Daniel Smutek

Segmentation of multiple organs in non-contrast 3D abdominal CT images

ObjectiveWe propose a simultaneous extraction method for 12 organs from non-contrast three-dimensional abdominal CT images.Materials and methodsThe proposed method uses an abdominal cavity standardization process and atlas guided segmentation incorporating parameter estimation with the EM algorithm to deal with the large fluctuations in the feature distribution parameters between subjects. Segmentation is then performed using multiple level sets, which minimize the energy function that considers the hierarchy and exclusiveness between organs as well as uniformity of grey values in organs. To assess the performance of the proposed method, ten non-contrast 3D CT volumes were used.ResultsThe accuracy of the feature distribution parameter estimation was slightly improved using the proposed EM method, resulting in better performance of the segmentation process. Nine organs out of twelve were statistically improved compared with the results without the proposed parameter estimation process. The proposed multiple level sets also boosted the performance of the segmentation by 7.2 points on average compared with the atlas guided segmentation. Nine out of twelve organs were confirmed to be statistically improved compared with the atlas guided method.ConclusionThe proposed method was statistically proved to have better performance in the segmentation of 3D CT volumes.

Image texture analysis of sonograms in chronic inflammations of thyroid gland

The current practice in assessing sonographic findings of chronic inflamed thyroid tissue is mainly qualitative, based just on a physicians experience. This study shows that inflamed and healthy tissues can be differentiated by automatic texture analysis of B-mode sonographic images. Feature selection is the most important part of this procedure. We employed two selection schemes for finding recognition-optimal features: one based on compactness and separability and the other based on classification error. The full feature set included Muzzolinis spatial features and Haralicks co-occurrence features. These features were selected on a set of 2430 sonograms of 81 subjects, and the classifier performance was evaluated on a test set of 540 sonograms of 18 independent subjects. A classification success rate of 100% was achieved with as few as one optimal feature among the 129 texture characteristics tested. Both selection schemes agreed on the best features. The results were confirmed on the independent test set. The stability of the results with respect to sonograph setting, thyroid gland segmentation and scanning direction was tested.

Medical image analysis of 3D CT images based on extension of Haralick texture features

PURPOSE A new approach to the segmentation of 3D CT images is proposed in an attempt to provide texture-based segmentation of organs or disease diagnosis. 3D extension of Haralick texture features was studied calculating co-occurrences of all voxels in a small cubic region around the voxel. RESULTS For verification, the proposed method was tested on a set of abdominal 3D volumes of patients. Statistically, the improvement in segmentation was significant for most of the organs considered herein. CONCLUSIONS The proposed method has potential application in medical image segmentation, including diagnosis of diseases.

Systematic Construction of Texture Features for Hashimoto's Lymphocytic Thyroiditis Recognition from Sonographic Images

The success of discrimination between normal and inflamed parenchyma of thyroid gland by means of automatic texture analysis is largely determined by selecting descriptive yet simple and independent sonographic image features. We replace the standard non-systematic process of feature selection by systematic feature construction based on the search for the separation distances among a clique of n pixels that minimise conditional entropy of class label given all data. The procedure is fairly general and does not require any assumptions about the form of the class probability density function. We show that a network of weak Bayes classifiers using 4-cliques as features and combined by majority vote achieves diagnosis recognition accuracy of 92%, as evaluated on a set of 741 B-mode sonographic images from 39 subjects. The results suggest the possibility to use this method in clinical diagnostic process.

Image registration accuracy estimation without ground truth using bootstrap

We consider the problem of estimating the local accuracy of image registration when no ground truth data is available. The technique is based on a statistical resampling technique called bootstrap. Only the two input images are used, no other data are needed. The general bootstrap uncertainty estimation framework described here is in principle applicable to most of the existing pixel based registration techniques. In practice, a large computing power is required. We present experimental results for a block matching method on an ultrasound image sequence for elastography with both known and unknown deformation field.

Computational elastography from standard ultrasound image sequences by global trust region optimization

A new approach is proposed to estimate the spatial distribution of shear modulus of tissues in-vivo. An image sequence is acquired using a standard medical ultrasound scanner while varying the force applied to the handle. The elastic properties are then recovered simultaneously with the inter-frame displacement fields using a computational procedure based on finite element modeling and trust region constrained optimization. No assumption about boundary conditions is needed. The optimization procedure is global, taking advantage of all available images. The algorithm was tested on phantom, as well as on real clinical images.

Automatic Internal Medicine Diagnostics Using Statistical Imaging Methods

To develop a computer-aided diagnostic system for diagnosing different internal medicine diseases based on imaging methods. We focus on focal liver lesions in CT images. The diagnosing process follows the learning phase from known images. For image description, 22 first-order and 108 second-order texture features are used. They are used as input for network of Bayes classifiers. The best value of 100% success of classification between hepatocellular carcinoma and non-parasitic solitary liver cysts was achieved. The method allows discriminating between different liver diseases based on computer imaging. The method may be very useful in cases where any internal images of patients already diagnosed are available

On reproducibility of ultrasound image classification

Ultrasound B-mode images of thyroid gland were previously analyzed to distinguish normal tissue from inflamed tissue due to Hashimotos Lymphocytic Thyroiditis. This is a two-class recognition problem. Sensitivity and specificity of 100% was reported using Bayesian classifier with selected texture features. These results were obtained on 99 subjects at a fixed setting of one specific sonograph, for a given manual thyroid gland segmentation and sonographic scan orientation (longitudinal, transversal). To evaluate the reproducibility of the method, sensitivity analysis is the topic of this paper. A general method for determining feature sensitivity to variables influencing the scanning process is proposed. Jensen Shannon distances between modified and unmodified inter- and intra-class feature probability distributions capture the changes induced by the variables. Among selected features, the least sensitive one is found. The proposed sensitivity evaluation method can be used in other problems with complex and non-linear dependencies on variables that cannot be controlled.

Estimating elastic properties of tissues from standard 2D ultrasound images

We propose a way of measuring elastic properties of tissues in-vivo, using standard medical image ultrasound machine without any special hardware. Images are acquired while the tissue is being deformed by a varying pressure applied by the operator on the hand-held ultrasound probe. The local elastic shear modulus is either estimated from a local displacement field reconstructed by an elastic registration algorithm, or both the modulus and the displacement are estimated simultaneously. The relation between modulus and displacement is calculated using a finite element method (FEM). The estimation algorithms were tested on both synthetic, phantom and real subject data.

Genetic algorithms for thyroid gland ultrasound image feature reduction

The problem of automatic classification of ultrasound images is addressed. For texture analysis of ultrasound images quantifiable indexes, called features, are used. Classification was performed using Gaussian mixture model based on Bayes classifier. The common problem of texture analysis is a feature selection for classification tasks. In this work we use genetic algorithms for a feature subset selection. Total number of 387 features was used, consisting of spatial an co-occurance statistical texture features (proposed by Muzzolini and Haralick). The classification infers between healthy thyroid gland and thyroid gland with chronic inflammation.