Ground Truth Manual Segmentation Research Articles

Automated Segmentation of Optical Coherence Tomography Angiography Images: Benchmark Data and Clinically Relevant Metrics.

PurposeTo generate the first open dataset of retinal parafoveal optical coherence tomography angiography (OCTA) images with associated ground truth manual segmentations, and to establish a standard for OCTA image segmentation by surveying a broad range of state-of-the-art vessel enhancement and binarization procedures.MethodsHandcrafted filters and neural network architectures were used to perform vessel enhancement. Thresholding methods and machine learning approaches were applied to obtain the final binarization. Evaluation was performed by using pixelwise metrics and newly proposed topological metrics. Finally, we compare the error in the computation of clinically relevant vascular network metrics (e.g., foveal avascular zone area and vessel density) across segmentation methods.ResultsOur results show that, for the set of images considered, deep learning architectures (U-Net and CS-Net) achieve the best performance (Dice = 0.89). For applications where manually segmented data are not available to retrain these approaches, our findings suggest that optimally oriented flux (OOF) is the best handcrafted filter (Dice = 0.86). Moreover, our results show up to 25% differences in vessel density accuracy depending on the segmentation method used.ConclusionsIn this study, we derive and validate the first open dataset of retinal parafoveal OCTA images with associated ground truth manual segmentations. Our findings should be taken into account when comparing the results of clinical studies and performing meta-analyses. Finally, we release our data and source code to support standardization efforts in OCTA image segmentation.Translational RelevanceThis work establishes a standard for OCTA retinal image segmentation and introduces the importance of evaluating segmentation performance in terms of clinically relevant metrics.

Fully automated left atrium segmentation from anatomical cine long-axis MRI sequences using deep convolutional neural network with unscented Kalman filter.

This study proposes a fully automated approach for the left atrial segmentation from routine cine long-axis cardiac magnetic resonance image sequences using deep convolutional neural networks and Bayesian filtering. The proposed approach consists of a classification network that automatically detects the type of long-axis sequence and three different convolutional neural network models followed by unscented Kalman filtering (UKF) that delineates the left atrium. Instead of training and predicting all long-axis sequence types together, the proposed approach first identifies the image sequence type as to 2, 3 and 4 chamber views, and then performs prediction based on neural nets trained for that particular sequence type. The datasets were acquired retrospectively and ground truth manual segmentation was provided by an expert radiologist. In addition to neural net based classification and segmentation, another neural net is trained and utilized to select image sequences for further processing using UKF to impose temporal consistency over cardiac cycle. A cyclic dynamic model with time-varying angular frequency is introduced in UKF to characterize the variations in cardiac motion during image scanning. The proposed approach was trained and evaluated separately with varying amount of training data with images acquired from 20, 40, 60 and 80 patients. Evaluations over 1515 images with equal number of images from each chamber group acquired from an additional 20 patients demonstrated that the proposed model outperformed state-of-the-art and yielded a mean Dice coefficient value of 94.1%, 93.7% and 90.1% for 2, 3 and 4-chamber sequences, respectively, when trained with datasets from 80 patients.

Convolutional neural network-based automatic liver delineation on contrast-enhanced and non-contrast-enhanced CT images for radiotherapy planning.

This study evaluated a convolutional neural network (CNN) for automatically delineating the liver on contrast-enhanced or non-contrast-enhanced CT, making comparisons with a commercial automated technique (MIM Maestro®). Intensity-modulated radiation therapy requires careful labor-intensive planning involving delineation of the target and organs on CT or MR images to ensure delivery of the effective dose to the target while avoiding organs at risk. Contrast-enhanced planning CT images from 101 pancreatic cancer cases and accompanying mask images showing manually-delineated liver contours were used to train the CNN to segment the liver. The trained CNN then performed liver segmentation on a further 20 contrast-enhanced and 15 non-contrastenhanced CT image sets, producing three-dimensional mask images of the liver. For both contrast-enhanced and non-contrast-enhanced images, the mean Dice similarity coefficients between CNN segmentations and ground-truth manual segmentations were significantly higher than those between ground-truth and MIM Maestro software (p < 0.001). Although mean CT values of the liver were higher on contrast-enhanced than on non-contrast-enhanced CT, there were no significant differences in the Hausdorff distances of the CNN segmentations, indicating that the CNN could successfully segment the liver on both image types, despite being trained only on contrast-enhanced images. Our results suggest that a CNN can perform highly accurate automated delineation of the liver on CT images, irrespective of whether the CT images are contrast-enhanced or not.

