Chee-Kong Chui

Integrating spatial fuzzy clustering with level set methods for automated medical image segmentation

The performance of the level set segmentation is subject to appropriate initialization and optimal configuration of controlling parameters, which require substantial manual intervention. A new fuzzy level set algorithm is proposed in this paper to facilitate medical image segmentation. It is able to directly evolve from the initial segmentation by spatial fuzzy clustering. The controlling parameters of level set evolution are also estimated from the results of fuzzy clustering. Moreover the fuzzy level set algorithm is enhanced with locally regularized evolution. Such improvements facilitate level set manipulation and lead to more robust segmentation. Performance evaluation of the proposed algorithm was carried on medical images from different modalities. The results confirm its effectiveness for medical image segmentation.

Combined compression and elongation experiments and non-linear modelling of liver tissue for surgical simulation

Uniaxial stress-strain data were obtained from in vitro experiments on 20 porcine livers for compressions, elongations and cycles of compression and then elongation. There were about 70 cylindrical samples, with diameter 7 mm and varying height (4–11 mm). The combined compression and elongation test provide a unified framework for both compression and elongation for applications such as computer-aided surgical simulation. It enable the zero stress state of the experimental liver sample to be precisely determined. A new equation that combined both logarithmic and polynomial strain energy forms was proposed in modelling these experimental data. The assumption of incompressibility was justified from a preliminary Poissons ratio for elongation and compression at 0.43±0.16 and 0.47±0.15, respectively. This equation provided a good fit for the observed mechanical properties of liver during compression-elongation cycles and for separate compressions or elongations. The root mean square errors were 91.92±17.43 Pa, 57.55±13.23 Pa and 29.78±17.67 Pa, respectively. In comparison with existing strain energy functions, this combined model was the better constitutive equation. Application of this theoretical model to small liver samples and other tissues demonstrated its suitability as the material model of choice for soft tissue.

Transversely isotropic properties of porcine liver tissue: experiments and constitutive modelling.

Knowledge of the biomechanical properties of soft tissue, such as liver, is important in modelling computer aided surgical procedures. Liver tissue does not bear mechanical loads, and, in numerical simulation research, is typically assumed to be isotropic. Nevertheless, a typical biological soft tissue is anisotropic. In vitro uniaxial tension and compression experiments were conducted on porcine cylindrical and cubical liver tissue samples respectively assuming a simplistic architecture of liver tissue with its constituent lobule and connective tissues components. With the primary axis perpendicular to the cross sectional surface of samples, the tissue is stiffer with tensile or compressive force in the axial direction compared to that of the transverse direction. At 20% strain, about twice as much force is required to elongate a longitudinal tissue sample than that of a transverse sample. Results of the study suggest that liver tissue is transversely isotropic. A combined strain energy based constitutive equation for transversely isotropic material is proposed. The improved capability of this equation to model the experimental data compared to its previously disclosed isotropic version suggests that the assumption on the fourth invariant in the constitutive equation is probably correct and that anisotropy properties of liver tissue should be considered in surgical simulation.

Fast segmentation of bone in CT images using 3D adaptive thresholding

Fast bone segmentation is often important in computer-aided medical systems. Thresholding-based techniques have been widely used to identify the object of interest (bone) against dark backgrounds. However, the darker areas that are often present in bone tissue may adversely affect the results obtained using existing thresholding-based segmentation methods. We propose an automatic, fast, robust and accurate method for the segmentation of bone using 3D adaptive thresholding. An initial segmentation is first performed to partition the image into bone and non-bone classes, followed by an iterative process of 3D correlation to update voxel classification. This iterative process significantly improves the thresholding performance. A post-processing step of 3D region growing is used to extract the required bone region. The proposed algorithm can achieve sub-voxel accuracy very rapidly. In our experiments, the segmentation of a CT image set required on average less than 10s per slice. This execution time can be further reduced by optimizing the iterative convergence process.

A new unified level set method for semi-automatic liver tumor segmentation on contrast-enhanced CT images

Highlights? Computerized liver tumor segmentation is a challenging issue. ? Segmentation confronts great variations of shape, intensity and contrast-enhancement. ? Common level set models are not robust for contrast-enhanced liver tumor segmentation. ? We propose a level set model to unify image gradient, region competition and prior knowledge. ObjectiveComputerized liver tumor segmentation on computed tomography (CT) images is a challenging problem. Level set methods have been proposed for CT liver and tumor segmentation. However, the common models using image gradient or region competition have inherent drawbacks, and are not very robust for liver tumor segmentation. MethodsWe propose a new unified level set model to integrate image gradient, region competition and prior information for CT liver tumor segmentation. The probabilistic distribution of liver tumors is estimated by unsupervised fuzzy clustering, and is utilized to enhance the object indication function, define the directional balloon force and regulate region competition. This unified model has been evaluated on 25 two-dimensional (2D) CT scans and 4 three-dimensional (3D) CT scans with 10 tumors. ResultsFor the 2D dataset, the area overlapping error (AOE) is 12.75?5.76%, the relative area difference (RAD) is -4.28?9.58%, the average contour distance (ACD) is 1.66?1.09mm, and the maximum contour distance (MCD) is 4.29?2.75mm. For the 3D dataset, the volume overlapping error (VOE) is 26.31?5.79%, the relative volume difference (RVD) is -10.64?7.55%, the average surface distance (ASD) is 1.06?0.38mm, and the maximum surface distance (MSD) is 8.66?3.17mm. All results are competitive with that of the state-of-the-art methods. ConclusionThe new unified level set model is an effective solution for liver tumor segmentation on contrast-enhanced CT images.

