A model-based, semi-global segmentation approach for automatic 3-D point landmark localization in neuroimages

A functional pipeline framework for landmark identification on 3D surface extracted from volumetric data

Landmarks, also known as feature points, are one of the important geometry primitives that describe the predominant characteristics of a surface. In this study we proposed a self-contained framework to generate landmarks on surfaces extracted from volumetric data. The framework is designed to be a three-fold pipeline structure. The pipeline comprises three phases which are surface construction, crest line extraction and landmark identification. With input as a volumetric data and output as landmarks, the pipeline takes in 3D raw data and produces a 0D geometry feature. In each phase we investigate existing methods, extend and tailor the methods to fit the pipeline design. The pipeline is designed to be functional as it is modularised to have a dedicated function in each phase. We extended the implicit surface polygonizer for surface construction in first phase, developed an alternative way to compute the gradient of maximal curvature for crest line extraction in second phase and finally we combine curvature information and K-means clustering method to identify the landmarks in the third phase. The implementations are firstly carried on a controlled environment, i.e. synthetic data, for proof of concept. Then the method is tested on a small scale data set and subsequently on huge data set. Issues and justifications are addressed accordingly for each phase.

Does 3D Phenotyping Yield Substantial Insights in the Genetics of the Mouse Mandible Shape?

We describe the application of high-resolution 3D microcomputed tomography, together with 3D landmarks and geometric morphometrics, to validate and further improve previous quantitative genetic studies that reported QTL responsible for variation in the mandible shape of laboratory mice using a new backcross between C57BL/6J and A/J inbred strains. Despite the increasing availability of 3D imaging techniques, artificial flattening of the mandible by 2D imaging techniques seems at first an acceptable compromise for large-scale phenotyping protocols, thanks to an abundance of low-cost digital imaging systems such as microscopes or digital cameras. We evaluated the gain of information from considering explicitly this additional third dimension, and also from capturing variation on the bone surface where no precise anatomical landmark can be marked. Multivariate QTL mapping conducted with different landmark configurations (2D vs. 3D; manual vs. semilandmarks) broadly agreed with the findings of previous studies. Significantly more QTL (23) were identified and more precisely mapped when the mandible shape was captured with a large set of semilandmarks coupled with manual landmarks. It appears that finer phenotypic characterization of the mandibular shape with 3D landmarks, along with higher density genotyping, yields better insights into the genetic architecture of mandibular development. Most of the main variation is, nonetheless, preferentially embedded in the natural 2D plane of the hemi-mandible, reinforcing the results of earlier influential investigations.

Automated Detection of 3D Landmarks for the Elimination of Non-Biological Variation in Geometric Morphometric Analyses

Landmark-based morphometric analyses are used by anthropologists, developmental and evolutionary biologists to understand shape and size differences (eg. in the cranioskeleton) between groups of specimens. The standard, labor intensive approach is for researchers to manually place landmarks on 3D image datasets. As landmark recognition is subject to inaccuracies of human perception, digitization of landmark coordinates is typically repeated (often by more than one person) and the mean coordinates are used. In an attempt to improve efficiency and reproducibility between researchers, we have developed an algorithm to locate landmarks on CT mouse hemi-mandible data. The method is evaluated on 3D meshes of 28-day old mice, and results compared to landmarks manually identified by experts. Quantitative shape comparison between two inbred mouse strains demonstrate that data obtained using our algorithm also has enhanced statistical power when compared to data obtained by manual landmarking.

Parallelized seeded region growing using CUDA

This paper presents a novel method for parallelizing the seeded region growing (SRG) algorithm using Compute Unified Device Architecture (CUDA) technology, with intention to overcome the theoretical weakness of SRG algorithm of its computation time being directly proportional to the size of a segmented region. The segmentation performance of the proposed CUDA-based SRG is compared with SRG implementations on single-core CPUs, quad-core CPUs, and shader language programming, using synthetic datasets and 20 body CT scans. Based on the experimental results, the CUDA-based SRG outperforms the other three implementations, advocating that it can substantially assist the segmentation during massive CT screening tests.