Automatic Segmentation of Retinal Capillaries in Adaptive Optics Scanning Laser Ophthalmoscope Perfusion Images Using a Convolutional Neural Network.

PurposeAdaptive optics scanning laser ophthalmoscope (AOSLO) capillary perfusion images can possess large variations in contrast, intensity, and background signal, thereby limiting the use of global or adaptive thresholding techniques for automatic segmentation. We sought to develop an automated approach to segment perfused capillaries in AOSLO images.Methods12,979 image patches were extracted from manually segmented AOSLO montages from 14 eyes and used to train a convolutional neural network (CNN) that classified pixels as capillaries, large vessels, background, or image canvas. 1764 patches were extracted from AOSLO montages of four separate subjects, and were segmented manually by two raters (ground truth) and automatically by the CNN, an Otsu's approach, and a Frangi approach. A modified Dice coefficient was created to account for slight spatial differences between the same manually and CNN-segmented capillaries.ResultsCNN capillary segmentation had an accuracy (0.94), a Dice coefficient (0.67), and a modified Dice coefficient (0.90) that were significantly higher than other automated approaches (P < 0.05). There were no significant differences in capillary density and mean segment length between manual ground-truth and CNN segmentations (P > 0.05).ConclusionsClose agreement between the CNN and manual segmentations enables robust and objective quantification of perfused capillary metrics. The developed CNN is time and computationally efficient, and distinguishes capillaries from areas containing diffuse background signal and larger underlying vessels.Translational RelevanceThis automatic segmentation algorithm greatly increases the efficiency of quantifying AOSLO capillary perfusion images.

A deep learning algorithm may automate intracranial aneurysm detection on MR angiography with high diagnostic performance.

To develop a deep learning algorithm for automated detection and localization of intracranial aneurysms on time-of-flight MR angiography and evaluate its diagnostic performance. In a retrospective and multicenter study, MR images with aneurysms based on radiological reports were extracted. The examinations were randomly divided into two data sets: training set of 468 examinations and internal test set of 120 examinations. Additionally, 50 examinations without aneurysms were randomly selected and added to the internal test set. External test data set consisted of 56 examinations with intracranial aneurysms and 50 examinations without aneurysms, which were extracted based on radiological reports from a different institution. After manual ground truth segmentation of aneurysms, a deep learning algorithm based on 3D ResNet architecture was established with the training set. Its sensitivity, positive predictive value, and specificity were evaluated in the internal and external test sets. MR images included 551 aneurysms (mean diameter, 4.17 ± 2.49 mm) in the training, 147 aneurysms (mean diameter, 3.98 ± 2.11 mm) in the internal test, 63 aneurysms (mean diameter, 3.23 ± 1.69 mm) in the external test sets. The sensitivity, the positive predictive value, and the specificity were 87.1%, 92.8%, and 92.0% for the internal test set and 85.7%, 91.5%, and 98.0% for the external test set, respectively. A deep learning algorithm detected intracranial aneurysms with a high diagnostic performance which was validated using external data set. • A deep learning-based algorithm for the automated diagnosis of intracranial aneurysms demonstrated a high sensitivity, positive predictive value, and specificity. • The high diagnostic performance of the algorithm was validated using external test data set from a different institution with a different scanner. • The algorithm might be robust and effective for general use in real clinical settings.

Deep learning-based fully automatic segmentation of wrist cartilage in MR images.

