Lixu Gu

Implementation of medical image segmentation in CUDA

As the fast development of GPU, people tend to use it for more general purposes than its original graphic related work. The high parallel computation capabilities of GPU are welcomed by programmers who work at medical image processing which always have to deal with a large scale of voxel computation. The birth of NVIDIAreg CUDAtrade technology and CUDA-enabled GPUs brought a revolution in the general purpose GPU area. In this paper, we propose the implementation of several medical image segmentation algorithms using CUDA and CUDA-enabled GPUs, compare their performance and results to the previous implementation in old version of GPU and CPU, indicate the advantages of using CUDA technology and how to design algorithm to make full use of it.

An Improved Level Set for Liver Segmentation and Perfusion Analysis in MRIs

Determining liver segmentation accurately from MRIs is the primary and crucial step for any automated liver perfusion analysis, which provides important information about the blood supply to the liver. Although implicit contour extraction methods, such as level set methods (LSMs) and active contours, are often used to segment livers, the results are not always satisfactory due to the presence of artifacts and low-gradient response on the liver boundary. In this paper, we propose a multiple-initialization, multiple-step LSM to overcome the leakage and over-segmentation problems. The multiple-initialization curves are first evolved separately using the fast marching methods and LSMs, which are then combined with a convex hull algorithm to obtain a rough liver contour. Finally, the contour is evolved again using global level set smoothing to determine a precise liver boundary. Experimental results on 12 abdominal MRI series showed that the proposed approach obtained better liver segmentation results, so that a refined liver perfusion curve without respiration affection can be obtained by using a modified chamfer matching algorithm and the perfusion curve is evaluated by radiologists.

A homotopy-based sparse representation for fast and accurate shape prior modeling in liver surgical planning

Shape prior plays an important role in accurate and robust liver segmentation. However, liver shapes have complex variations and accurate modeling of liver shapes is challenging. Using large-scale training data can improve the accuracy but it limits the computational efficiency. In order to obtain accurate liver shape priors without sacrificing the efficiency when dealing with large-scale training data, we investigate effective and scalable shape prior modeling method that is more applicable in clinical liver surgical planning system. We employed the Sparse Shape Composition (SSC) to represent liver shapes by an optimized sparse combination of shapes in the repository, without any assumptions on parametric distributions of liver shapes. To leverage large-scale training data and improve the computational efficiency of SSC, we also introduced a homotopy-based method to quickly solve the L1-norm optimization problem in SSC. This method takes advantage of the sparsity of shape modeling, and solves the original optimization problem in SSC by continuously transforming it into a series of simplified problems whose solution is fast to compute. When new training shapes arrive gradually, the homotopy strategy updates the optimal solution on the fly and avoids re-computing it from scratch. Experiments showed that SSC had a high accuracy and efficiency in dealing with complex liver shape variations, excluding gross errors and preserving local details on the input liver shape. The homotopy-based SSC had a high computational efficiency, and its runtime increased very slowly when repositorys capacity and vertex number rose to a large degree. When repositorys capacity was 10,000, with 2000 vertices on each shape, homotopy method cost merely about 11.29 s to solve the optimization problem in SSC, nearly 2000 times faster than interior point method. The dice similarity coefficient (DSC), average symmetric surface distance (ASD), and maximum symmetric surface distance measurement was 94.31 ± 3.04%, 1.12 ± 0.69 mm and 3.65 ± 1.40 mm respectively.

A hybrid deformable model for real-time surgical simulation.

Modeling organ deformation in real remains a challenge in virtual minimally invasive (MIS) surgery simulation. In this paper, we propose a new hybrid deformable model to simulate deformable organs in the real-time surgical training system. Our hybrid model uses boundary element method (BEM) to compute global deformation based on a coarse surface mesh and uses a mass-spring model to simulate the dynamic behaviors of soft tissue interacting with surgical instruments. The simulation result is coupled with a high-resolution rendering mesh through a particle surface interpolation algorithm. Accurate visual and haptic feedbacks are provided in real time and temporal behaviors of biological soft tissues including viscosity and creeping are modeled as well. We prove our model to be suitable to work in complex virtual surgical environment by integrating it into a MIS training system. The hybrid model is evaluated with respect to efficiency, accuracy and robustness by a series of experiments.

A Novel Level Set Based Shape Prior Method for Liver Segmentation from MRI Images

Liver segmentation in MR Image is the first step of our automated liver perfusion analysis project. Traditional Level Set methods and active contours were often used to segment the liver, but the results were not always promising due to noise and the low gradient response on the liver boundary. In this paper we propose a novel level set based variational approach that incorporates shape prior knowledge into the improved Chan-Veses model [1] which can overcome the leakage and over-segmentation problems. The experiments are taken on abdomen MRI series and the results reveal that our improved level set based shape prior method can segment liver shape precisely and a refined liver perfusion curve without respiration affection can be achieved.

