was adequate, and further increase in the LS only showed insignificant gain. Multiple atlas performed better than single atlas for most of the time. Using more atlases did not guarantee better performance, with five atlases performing better than ten atlases. With our recommended setting, the median DSC for the bladder, rectum, prostate, seminal vesicle and femurs was 0.90, 0.77, 0.84, 0.56 and 0.95, respectively. Our study shows that multiple-atlas-based strategies have better accuracy than single-atlas approach. STAPLE is preferred, and a LS of ten is adequate for prostate cases. Using five atlases for contour averaging is recommended. The contouring accuracy of seminal vesicle still needs improvement, and manual editing is still required for the other structures. This article provides a better understanding of the influence of the parameters used in multiple-atlas-based segmentation of prostate cancers.

A multimodality segmentation framework for automatic target delineation in head and neck radiotherapy.

To develop an automatic segmentation algorithm integrating imaging information from computed tomography (CT), positron emission tomography (PET), and magnetic resonance imaging (MRI) to delineate target volume in head and neck cancer radiotherapy. Eleven patients with unresectable disease at the tonsil or base of tongue who underwent MRI, CT, and PET/CT within two months before the start of radiotherapy or chemoradiotherapy were recruited for the study. For each patient, PET/CT and T1-weighted contrast MRI scans were first registered to the planning CT using deformable and rigid registration, respectively, to resample the PET and magnetic resonance (MR) images to the planning CT space. A binary mask was manually defined to identify the tumor area. The resampled PET and MR images, the planning CT image, and the binary mask were fed into the automatic segmentation algorithm for target delineation. The algorithm was based on a multichannel Gaussian mixture model and solved using an expectation-maximization algorithm with Markov random fields. To evaluate the algorithm, we compared the multichannel autosegmentation with an autosegmentation method using only PET images. The physician-defined gross tumor volume (GTV) was used as the "ground truth" for quantitative evaluation. The median multichannel segmented GTV of the primary tumor was 15.7 cm(3) (range, 6.6-44.3 cm(3)), while the PET segmented GTV was 10.2 cm(3) (range, 2.8-45.1 cm(3)). The median physician-defined GTV was 22.1 cm(3) (range, 4.2-38.4 cm(3)). The median difference between the multichannel segmented and physician-defined GTVs was -10.7%, not showing a statistically significant difference (p-value = 0.43). However, the median difference between the PET segmented and physician-defined GTVs was -19.2%, showing a statistically significant difference (p-value =0.0037). The median Dice similarity coefficient between the multichannel segmented and physician-defined GTVs was 0.75 (range, 0.55-0.84), and the median sensitivity and positive predictive value between them were 0.76 and 0.81, respectively. The authors developed an automated multimodality segmentation algorithm for tumor volume delineation and validated this algorithm for head and neck cancer radiotherapy. The multichannel segmented GTV agreed well with the physician-defined GTV. The authors expect that their algorithm will improve the accuracy and consistency in target definition for radiotherapy.

Automatic tissue segmentation of head and neck MR images for hyperthermia treatment planning

A hyperthermia treatment requires accurate, patient-specific treatment planning. This planning is based on 3D anatomical models which are generally derived from computed tomography. Because of its superior soft tissue contrast, magnetic resonance imaging (MRI) information can be introduced to improve the quality of these 3D patient models and therefore the treatment planning itself. Thus, we present here an automatic atlas-based segmentation algorithm for MR images of the head and neck.Our method combines multiatlas local weighting fusion with intensity modelling. The accuracy of the method was evaluated using a leave-one-out cross validation experiment over a set of 11 patients for which manual delineation were available.The accuracy of the proposed method was high both in terms of the Dice similarity coefficient (DSC) and the 95th percentile Hausdorff surface distance (HSD) with median DSC higher than 0.8 for all tissues except sclera. For all tissues, except the spine tissues, the accuracy was approaching the interobserver agreement/variability both in terms of DSC and HSD. The positive effect of adding the intensity modelling to the multiatlas fusion decreased when a more accurate atlas fusion method was used.Using the proposed approach we improved the performance of the approach previously presented for H&N hyperthermia treatment planning, making the method suitable for clinical application.

Robust optic nerve segmentation on clinically acquired computed tomography.

The optic nerve (ON) plays a critical role in many devastating pathological conditions. Segmentation of the ON has the ability to provide understanding of anatomical development and progression of diseases of the ON. Recently, methods have been proposed to segment the ON but progress toward full automation has been limited. We optimize registration and fusion methods for a new multi-atlas framework for automated segmentation of the ONs, eye globes, and muscles on clinically acquired computed tomography (CT) data. Briefly, the multi-atlas approach consists of determining a region of interest within each scan using affine registration, followed by nonrigid registration on reduced field of view atlases, and performing statistical fusion on the results. We evaluate the robustness of the approach by segmenting the ON structure in 501 clinically acquired CT scan volumes obtained from 183 subjects from a thyroid eye disease patient population. A subset of 30 scan volumes was manually labeled to assess accuracy and guide method choice. Of the 18 compared methods, the ANTS Symmetric Normalization registration and nonlocal spatial simultaneous truth and performance level estimation statistical fusion resulted in the best overall performance, resulting in a median Dice similarity coefficient of 0.77, which is comparable with inter-rater (human) reproducibility at 0.73.

Quantification of delineation errors of the gross tumor volume on magnetic resonance imaging in uterine cervical cancer using pathology data and deformation correction

Background. To safely optimize target volumes using magnetic resonance imaging (MRI) for uterine cervical cancer radiation therapy, MRI findings need to be validated. The aim of this study was to correlate pre-operatively acquired MRI and surgical specimen imaging for uterine cervical cancer patients using deformable image registration and quantify gross tumor volume (GTV) delineation discrepancies.Material and methods. For 16 retrospectively selected early-stage uterine cervical cancer patients, the cervix-uterus structure, uterine cavity and the GTV were delineated on 2D pathology photos after macroscopic intersection and corresponding pre-operatively acquired T2-weighted 2D sagittal MR images. Segmentations of pathology photos and MR images were simultaneously registered using a three-step multi-image registration strategy. The registration outcome was evaluated by the Dice similarity coefficient (DSC) and the surface distance error (SDE). In addition, GTV expansions within the cervix-uterus structure needed to obtain 95% GTV coverage were determined.Results. After three-step multi-image registration, the median DSC and median SDE were 0.98 and 0.4 mm (cervix-uterus) and 0.90 and 0.4 mm (uterine cavity), respectively. The average SDE around the GTV was 0.7 mm (range, 0.1 mm – 2.6 mm). An underestimation of MRI-based GTV delineations was found when no margin was applied, indicated by a mean GTV coverage of 61%. To obtain 95% GTV coverage for 90% of the patients, a minimum 12.0 mm margin around MRI-based GTVs was needed.Conclusion. The presented three-step multi-image registration strategy was suitable and accurate to correlate MRI and pathology data for uterine cervical cancer patients. To cover the pathology-based GTV, a margin of at least 12.0 mm around GTV delineations on T2-weighted MRI is needed.