The study objective was to investigate the performance of a dedicated convolutional neural network (CNN) optimized for wrist cartilage segmentation from 2D MR images. CNN utilized a planar architecture and patch-based (PB) training approach that ensured optimal performance in the presence of a limited amount of training data. The CNN was trained and validated in 20 multi-slice MRI datasets acquired with two different coils in 11 subjects (healthy volunteers and patients). The validation included a comparison with the alternative state-of-the-art CNN methods for the segmentation of joints from MR images and the ground-truth manual segmentation. When trained on the limited training data, the CNN outperformed significantly image-based and PB-U-Net networks. Our PB-CNN also demonstrated a good agreement with manual segmentation (Sørensen-Dice similarity coefficient [DSC] = 0.81) in the representative (central coronal) slices with a large amount of cartilage tissue. Reduced performance of the network for slices with a very limited amount of cartilage tissue suggests the need for fully 3D convolutional networks to provide uniform performance across the joint. The study also assessed inter- and intra-observer variability of the manual wrist cartilage segmentation (DSC = 0.78-0.88 and 0.9, respectively). The proposed deep learning-based segmentation of the wrist cartilage from MRI could facilitate research of novel imaging markers of wrist osteoarthritis to characterize its progression and response to therapy.

Computational techniques to segment and classify lumbar compression fractures.

Vertebral fractures are important indicators of osteoporosis. Fractures with partial collapse of vertebral bodies are referred to as vertebral compression fractures (VCFs) that are usually non-traumatic in nature. Some common causes of VCFs are trauma, bone failure related to osteoporosis (benign) and metastatic cancer (malignant). This paper aims at developing a system for computer-aided diagnosis to help in the detection, labeling and segmentation of lumbar vertebral body (VB) and to further classify each VB into normal, malignant and benign VCFs. After the initial preprocessing, morphological, shape and angular features are used in the detection, labeling and segmentation steps. Various shape and statistical texture features are extracted from the segmented VB and are fed to the classifier for the final decision. The segmentation and classification results obtained were compared with the ground truth manual segmentation of the lumbar VB and the decision labels of the fractures provided by the experts. The dice similarity coefficient (DSC) for segmentation reached up to 94.27%, and the classification results show that shape and texture features together are able to correctly classify with an accuracy rate of 95.34%. The final outcomes are expected to be useful in the analysis of vertebral compression fractures.

Segmentation of head-and-neck organs-at-risk in longitudinal CT scans combining deformable registrations and convolutional neural networks

ABSTRACT Automated segmentation of organs-at-risk (OAR) in follow-up images of the patient acquired during the course of treatment could greatly facilitate adaptive treatment planning in radiotherapy. Instead of segmenting each image separately, the segmentation could be improved by making use of the additional information provided by longitudinal data of previously segmented images of the same patient. We propose a tool for automated segmentation of longitudinal data that combines deformable image registration (DIR) and convolutional neural networks (CNN). The segmentation propagated by DIR from a previous image onto the current image and the segmentation obtained by a separately trained cross-sectional CNN applied to the current image are given as input to a longitudinal CNN, together with the images itself, that is trained to optimally predict the manual ground truth segmentation using all available information. Despite the fairly limited amount of training data available in this study, an improvement of the segmentation of five different OAR in head-and-neck CT scans was found compared to both the results of DIR and the cross-sectional CNN separately in terms of Dice coefficient.

Stroke Lesion Segmentation in FLAIR MRI Datasets Using Customized Markov Random Fields.

