Sylvia Rueda

Evaluation and Comparison of Current Fetal Ultrasound Image Segmentation Methods for Biometric Measurements: A Grand Challenge

This paper presents the evaluation results of the methods submitted to Challenge US: Biometric Measurements from Fetal Ultrasound Images, a segmentation challenge held at the IEEE International Symposium on Biomedical Imaging 2012. The challenge was set to compare and evaluate current fetal ultrasound image segmentation methods. It consisted of automatically segmenting fetal anatomical structures to measure standard obstetric biometric parameters, from 2D fetal ultrasound images taken on fetuses at different gestational ages (21 weeks, 28 weeks, and 33 weeks) and with varying image quality to reflect data encountered in real clinical environments. Four independent sub-challenges were proposed, according to the objects of interest measured in clinical practice: abdomen, head, femur, and whole fetus. Five teams participated in the head sub-challenge and two teams in the femur sub-challenge, including one team who tackled both. Nobody attempted the abdomen and whole fetus sub-challenges. The challenge goals were two-fold and the participants were asked to submit the segmentation results as well as the measurements derived from the segmented objects. Extensive quantitative (region-based, distance-based, and Bland-Altman measurements) and qualitative evaluation was performed to compare the results from a representative selection of current methods submitted to the challenge. Several experts (three for the head sub-challenge and two for the femur sub-challenge), with different degrees of expertise, manually delineated the objects of interest to define the ground truth used within the evaluation framework. For the head sub-challenge, several groups produced results that could be potentially used in clinical settings, with comparable performance to manual delineations. The femur sub-challenge had inferior performance to the head sub-challenge due to the fact that it is a harder segmentation problem and that the techniques presented relied more on the femurs appearance.

Automatic segmentation of jaw tissues in CT using active appearance models and semi-automatic landmarking

Preoperative planning systems are commonly used for oral implant surgery. One of the objectives is to determine if the quantity and quality of bone is sufficient to sustain an implant while avoiding critical anatomic structures. We aim to automate the segmentation of jaw tissues on CT images: cortical bone, trabecular core and especially the mandibular canal containing the dental nerve. This nerve must be avoided during implant surgery to prevent lip numbness. Previous work in this field used thresholds or filters and needed manual initialization. An automated system based on the use of Active Appearance Models (AAMs) is proposed. Our contribution is a completely automated segmentation of tissues and a semi-automatic landmarking process necessary to create the AAM model. The AAM is trained using 215 images and tested with a leave-4-out scheme. Results obtained show an initialization error of 3.25% and a mean error of 1.63mm for the cortical bone, 2.90 mm for the trabecular core, 4.76 mm for the mandibular canal and 3.40 mm for the dental nerve.

An approach for the automatic cephalometric landmark detection using mathematical morphology and active appearance models

Cephalometric analysis of lateral radiographs of the head is an important diagnosis tool in orthodontics. Based on manually locating specific landmarks, it is a tedious, time-consuming and error prone task. In this paper, we propose an automated system based on the use of Active Appearance Models (AAMs). Special attention has been paid to clinical validation of our method since previous work in this field used few images, was tested in the training set and/or did not take into account the variability of the images. In this research, a top-hat transformation was used to correct the intensity inhomogeneity of the radiographs generating a consistent training set that overcomes the above described drawbacks. The AAM was trained using 96 hand-annotated images and tested with a leave-one-out scheme obtaining an average accuracy of 2.48mm. Results show that AAM combined with mathematical morphology is the suitable method for clinical cephalometric applications.

Local curvature scale : a new concept of shape description

Shape description plays a fundamental role in computer vision and pattern recognition, especially in the fields of shape analysis, image segmentation, and registration. Shape representations must be unique, complete and should be able to reflect the differences between similar objects while abstracting from detail and keeping the basic features. Although many methods for shape description exist, they are usually application dependent. The proposed method of boundary shape description is based on the notion of curvature-scale, which is a new local scale concept, defined at each boundary element. From this representation, we can extract special points of interest such as convex and concave corners, straight lines, circular segments, and inflection points. This method is different from existing methods of curvature estimation and can be directly applied to digital boundaries without requiring prior approximation of the boundary. The results show that it produces a complete boundary shape description capable of handling different levels of shape detail. It also has numerous potential applications such as automatic landmark tagging which becomes necessary to build model-based approaches toward the goal of organ modeling and segmentation. The method is applicable to spaces of any dimensionality, although we have focused in this paper on 2D shapes.

