Grzegorz Soza

Fast volumetric deformation on general purpose hardware

High performance deformation of volumetric objects is a common problem in computer graphics that has not yet been handled sufficiently. As a supplement to 3D texture based volume rendering, a novel approach is presented, which adaptively subdivides the volume into piecewise linear patches. An appropriate mathematical model based on tri-linear interpolation and its approximations is proposed. New optimizations are introduced in this paper which are especially tailored to an efficient implementation using general purpose rasterization hardware, including new technologies, such as vertex programs and pixel shaders. Additionally, a high performance model for local illumination calculation is introduced, which meets the aesthetic requirements of visual arts and entertainment. The results demonstrate the significant performance benefit and allow for time-critical applications, such as computer assisted surgery.

A Generic Probabilistic Active Shape Model for Organ Segmentation

Probabilistic models are extensively used in medical image segmentation. Most of them employ parametric representations of densities and make idealizing assumptions, e.g., normal distribution of data. Often, such assumptions are inadequate and limit a broader application. We propose here a novel probabilistic active shape model for organ segmentation, which is entirely built upon non-parametric density estimates. In particular, a nearest neighbor boundary appearance model is complemented by a cascade of boosted classifiers for region information and combined with a shape model based on Parzen density estimation. Image and shape terms are integrated into a single level set equation. Our approach has been evaluated for 3-D liver segmentation using a public data base originating from a competition (http://sliver07.org). With an average surface distance of 1.0 mm and an average volume overlap error of 6.5%, it outperforms other automatic methods and provides accuracy close to interactive ones. Since no adaptions specific to liver segmentation have been made, our probabilistic active shape model can be applied to other segmentation tasks easily.

Correction of susceptibility artifacts in diffusion tensor data using non-linear registration

Diffusion tensor imaging can be used to localize major white matter tracts within the human brain. For surgery of tumors near eloquent brain areas such as the pyramidal tract this information is of importance to achieve an optimal resection while avoiding post-operative neurological deficits. However, due to the small bandwidth of echo planar imaging, diffusion tensor images suffer from susceptibility artifacts resulting in positional shifts and distortion. As a consequence, the fiber tracts computed from echo planar imaging data are spatially distorted. We present an approach based on non-linear registration using Bézier functions to efficiently correct distortions due to susceptibility artifacts. The approach makes extensive use of graphics hardware to accelerate the non-linear registration procedure. An improvement presented in this paper is a more robust and efficient optimization strategy based on simultaneous perturbation stochastic approximation (SPSA). Since the accuracy of non-linear registration is crucial for the value of the presented correction method, two techniques were applied in order to prove the quality of the proposed framework. First, the registration accuracy was evaluated by recovering a known transformation with non-linear registration. Second, landmark-based evaluation of the registration method for anatomical and diffusion tensor data was performed. The registration was then applied to patients with lesions adjacent to the pyramidal tract in order to compensate for susceptibility artifacts. The effect of the correction on the pyramidal tract was then quantified by measuring the position of the tract before and after registration. As a result, the distortions observed in phase encoding direction were most prominent at the cortex and the brainstem. The presented approach allows correcting fiber tract distortions which is an important prerequisite when tractography data are integrated into a stereotactic setup for intra-operative guidance.

A learning based deformable template matching method for automatic rib centerline extraction and labeling in CT images

The automatic extraction and labeling of the rib centerlines is a useful yet challenging task in many clinical applications. In this paper, we propose a new approach integrating rib seed point detection and template matching to detect and identify each rib in chest CT scans. The bottom-up learning based detection exploits local image cues and top-down deformable template matching imposes global shape constraints. To adapt to the shape deformation of different rib cages whereas maintain high computational efficiency, we employ a Markov Random Field (MRF) based articulated rigid transformation method followed by Active Contour Model (ACM) deformation. Compared with traditional methods that each rib is individually detected, traced and labeled, the new approach is not only much more robust due to prior shape constraints of the whole rib cage, but removes tedious post-processing such as rib pairing and ordering steps because each rib is automatically labeled during the template matching. For experimental validation, we create an annotated database of 112 challenging volumes with ribs of various sizes, shapes, and pathologies such as metastases and fractures. The proposed approach shows orders of magnitude higher detection and labeling accuracy than state-of-the-art solutions and runs about 40 seconds for a complete rib cage on the average.

Non-rigid Registration with Use of Hardware-Based 3D Bézier Functions

In this paper we introduce a new method for non-rigid voxel-based registration. In many medical applications there is a need to establish an alignment between two image datasets. Often a registration of a time-shifted medical image sequence with appearing deformation of soft tissue (e.g. pre- and intraoperative data) has to be conducted. Soft tissue deformations are usually highly non-linear. For the handling of this phenomenon and for obtaining an optimal non-linear alignment of respective datasets we transform one of them using 3D Bezier functions, which provides some inherent smoothness as well as elasticity. In order to find the optimal transformation, many evaluations of this Bezier function are necessary. In order to make the method more efficient, graphics hardware is extensively used. We applied our non-rigid algorithm successfully to MR brain images in several clinical cases and showed its value.