Tumor burden analysis on computed tomography by automated liver and tumor segmentation

The paper presents the automated computation of hepatic tumor burden from abdominal computed tomography (CT) images of diseased populations with images with inconsistent enhancement. The automated segmentation of livers is addressed first. A novel 3-D affine invariant shape parameterization is employed to compare local shape across organs. By generating a regular sampling of the organ's surface, this parameterization can be effectively used to compare features of a set of closed 3-D surfaces point-to-point, while avoiding common problems with the parameterization of concave surfaces. From an initial segmentation of the livers, the areas of atypical local shape are determined using training sets. A geodesic active contour corrects locally the segmentations of the livers in abnormal images. Graph cuts segment the hepatic tumors using shape and enhancement constraints. Liver segmentation errors are reduced significantly and all tumors are detected. Finally, support vector machines and feature selection are employed to reduce the number of false tumor detections. The tumor detection true position fraction of 100% is achieved at 2.3 false positives/case and the tumor burden is estimated with 0.9% error. Results from the test data demonstrate the method's robustness to analyze livers from difficult clinical cases to allow the temporal monitoring of patients with hepatic cancer.

Group average difference: a termination criterion for active contour

This paper presents a termination criterion for active contour that does not involve alteration of the energy functional. The criterion is based on the area difference of the contour during evolution. In this criterion, the evolution of the contour terminates when the area difference fluctuates around a constant. The termination criterion is tested using parametric gradient vector flow active contour with contour resampling and normal force selection. The usefulness of the criterion is shown through its trend, speed, accuracy, shape insensitivity, and insensitivity to contour resampling. The metric used in the proposed criterion demonstrated a steadily decreasing trend. For automatic implementation in which different shapes need to be segmented, the proposed criterion demonstrated almost 50% and 60% total time reduction while achieving similar accuracy as compared with the pixel movement-based method in the segmentation of synthetic and real medical images, respectively. Our results also show that the proposed termination criterion is insensitive to shape variation and contour resampling. The criterion also possesses potential to be used for other kinds of snakes.

Detection of point landmarks in 3D medical images via phase congruency model

This paper presents a novel technique for detection of point landmarks in volumetric medical images based on a three-dimensional (3D) Phase Congruency (PC) model. A bank of 3D log-Gabor filters is specially designed in the frequency domain and used to compute 3D energy maps, which are further combined to form the phase congruency measure. The PC measure is invariant to intensity variations and contrast resolution and provides a good indication of feature significance in an image. To detect significant 3D point landmarks, eigen-analysis of a 3×3 matrix of second-order PC moments, computed for each point in the image, is performed followed by local maxima detection. Two different application scenarios in radiation therapy planning of the head and neck anatomy are used to illustrate the feasibility and usefulness of the proposed method.

Detecting corpus callosum abnormalities in autism based on anatomical landmarks

Autism is a severe developmental disorder whose neurological basis is largely unknown. The aim of this study was to identify the shape differences of the corpus callosum between patients with autism and control subjects. Anatomical landmarks were collected from midsagittal magnetic resonance images of 25 patients and 18 controls. Euclidean distance matrix analysis and thin-plate spline analyses were used to examine the landmark forms. Point-by-point shape comparison was performed both globally and locally. A new local shape comparison scheme was proposed which compared each part of the shape in its local coordinate system. Point correspondence was established among individual shapes based on the inherent landmark correspondence. No significant difference was found in the landmark form between patients and controls, but the distance between the interior genu and the posterior-most section was found to be significantly shorter in patients. Thin-plate spline analysis showed significant group differences between the landmark configurations in terms of the deformation from the overall mean configuration. Significant global shape differences were found in the anterior lower body and posterior bottom, and there was a local shape difference in the anterior bottom. This study can serve as both a clinical reference and a detailed procedural guideline for similar studies in the future.

Automatic model-guided segmentation of the human brain ventricular system from CT images

RATIONALE AND OBJECTIVES

Accurate segmentation of the brain ventricular system on computed tomographic (CT) imaging is useful in neurodiagnosis and neurosurgery. Manual segmentation is time consuming, usually not reproducible, and subjective. Because of image noise, low contrast between soft tissues, large interslice distance, large shape, and size variations of the ventricular system, no automatic method is presently available. The authors propose a model-guided method for the automated segmentation of the ventricular system.

