Automatic landmark extraction from image data using modified growing neural gas network

A new method for automatic landmark extraction from MR brain images is presented. In this method, landmark extraction is accomplished by modifying growing neural gas (GNG), which is a neural-network-based cluster-seeking algorithm. Using modified GNG (MGNG) corresponding dominant points of contours extracted from two corresponding images are found. These contours are borders of segmented anatomical regions from brain images. The presented method is compared to: 1) the node splitting-merging Kohonen model and 2) the Teh-Chin algorithm (a well-known approach for dominant points extraction of ordered curves). It is shown that the proposed algorithm has lower distortion error, ability of extracting landmarks from two corresponding curves simultaneously, and also generates the best match according to five medical experts.     

Multiwavelet grading of pathological images of prostate

Histological grading of pathological images is used to determine level of malignancy of cancerous tissues. This is a very important task in prostate cancer prognosis, since it is used for treatment planning. If infection of cancer is not rejected by non-invasive diagnostic techniques like magnetic resonance imaging, computed tomography scan, and ultrasound, then biopsy specimens of tissue are tested. For prostate, biopsied tissue is stained by hematoxyline and eosine method and viewed by pathologists under a microscope to determine its histological grade. Human grading is very subjective due to interobserver and intraobserver variations and in some cases difficult and time-consuming. Thus, an automatic and repeatable technique is needed for grading. Gleason grading system is the most common method for histological grading of prostate tissue samples. According to this system, each cancerous specimen is assigned one of five grades. Although some automatic systems have been developed for analysis of pathological images, Gleason grading has not yet been automated; the goal of this research is to automate it. To this end, we calculate energy and entropy features of multiwavelet coefficients of the image. Then, we select most discriminative features by simulated annealing and use a k-nearest neighbor classifier to classify each image to appropriate grade (class). The leaving-one-out technique is used for error rate estimation. We also obtain the results using features extracted by wavelet packets and co-occurrence matrices and compare them with the multiwavelet method. Experimental results show the superiority of the multiwavelet transforms compared with other techniques. For multiwavelets, critically sampled preprocessing outperforms repeated-row preprocessing and has less sensitivity to noise for second level of decomposition. The first level of decomposition is very sensitive to noise and, thus, should not be used for feature extraction. The best multiwavelet method grades prostate pathological images correctly 97% of the time.     

A multiresolution approach for contour extraction from brain images

Many image registration methods use head surface, brain surface, or inner/outer surface of the skull to estimate rotation and translation parameters. The inner surface of the skull is also used for intracranial volume segmentation which is considered the first step in segmentation and analysis of brain images. The surface is usually characterized by a set of edge or contour points extracted from cross-sectional images. Automatic extraction of contour points is complicated by discontinuity of edges in the back of the eyes and ears and sometimes by a previous surgery or an inadequate field of view. We have developed an automated method for contour extraction that connects discontinuities using a multiresolution pyramid. Steps of the method are: (1) Contour points are found by an edge-tracking algorithm; (2) A multiresolution pyramid of contour points is constructed; (3) Contour points of reduced images are found; (4) From the continuous contour found at the lowest resolution, contour points at a higher resolution are found; (5) Step 4 is repeated until contour points at the highest resolution (original image) are found. The method runs fast and has been successful in extracting contours from MRI and CT images. We illustrate the method and its performance using MRI and CT images of the human brain.     

A comparative analysis of several transformations for enhancement and segmentation of magnetic resonance image scene sequences

The performance of the eigenimage filter is compared with those of several other filters as applied to magnetic resonance image (MRI) scene sequences for image enhancement and segmentation. Comparisons are made with principal component analysis, matched, modified-matched, maximum contrast, target point, ratio, log-ratio, and angle image filters. Signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), segmentation of a desired feature (SDF), and correction for partial volume averaging effects (CPV) are used as performance measures. For comparison, analytical expressions for SNRs and CNRs of filtered images are derived, and CPV by a linear filter is studied. Properties of filters are illustrated through their applications to simulated and acquired MRI sequences of a phantom study and a clinical case; advantages and weaknesses are discussed. The conclusion is that the eigenimage filter is the optimal linear filter that achieves SDF and CPV simultaneously.     

Estimating Brain Deformation During Surgery Using Finite Element Method: Optimization and Comparison of Two Linear Models

This paper addresses estimation of brain deformation during craniotomy using finite element modeling. Two mechanical models are optimized and compared for this purpose: linear solid-mechanic model and linear elastic model. Both models assume the realistic finite deformation of the brain after opening the skull. In this study, we use pre-operative and intra-operative magnetic resonance images (MRI) of five patients undergoing brain tumor surgery. Anatomical landmarks are identified by an expert radiologist on MRI and used for the method development and comparison studies. We use tetrahedral finite element meshes and optimize model parameters by minimizing the mean distance between the predicted locations of the anatomical landmarks using the pre-operative images and their actual locations on the intra-operative images. Evaluation of the objective function using a second set of landmarks not used in the optimization process suggests that accuracy of the solid mechanic model is higher than that of the elastic model for our application. Visual inspection of the results confirms this conclusion. The proposed method along with the location information of the surface landmarks measured in the operating room and marked on the pre-operative images can be used to estimate the brain deformations without needing intra-operative images. In this case, since the parameters of the brain tissue are not the same for different patients, the proposed optimization process is crucial for obtaining accurate results.     

A discrete curvature-based deformable surface model with application to segmentation of volumetric images

In this paper, we present a new curvature-based three-dimensional (3-D) deformable surface model. The model deforms under defined force terms. Internal forces are calculated from local model curvature, using a robust method by a least-squares error (LSE) approximation to the Dupin indicatrix. External forces are calculated by applying a step expansion and restoration filter (SEF) to the image data. A solution for one of the most common problems associated with deformable models, self-cutting, has been proposed in this work. We use a principal axis analysis and reslicing of the deformable model, followed by triangulation of the slices, to remedy self-cutting. We use vertex resampling, multiresolution deformation, and refinement of the mesh grid to improve the quality of the model deformation, which leads to better results. Examples of the model application to different cases (simulation, magnetic resonance imaging (MRI), computerized tomography (CT), and ultrasound images) are presented, showing diversity and flexibility of the model.     

Error propagation in eigenimage filtering

Mathematical derivation of error (noise) propagation in eigenimage filtering is presented. Based on the mathematical expressions, a method for decreasing the propagated noise given a sequence of images is suggested. The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of the final composite image are compared to the SNRs and CNRs of the images in the sequence. The consistency of the assumptions and accuracy of the mathematical expressions are investigated using sequences of simulated and real magnetic resonance (MR) images of an agarose phantom and a human brain.