Robust and reliable stroke lesion segmentation is a crucial step toward employing lesion volume as an independent endpoint for randomized trials. The aim of this work was to develop and evaluate a novel method to segment sub-acute ischemic stroke lesions from fluid-attenuated inversion recovery (FLAIR) magnetic resonance imaging (MRI) datasets. After preprocessing of the datasets, a Bayesian technique based on Gabor textures extracted from the FLAIR signal intensities is utilized to generate a first estimate of the lesion segmentation. Using this initial segmentation, a customized voxel-level Markov random field model based on intensity as well as Gabor texture features is employed to refine the stroke lesion segmentation. The proposed method was developed and evaluated based on 151 multi-center datasets from three different databases using a leave-one-patient-out validation approach. The comparison of the automatically segmented stroke lesions with manual ground truth segmentation revealed an average Dice coefficient of 0.582, which is in the upper range of previously presented lesion segmentation methods using multi-modal MRI datasets. Furthermore, the results obtained by the proposed technique are superior compared to the results obtained by two methods based on convolutional neural networks and three phase level-sets, respectively, which performed best in the ISLES 2015 challenge using multi-modal imaging datasets. The results of the quantitative evaluation suggest that the proposed method leads to robust lesion segmentation results using FLAIR MRI datasets only as a follow-up sequence.

Automated quantification and evaluation of motion artifact on coronary CT angiography images.

This study developed and validated a Motion Artifact Quantification algorithm to automatically quantify the severity of motion artifacts on coronary computed tomography angiography (CCTA) images. The algorithm was then used to develop a Motion IQ Decision method to automatically identify whether a CCTA dataset is of sufficient diagnostic image quality or requires further correction. The developed Motion Artifact Quantification algorithm includes steps to identify the right coronary artery (RCA) regions of interest (ROIs), segment vessel and shading artifacts, and to calculate the motion artifact score (MAS) metric. The segmentation algorithms were verified against ground-truth manual segmentations. The segmentation algorithms were also verified by comparing and analyzing the MAS calculated from ground-truth segmentations and the algorithm-generated segmentations. The Motion IQ Decision algorithm first identifies slices with unsatisfactory image quality using a MAS threshold. The algorithm then uses an artifact-length threshold to determine whether the degraded vessel segment is large enough to cause the dataset to be nondiagnostic. An observer study on 30 clinical CCTA datasets was performed to obtain the ground-truth decisions of whether the datasets were of sufficient image quality. A five-fold cross-validation was used to identify the thresholds and to evaluate the Motion IQ Decision algorithm. The automated segmentation algorithms in the Motion Artifact Quantification algorithm resulted in Dice coefficients of 0.84 for the segmented vessel regions and 0.75 for the segmented shading artifact regions. The MAS calculated using the automated algorithm was within 10% of the values obtained using ground-truth segmentations. The MAS threshold and artifact-length thresholds were determined by the ROC analysis to be 0.6 and 6.25mm by all folds. The Motion IQ Decision algorithm demonstrated 100% sensitivity, 66.7%±27.9% specificity, and a total accuracy of 86.7%±12.5% for identifying datasets in which the RCA required correction. The Motion IQ Decision algorithm demonstrated 91.3% sensitivity, 71.4% specificity, and a total accuracy of 86.7% for identifying CCTA datasets that need correction for any of the three main vessels. The Motion Artifact Quantification algorithm calculated accurate (<10% error) motion artifact scores using the automated segmentation methods. The developed algorithms demonstrated high sensitivity (91.3%) and specificity (71.4%) in identifying datasets of insufficient image quality. The developed algorithms for automatically quantifying motion artifact severity may be useful for comparing acquisition techniques, improving best-phase selection algorithms, and evaluating motion compensation techniques.

Fully automatic segmentation and objective assessment of atrial scars for long-standing persistent atrial fibrillation patients using late gadolinium-enhanced MRI.