In vitro measurement of mechanical properties of liver tissue under compression and elongation using a new test piece holding method with surgical glue

There is a need to determine biomechanical properties of liver tissue to develop realistic elastic deformable liver model for computer aided surgery. In this report, we introduced a method to measure mechanical properties using surgical instant adhesive (surgical glue). The method made easier to define the mechanical boundary conditions for test pieces. It also makes it possible to conduct both compression and elongation test on the same test piece. In actual deformation of liver during surgical intervention, the tissue is subject both to compression and elongation. Identification of mechanical properties in the range where mechanical force changes from compression to elongation is important. We can identify the stress-strain relationship of liver samples in the transition range from compression to elongation. We also investigated viscoelastic properties by compressing the sample at different velocities. The obtained results can be applied to non linear FEM analysis of liver tissue.

Real-Time Interactive Simulator for Percutaneous Coronary Revascularization Procedures

This article describes the simulation of real-time catheter navigation in our interactive interventional cardiology simulation system (ICard). ICard is designed to enable medical students or physicians to familiarize themselves with the techniques of interventional catheterization procedures. The ICard software provides three-dimensional (3-D) views of the blood vessels and fluoroscopic images for real-time visualization of the catheter position. The 3-D human vasculature is built from various image data sets and is represented with a central line hierarchy model. Navigation of the catheter and guide wire and their interaction with blood vessels are implemented by applying the finite element method. Physical modeling that features the elasticity of the catheter and guide wire was developed to provide a realistic simulation of catheterization procedures. An electromechanical device was also developed in the system, allowing physical manipulation of catheter and guide wire movements. ICard can be used for training and design of equipment for interventional cardiology and may be further extended for pretreatment planning.

Simulation of interventional neuroradiology procedures

We describe the design and development of a computer environment for planning interventional neuroradiology procedures. The Neuroradiology Catheterization Simulator called NeuroCath is intended for interventional procedures involving vascular malformations, such as aneurysms, stenosis, and AVMs. NeuroCath include extraction and construction of a vascular model from different imaging modalities that represents the anatomy of patient in a computationally efficient manner, and a FEM-based physical model that simulates the behavior between the devices and cerebral vasculature. This model comprises topology, geometry (normal and pathological), and physical properties of the patient-specific vasculature. It also provides a reliable measurement of distance and volume allowing calculation of the size of vessels and aneurysms. A realistic visual interface with multiple, synchronized windows is developed. The visual interface comprises of fluoroscopic display that duplicates the views to be seen in actual intentional procedures, and other displays that enhance interpretation of the anatomy of the patient. The hybrid volume and surface renderer provides insight into inferior and exterior of patients vasculature. NeuroCath is also provided with the haptic apparatus that gives the interventional neuroradiologist the sense of touch during intervention planning and training.

Computational biomechanical modelling of the lumbar spine using marching-cubes surface smoothened finite element voxel meshing

There is a need for the development of finite element (FE) models based on medical datasets, such as magnetic resonance imaging and computerized tomography in computation biomechanics. Direct conversion of graphic voxels to FE elements is a commonly used method for the generation of FE models. However, conventional voxel-based methods tend to produce models with jagged surfaces. This is a consequence of the inherent characteristics of voxel elements; such a model is unable to capture the geometries of anatomical structures satisfactorily. We have developed a robust technique for the automatic generation of voxel-based patient-specific FE models. Our approach features a novel tetrahedronization scheme that incorporates marching-cubes surface smoothing together with a smooth-distortion factor (SDF). The models conform to the actual geometries of anatomical structures of a lumbar spine segment (L3). The resultant finite element analysis (FEA) at the surfaces is more accurate compared to the use of conventional voxel-based generated FE models. In general, models produced by our method were superior compared to that obtained using the commercial software ScanFE.

Rapid surface registration of 3D volumes using a neural network approach

An automatic surface-based rigid registration system using a neural network representation is proposed. The system has been applied to register human bone structures for image-guided surgery. A multilayer perceptron neural network is used to construct a patient-specific surface model from pre-operative images. A surface representation function derived from the resultant neural network model is then employed for intra-operative registration. The optimal transformation parameters are obtained via an optimization process. This segmentation/registration system achieves sub-voxel accuracy comparable to that of conventional techniques, and is significantly faster. These advantages are demonstrated using image datasets of the calcaneus and vertebrae.