MATERIALS AND METHODS

Fifty CT scans of patients with strokes at different sites were collected for this study. Given a brain CT image, its ventricular system was segmented in five steps: (1) a predefined volumetric model was registered (or deformed) onto the image; (2) according to the deformed model, eight regions of interest were automatically specified; (3) the intensity threshold of cerebrospinal fluid was calculated in a region of interest and used to segment all regions of cerebrospinal fluid from the entire brain volume; (4) each ventricle was segmented in its specified region of interest; and (5) intraventricular calcification regions were identified to refine the ventricular segmentation.

RESULTS

Compared to ground truths provided by experts, the segmentation results of this method achieved an average overlap ratio of 85% for the entire ventricular system. On a desktop personal computer with a dual-core central processing unit running at 2.13 GHz, about 10 seconds were required to analyze each data set.

CONCLUSION

Experiments with clinical CT images showed that the proposed method can generate acceptable results in the presence of image noise, large shape, and size variations of the ventricular system, and therefore it is potentially useful for the quantitative interpretation of CT images in neurodiagnosis and neurosurgery.

Automatic identification of the reference system based on the fourth ventricular landmarks in T1-weighted MR images

RATIONALE AND OBJECTIVES

The reference system based on the fourth ventricular landmarks (including the fastigial point and ventricular floor plane) is used in medical image analysis of the brain stem. The objective of this study was to develop a rapid, robust, and accurate method for the automatic identification of this reference system on T1-weighted magnetic resonance images.

MATERIALS AND METHODS

The fully automated method developed in this study consisted of four stages: preprocessing of the data set, expectation-maximization algorithm-based extraction of the fourth ventricle in the region of interest, a coarse-to-fine strategy for identifying the fastigial point, and localization of the base point. The method was evaluated on 27 Brain Web data sets qualitatively and 18 Internet Brain Segmentation Repository data sets and 30 clinical scans quantitatively.

RESULTS

The results of qualitative evaluation indicated that the method was robust to rotation, landmark variation, noise, and inhomogeneity. The results of quantitative evaluation indicated that the method was able to identify the reference system with an accuracy of 0.7 +/- 0.2 mm for the fastigial point and 1.1 +/- 0.3 mm for the base point. It took <6 seconds for the method to identify the related landmarks on a personal computer with an Intel Core 2 6300 processor and 2 GB of random-access memory.

CONCLUSION

The proposed method for the automatic identification of the reference system based on the fourth ventricular landmarks was shown to be rapid, robust, and accurate. The method has potentially utility in image registration and computer-aided surgery.

Automatic segmentation of the human brain ventricles from MR images by knowledge-based region growing and trimming

Automatic segmentation of the human brain ventricular system from MR images is useful in studies of brain anatomy and its diseases. Existing intensity-based segmentation methods are adaptive to large shape and size variations of the ventricular system, but may leak to the non-ventricular regions due to the non-homogeneity, noise and partial volume effect in the images. Deformable model-based methods are more robust to noise and alleviate the leakage problem, but may generate wrong results when the shape or size of the ventricle to be segmented in the images has a large difference in comparison to its model. In this paper, we propose a knowledge-based region growing and trimming approach where: (1) a model of a ventricular system is used to define regions of interest (ROI) for the four ventricles (i.e., left, right, third and fourth); (2) to segment a ventricle in its ROI, a region growing procedure is first applied to obtain a connected region that contains the ventricle, and (3) a region trimming procedure is then employed to trim the non-ventricle regions. A hysteretic thresholding is developed for the region growing procedure to cope with the partial volume effect and minimize non-ventricular regions. The domain knowledge on the shape and intensity features of the ventricular system is used for the region trimming procedure. Due to the joint use of the model-based and intensity-based approaches, our method is robust to noise and large shape and size variations. Experiments on 18 simulated and 58 clinical MR images show that the proposed approach is able to segment the ventricular system accurately with the dice similarity coefficient ranging from 91% to 99%.