PurposeAtrial fibrillation (AF) is the most common heart rhythm disorder and causes considerable morbidity and mortality, resulting in a large public health burden that is increasing as the population ages. It is associated with atrial fibrosis, the amount and distribution of which can be used to stratify patients and to guide subsequent electrophysiology ablation treatment. Atrial fibrosis may be assessed noninvasively using late gadolinium‐enhanced (LGE) magnetic resonance imaging (MRI) where scar tissue is visualized as a region of signal enhancement. However, manual segmentation of the heart chambers and of the atrial scar tissue is time consuming and subject to interoperator variability, particularly as image quality in AF is often poor. In this study, we propose a novel fully automatic pipeline to achieve accurate and objective segmentation of the heart (from MRI Roadmap data) and of scar tissue within the heart (from LGE MRI data) acquired in patients with AF.MethodsOur fully automatic pipeline uniquely combines: (a) a multiatlas‐based whole heart segmentation (MA‐WHS) to determine the cardiac anatomy from an MRI Roadmap acquisition which is then mapped to LGE MRI, and (b) a super‐pixel and supervised learning based approach to delineate the distribution and extent of atrial scarring in LGE MRI. We compared the accuracy of the automatic analysis to manual ground truth segmentations in 37 patients with persistent long‐standing AF.ResultsBoth our MA‐WHS and atrial scarring segmentations showed accurate delineations of cardiac anatomy (mean Dice = 89%) and atrial scarring (mean Dice = 79%), respectively, compared to the established ground truth from manual segmentation. In addition, compared to the ground truth, we obtained 88% segmentation accuracy, with 90% sensitivity and 79% specificity. Receiver operating characteristic analysis achieved an average area under the curve of 0.91.ConclusionCompared with previously studied methods with manual interventions, our innovative pipeline demonstrated comparable results, but was computed fully automatically. The proposed segmentation methods allow LGE MRI to be used as an objective assessment tool for localization, visualization, and quantitation of atrial scarring and to guide ablation treatment.

Open Source Software Implementation of Anatomical Segmentation

Purpose: Intraventricular hemorrhage (IVH) affects nearly 15% of preterm infants. It can lead to ventricular dilation and cognitive impairment. To ablate IVH clots, MR-guided focused ultrasound surgery (MRgFUS) is investigated. This procedure requires accurate, fast and consistent quantification of ventricle and clot volumes.&#x0D; Methods: We developed a semi-autonomous segmentation (SAS) algorithm for measuring changes in the ventricle and clot volumes. Images are normalized, and then ventricle and clot masks are registered to the images. Voxels of the registered masks and voxels obtained by thresholding the normalized images are used as seed points for competitive region growing, which provides the final segmentation. The user selects the areas of interest for correspondence after thresholding and these selections are the final seeds for region growing. SAS was evaluated on an IVH porcine model.&#x0D; Results: SAS was compared to ground truth manual segmentation (MS) for accuracy, efficiency, and consistency. Accuracy was determined by comparing clot and ventricle volumes produced by SAS and MS. In Two-One-Sided Test, SAS and MS were found to be significantly equivalent (p &lt; 0.01). SAS on average was found to be 15 times faster than MS (p &lt; 0.01). Consistency was determined by repeated segmentation of the same image by both SAS and manual methods, SAS being significantly more consistent than MS (p &lt; 0.05). &#x0D; Conclusion: SAS is a viable method to quantify the IVH clot and the lateral brain ventricles and it is serving in a large- scale porcine study of MRgFUS treatment of IVH clot lysis.

Automatic Measurement of Fetal Brain Development from Magnetic Resonance Imaging: New Reference Data

Background: Accurate fetal brain volume estimation is of paramount importance in evaluating fetal development. The aim of this study was to develop an automatic method for fetal brain segmentation from magnetic resonance imaging (MRI) data, and to create for the first time a normal volumetric growth chart based on a large cohort. Subjects and Methods: A semi-automatic segmentation method based on Seeded Region Growing algorithm was developed and applied to MRI data of 199 typically developed fetuses between 18 and 37 weeks' gestation. The accuracy of the algorithm was tested against a sub-cohort of ground truth manual segmentations. A quadratic regression analysis was used to create normal growth charts. The sensitivity of the method to identify developmental disorders was demonstrated on 9 fetuses with intrauterine growth restriction (IUGR). Results: The developed method showed high correlation with manual segmentation (r² = 0.9183, p < 0.001) as well as mean volume and volume overlap differences of 4.77 and 18.13%, respectively. New reference data on 199 normal fetuses were created, and all 9 IUGR fetuses were at or below the third percentile of the normal growth chart. Discussion: The proposed method is fast, accurate, reproducible, user independent, applicable with retrospective data, and is suggested for use in routine clinical practice.