Magnetic Navigation for Thoracic Aortic Stent-graft Deployment Using Ultrasound Image Guidance

We propose a system for thoracic aortic stent-graft deployment that employs a magnetic tracking system (MTS) and intraoperative ultrasound (US). A preoperative plan is first performed using a general public utilities-accelerated cardiac modeling method to determine the target position of the stent-graft. During the surgery, an MTS is employed to track sensors embedded in the catheter, cannula, and the US probe, while a fiducial landmark based registration is used to map the patients coordinate to the image coordinate. The surgical target is tracked in real time via a calibrated intraoperative US image. Under the guidance of the MTS integrated with the real-time US images, the stent-graft can be deployed to the target position without the use of ionizing radiation. This navigation approach was validated using both phantom and animal studies. In the phantom study, we demonstrate a US calibration accuracy of 1.5 ± 0.47 mm, and a deployment error of 1.4 ± 0.16 mm. In the animal study, we performed experiments on five porcine subjects and recorded fiducial, target, and deployment errors of 2.5 ± 0.32, 4.2 ± 0.78, and 2.43 ± 0.69 mm, respectively. These results demonstrate that delivery and deployment of thoracic stent-graft under MTS-guided navigation using US imaging is feasible and appropriate for clinical application.

An Automatic and Fast Centerline Extraction Algorithm for Virtual Colonoscopy

This paper introduces a new refined centerline extraction algorithm, which is based on and significantly improved from distance mapping algorithms. The new approach include three major parts: employing a colon segmentation method; designing and realizing a fast Euclidean transform algorithm and inducting boundary voxels cutting (BVC) approach. The main contribution is the BVC processing, which greatly speeds up the Dijkstra algorithm and improves the whole performance of the new algorithm. Experimental results demonstrate that the new centerline algorithm was more efficient and accurate comparing with existing algorithms

A robust and real-time vascular intervention simulation based on Kirchhoff elastic rod

A virtual reality (VR) based vascular intervention simulation system is introduced in this paper, which helps trainees develop surgical skills and experience complications in safety remote from patients. The system simulates interventional radiology procedures, in which flexible tipped guidewires are employed to advance diagnostic or therapeutic catheters into vascular anatomy of a patient. A real-time physically-based modeling approach ground on Kirchhoff elastic rod is proposed to simulate complicated behaviors of guidewires and catheters. The slender body of guidewire and catheter is modeled using more efficient special case of naturally straight, isotropic Kirchhoff rods, and the shorter flexible tip composed of straight or angled design is modeled using more complex generalized Kirchhoff rods. The motion equations for guidewire and catheter were derived with continuous elastic energy, followed by a discretization using a linear implicit scheme that guarantees stability and robustness. In addition, we used a fast-projection method to enforce the inextensibility of guidewire and catheter. An adaptive sampling algorithm was also implemented to improve the simulation efficiency without decrease of accuracy. Experimental results revealed that our system is both robust and efficient in a real-time performance.

Virtual Surgery Deformable Modelling Employing GPU Based Computation

The development of Virtual Environment (VE) systems is a challenging endeavor with a complex problem domain. The experience in the past decade has helped contribute significantly to various measures of software quality of the resulting VE systems. However, the resulting solutions remain monolithic in nature without addressing successfully the issue of system interoperability and software aging. This paper argues that the problem resides in the traditional system centric approach and that an alternative approach based on system of systems engineering is necessary. As a result, the paper presents a reference architecture based on layers, where only the core is required for deployment and all others are optional. The paper also presents an evaluation methodology to assess the validity of the resulting architecture, which was applied to the proposed core layer and involving individual sessions with 12 experts in developing VE systems.To achieve the real-time requirement of realistic deformable modelling, it is necessary to use the acceleration techniques such as GPU computing for FEM and employ the feasible hybrid structures in a virtual surgery simulation system. In this paper, we present a linear or nonlinear deformable model of soft tissue. In addition to the efficient meshing and basic finite element method, the high computation rate is achieved through two novel methods. Firstly, the major calculation work in the conjugate gradient solver for deformation is moved from the CPU to GPU in order to promote the calculation. Secondly, we apply the hybrid structures of deformable models, by fully calculating the volumetric deformation in the local operation part while only calculating the global deformation by medial representation method. Experiments have been given to show the feasibility and efficiency of the model.

Soft Tissue Deformation Simulation in Virtual Surgery using Nonlinear Finite Element Method

Simulation for soft tissues realistic deformation is an important part in Virtual Surgery. For large global deformation of soft tissue, linear elastic models are inappropriate, such as Mass-Spring and linear Finite Element Method (FEM). In this paper we present a simulation for 3D soft tissue using nonlinear strain computation. To get a finer mesh for FEM, we consider meshing algorithm based on Improved Delaunay criterion. Besides, we would present Spatial Hashing Collision Detection method and some improvement for real-time computation.