Shape modeling via local curvature scale

Segmentation and modeling of organs using model-based approaches require a priori information which is often given by manually tagging landmarks on a training set of shapes. This is a tedious, time-consuming, and error prone task. To overcome some of these drawbacks, focusing on 2D shapes, we devised an automatic method based on the notion of curvature scale - a new local scale concept. This shape descriptor is used to automatically locate mathematical landmarks on the mean of the shapes in the training set, which are then propagated to the training shapes. Altogether 12 different strategies are described and are evaluated in different combinations in terms of compactness on two data sets - 40 CT images of the liver and 40 MR images of the talus bone of the foot. The results show that, for the same number of landmarks, the proposed methods are more compact than manual and equally spaced annotations.

Local Scale for Boundary Shape Description: Application in Locating Landmarks Automatically

A good shape descriptor is necessary for automatically identifying landmarks on boundaries. Our method of boundary shape description is based on the notion of c- scale, which is a new local scale concept, defined at each boundary element. From this representation we can extract special points of interest such as convex and concave corners, straight lines, circular segments, and inflection points. The results show that this method gives a complete description of shape and allows the automatic positioning of mathematical landmarks, which agree with our intuitive ideas of where landmarks may be defined. This method is applicable to spaces of any dimensionality, although we have focused in this paper on 2D shapes.

Plane Localization in 3-D Fetal Neurosonography for Longitudinal Analysis of the Developing Brain

The parasagittal (PS) plane is a 2-D diagnostic plane used routinely in cranial ultrasonography of the neonatal brain. This paper develops a novel approach to find the PS plane in a 3-D fetal ultrasound scan to allow image-based biomarkers to be tracked from prebirth through the first weeks of postbirth life. We propose an accurate plane-finding solution based on regression forests (RF). The method initially localizes the fetal brain and its midline automatically. The midline on several axial slices is used to detect the midsagittal plane, which is used as a constraint in the proposed RF framework to detect the PS plane. The proposed learning algorithm guides the RF learning method in a novel way by: 1) using informative voxels and voxel informative strength as a weighting within the training stage objective function, and 2) introducing regularization of the RF by proposing a geometrical feature within the training stage. Results on clinical data indicate that the new automated method is more reproducible than manual plane finding obtained by two clinicians.

Feature-based fuzzy connectedness segmentation of ultrasound images with an object completion step.

Highligts• Novel US segmentation approach based on the fuzzy connectedness framework.• Use of local phase and feature asymmetry to define affinity function.• Shape-based object completion step to detect and complete one or more gaps.• Novel regional entropy-based quantitative image quality assessment approach.• Method performs well across a variety of image qualities from clinical practice.

A Constrained Regression Forests Solution to 3D Fetal Ultrasound Plane Localization for Longitudinal Analysis of Brain Growth and Maturation

This paper develops a novel approach to find the plane in a 3D fetal ultrasound scan which corresponds to the 2D diagnostic plane used in cranial ultrasound of a neonate to allow image-based biomarkers to be tracked from pre-birth through the first weeks of post-birth life. We propose a method based on regression forests (RF) with important algorithm design considerations taken into account to provide an accurate plane-finding solution. Specifically, the new method constrains the RF method by 1) using informative voxels and voxel informative strength as a weighting within the training stage objective function u, and 2) introducing regularization of the RF by proposing a geometrical feature within the training stage. Results on clinical data indicate that the new automated method is more reproducible than manual plane finding.

Difference of Gaussians revolved along elliptical paths for ultrasound fetal head segmentation.

We present a fully automatic method to segment the skull from 2-D ultrasound images of the fetal head and to compute the standard biometric measurements derived from the segmented images. The method is based on the minimization of a novel cost function. The cost function is formulated assuming that the fetal skull has an approximately elliptical shape in the image and that pixel values within the skull are on average higher than in surrounding tissues. The main idea is to construct a template image of the fetal skull parametrized by the ellipse parameters and the calvarial thickness. The cost function evaluates the match between the template image and the observed ultrasound image. The optimum solution that minimizes the cost is found by using a global multiscale, multistart Nelder-Mead algorithm. The method was qualitatively and quantitatively evaluated using 90 ultrasound images from a recent segmentation grand challenge. These images have been manually analyzed by three independent experts. The segmentation accuracy of the automatic method was similar to the inter-expert segmentation variability. The automatically derived biometric measurements were as accurate as the manual measurements. Moreover, the segmentation accuracy of the presented method was superior to the accuracy of the other automatic methods that have previously been evaluated using the same data.