A Multi-Objective Decision Making Approach for Solving the Image Segmentation Fusion Problem

Image segmentation fusion is defined as the set of methods which aim at merging several image segmentations, in a manner that takes full advantage of the complementarity of each one. Previous relevant researches in this field have been impeded by the difficulty in identifying an appropriate single segmentation fusion criterion, providing the best possible, i.e., the more informative, result of fusion. In this paper, we propose a new model of image segmentation fusion based on multi-objective optimization which can mitigate this problem, to obtain a final improved result of segmentation. Our fusion framework incorporates the dominance concept in order to efficiently combine and optimize two complementary segmentation criteria, namely, the global consistency error and the F-measure (precision-recall) criterion. To this end, we present a hierarchical and efficient way to optimize the multi-objective consensus energy function related to this fusion model, which exploits a simple and deterministic iterative relaxation strategy combining the different image segments. This step is followed by a decision making task based on the so-called "technique for order performance by similarity to ideal solution". Results obtained on two publicly available databases with manual ground truth segmentations clearly show that our multi-objective energy-based model gives better results than the classical mono-objective one.

Interactive Outlining of Pancreatic Cancer Liver Metastases in Ultrasound Images

Ultrasound (US) is the most commonly used liver imaging modality worldwide. Due to its low cost, it is increasingly used in the follow-up of cancer patients with metastases localized in the liver. In this contribution, we present the results of an interactive segmentation approach for liver metastases in US acquisitions. A (semi-) automatic segmentation is still very challenging because of the low image quality and the low contrast between the metastasis and the surrounding liver tissue. Thus, the state of the art in clinical practice is still manual measurement and outlining of the metastases in the US images. We tackle the problem by providing an interactive segmentation approach providing real-time feedback of the segmentation results. The approach has been evaluated with typical US acquisitions from the clinical routine, and the datasets consisted of pancreatic cancer metastases. Even for difficult cases, satisfying segmentations results could be achieved because of the interactive real-time behavior of the approach. In total, 40 clinical images have been evaluated with our method by comparing the results against manual ground truth segmentations. This evaluation yielded to an average Dice Score of 85% and an average Hausdorff Distance of 13 pixels.

EFA-BMFM: A multi-criteria framework for the fusion of colour image segmentation

Considering the recent progress in the development of practical applications in the field of image processing, it is increasingly important to develop new, efficient and more reliable algorithms to solve an image segmentation problem. To this end, various fusion-based segmentation approaches which use consensus clustering, and which are based on the optimization of a single criterion, have been proposed. One of the greatest challenges with these approaches is to select the best fusion criterion, which gives the best performance for the image segmentation model. In this paper, we propose a new fusion model of image segmentation based on multi-objective optimization, which aims to overcome the limitation and bias caused by a single criterion, and to provide a final improved segmentation. To address the ill-posedness for the search of the best criterion, the proposed fusion model combines two conflicting and complementary criteria for segmentation fusion, namely, the region-based variation of information (VoI) criterion and the contour-based F-measure (precision-recall) criterion using an entropy-based confidence weighting factor. To optimize our energy-based model, we propose an extended local optimization procedure based on superpixels and derived from the iterative conditional mode (ICM) algorithm. This new multi-objective median partition-based approach, which relies on the fusion of inaccurate, quick and spatial clustering results, has emerged as an appealing alternative to the use of traditional segmentation fusion models which exist in the literature. We perform experiments using the Berkeley database with manual ground truth segmentations, and the results clearly show the feasibility and efficiency of the proposed methodology.