Multi-stage osteolytic spinal bone lesion detection from CT data with internal sensitivity control

Spinal bone lesion detection is a challenging and important task in cancer diagnosis and treatment monitoring. In this paper we present a method for fully-automatic osteolytic spinal bone lesion detection from 3D CT data. It is a multi-stage approach subsequently applying multiple discriminative models, i.e., multiple random forests, for lesion candidate detection and rejection to an input volume. For each detection stage an internal control mechanism ensures maintaining sensitivity on unseen true positive lesion candidates during training. This way a pre-defined target sensitivity score of the overall system can be taken into account at the time of model generation. For a lesion not only the center is detected but also, during post-processing, its spatial extension along the three spatial axes defined by the surrounding vertebral bodys local coordinate system. Our method achieves a cross-validated sensitivity score of 75% and a mean false positive rate of 3.0 per volume on a data collection consisting of 34 patients with 105 osteolytic spinal bone lesions. The median sensitivity score is 86% at 2.0 false positives per volume.

Fast and Adaptive Finite Element Approach for Modeling Brain Shift

Objective: In this paper we introduce a finite element-based strategy for simulation of brain deformation occurring during neurosurgery. The phenomenon, known as brain shift, causes a decrease in the accuracy of neuronavigation systems that rely on preoperatively acquired data. This can be compensated for with a computational model of the brain deformation process. By applying model calculations to preoperative images, an update within the operating room can be performed. Methods: One of the crucial concerns in the context of developing a physical-based model is the choice of governing equations describing the physics of the phenomenon. In this work, deformation of brain tissue is expressed in terms of a 3D consolidation model for a linearly elastic and porous fluid. The next crucial issue is ensuring stable calculations within the chosen model. For this purpose, we developed a special technique for generating the underlying geometry for the simulation. With this technique an unstructured grid consisting of regular tetrahedra is created, whereupon time-dependent finite element simulation is performed in an adaptive manner. Results: We applied our algorithm to preoperative MR scans and investigated the value of the method. Due to the adaptivity of the method, only 5-10% of the computing time was needed as compared to traditional finite element approaches based on a uniformly subdivided grid. The results of the experiments were compared to the corresponding intraoperative MR scans. A close match between the computed deformation of the brain and the displacement resulting from the intraoperative data was observed. Conclusion: A model-based approach for the simulation of brain shift is presented. In this computational model the brain tissue is described as an elastic and porous material using Biot consolidation theory. Validating experiments conducted with MR data provided promising results.

Estimating Mechanical Brain Tissue Properties with Simulation and Registration

In this work a new method for the determination of the mechanical properties of brain tissue is introduced. Young’s modulus E and Poisson’s ratio ν are iteratively estimated based on a finite element model for brain shift and on the information contained in pre- and intraoperative MR data after registration. In each iteration, a 3D dataset is generated according to the displacement vector field resulting from a numerical simulation of the intraoperative brain deformation. This reconstruction is parametrized by elastic moduli of tissue. They are automatically varied in order to achieve the best correspondence between the grey value distribution in the reconstructed image and the intensity entropy in the MR image of the brain undergoing deformation. This work contributes to the difficult problem of defining correct mechanical parameters to perform reliable model calculations of brain deformation. Proper boundary conditions that are crucial in this context are also addressed.

Sketch-based Image-independent Editing of 3D Tumor Segmentations using Variational Interpolation

In the past years sophisticated automatic segmentation algorithms for various medical image segmentation problems have been developed. However, there are always cases where automatic algorithms fail to provide an acceptable segmentation. In these cases the user needs efficient segmentation correction tools, a problem which has not received much attention in research. Cases to be manually corrected are often particularly difficult and the image does often not provide enough information for segmentation, so we present an image-independent method for intuitive sketch-based editing of 3D tumor segmentations. It is based on an object reconstruction using variational interpolation and can be used in any 3D modality, such as CT or MRI. We also discuss sketch-based editing in 2D as well as a hole-correction approach for variational interpolation. Our manual correction algorithm has been evaluated on 89 segmentations of tumors in CT by 2 technical experts with 6+ years of experience in tumor segmentation and assessment. The experts rated the quality of our correction tool as acceptable or better in 92.1% of the cases. They needed a median number of 4 correction steps with one step taking 0.4s on average.

Automatic Liver Segmentation Using the Random Walker Algorithm

In this paper we present a new method for fully automatic liver segmentation in computed tomography images. First, an initial set of seed points for the random walker algorithm is created. In this context, voxels belonging to air, fat tissue and ribcage are labeled as background. Furthermore, depending on the shape of the ribcage and voxel intensities, several seed points inside the liver are automatically selected as foreground. This seed mask is then used to initialize the segmentation algorithm. Our method was successfully tested on data of 22 patients.