Fully automatic segmentation of arbitrarily shaped fiducial markers in cone-beam CT projections

Radio-opaque fiducial markers of different shapes are often implanted in or near abdominal or thoracic tumors to act as surrogates for the tumor position during radiotherapy. They can be used for real-time treatment adaptation, but this requires a robust, automatic segmentation method able to handle arbitrarily shaped markers in a rotational imaging geometry such as cone-beam computed tomography (CBCT) projection images and intra-treatment images. In this study, we propose a fully automatic dynamic programming (DP) assisted template-based (TB) segmentation method. Based on an initial DP segmentation, the DPTB algorithm generates and uses a 3D marker model to create 2D templates at any projection angle. The 2D templates are used to segment the marker position as the position with highest normalized cross-correlation in a search area centered at the DP segmented position. The accuracy of the DP algorithm and the new DPTB algorithm was quantified as the 2D segmentation error (pixels) compared to a manual ground truth segmentation for 97 markers in the projection images of CBCT scans of 40 patients. Also the fraction of wrong segmentations, defined as 2D errors larger than 5 pixels, was calculated. The mean 2D segmentation error of DP was reduced from 4.1 pixels to 3.0 pixels by DPTB, while the fraction of wrong segmentations was reduced from 17.4% to 6.8%. DPTB allowed rejection of uncertain segmentations as deemed by a low normalized cross-correlation coefficient and contrast-to-noise ratio. For a rejection rate of 9.97%, the sensitivity in detecting wrong segmentations was 67% and the specificity was 94%. The accepted segmentations had a mean segmentation error of 1.8 pixels and 2.5% wrong segmentations.

Automatic segmentation of lungs in SPECT images using active shape model trained by meshes delineated in CT images.

This paper presents a fully automated method for segmentation of 3D SPECT ventilation and perfusion images. It relies on statistical information on lung shape derived by CT manual segmentation and its main processing steps are: shape model extraction, binary segmentation, positioning of mean shape in SPECT images and iterative shape adaptation based on intensity profiles and on what is considered `plausible' lung shape. The Active Shape Model is used to generate accurate anatomic results in SPECT images with functional information and thus unclear borders, especially in the case of pathologies. The method was compared against ground truth manual segmentation on CT images, using volumetric, difference dice coefficient, sensitivity and precision.

A Novel Fusion Approach Based on the Global Consistency Criterion to Fusing Multiple Segmentations

In this paper, we introduce a new fusion model whose objective is to fuse multiple region-based segmentation maps to get a final better segmentation result. The suggested new fusion model is based on an energy function originated from the global consistency error (GCE), a perceptual measure which takes into account the inherent multiscale nature of an image segmentation by measuring the level of refinement existing between two spatial partitions. Combined with a region merging/splitting prior, this new energy-based fusion model of label fields allows to define an interesting penalized likelihood estimation procedure based on the GCE criterion with which the fusion of basic, rapidly-computed segmentation results appears as a relevant alternative compared with other (possibly complex) segmentation techniques proposed in the image segmentation field. The performance of our fusion model was evaluated on the Berkeley dataset including various segmentations given by humans (manual ground truth segmentations). The obtained results clearly demonstrate the efficiency of this fusion model.

A Precision-Recall Criterion Based Consensus Model For Fusing Multiple Segmentations

This paper presents a general framework for seamlessly combining multiple low cost and inaccurate estimated segmentation maps (with an arbitrary number of regions) of the same scene to achieve a final improved segmentation. The proposed fusion model is derived from the well-known precision-recall criterion, specially dedicated to the specific clustering problem of any spatially indexed data and which is also efficient and widely used in the vision community for evaluating both a region-based segmentation and the quality of contours produced by this segmentation map compared to one or multiple ground-truth segmentations of the same image. The proposed combination framework is here specifically designed to be robust with respect to outlier segmentations (that appear to be inconsistent with the remainder of the segmentation ensemble) and includes an explicit internal regularization factor reflecting the inherent ill-posed nature of the segmentation problem. We propose also a hierarchical and efficient way to optimize the consensus energy function related to this fusion model that exploits a simple and deterministic iterative relaxation strategy combining the different segments or individual regions belonging to the segmentation ensemble in the final solution. The experimental results on the Berkeley database with manual ground truth segmentations show the effectiveness of our combination model.