Sedat Ozer

Supervised and unsupervised methods for prostate cancer segmentation with multispectral MRI

PURPOSE Magnetic resonance imaging (MRI) has been proposed as a promising alternative to transrectal ultrasound for the detection and localization of prostate cancer and fusing the information from multispectral MR images is currently an active research area. In this study, the goal is to develop automated methods that combine the pharmacokinetic parameters derived from dynamic contrast enhanced (DCE) MRI with quantitative T2 MRI and diffusion weighted imaging (DWI) in contrast to most of the studies which were performed with human readers. The main advantages of the automated methods are that the observer variability is removed and easily reproducible results can be efficiently obtained when the methods are applied to a test data. The goal is also to compare the performance of automated supervised and unsupervised methods for prostate cancer localization with multispectral MRI. METHODS The authors use multispectral MRI data from 20 patients with biopsy-confirmed prostate cancer patients, and the image set consists of parameters derived from T2, DWI, and DCE-MRI. The authors utilize large margin classifiers for prostate cancer segmentation and compare them to an unsupervised method the authors have previously developed. The authors also develop thresholding schemes to tune support vector machines (SVMs) and their probabilistic counterparts, relevance vector machines (RVMs), for an improved performance with respect to a selected criterion. Moreover, the authors apply a thresholding method to make the unsupervised fuzzy Markov random fields method fully automatic. RESULTS The authors have developed a supervised machine learning method that performs better than the previously developed unsupervised method and, additionally, have found that there is no significant difference between the SVM and RVM segmentation results. The results also show that the proposed methods for threshold selection can be used to tune the automated segmentation methods to optimize results for certain criteria such as accuracy or sensitivity. The test results of the automated algorithms indicate that using multispectral MRI improves prostate cancer segmentation performance when compared to single MR images, a result similar to the human reader studies that were performed before. CONCLUSIONS The automated methods presented here can help diagnose and detect prostate cancer, and improve segmentation results. For that purpose, multispectral MRI provides better information about cancer and normal regions in the prostate when compared to methods that use single MRI techniques; thus, the different MRI measurements provide complementary information in the automated methods. Moreover, the use of supervised algorithms in such automated methods remain a good alternative to the use of unsupervised algorithms.

A set of new Chebyshev kernel functions for support vector machine pattern classification

In this study, we introduce a set of new kernel functions derived from the generalized Chebyshev polynomials. The proposed generalized Chebyshev polynomials allow us to derive different kernel functions. By using these polynomial functions, we generalize recently introduced Chebyshev kernel function for vector inputs and, as a result, we obtain a robust set of kernel functions for Support Vector Machine (SVM) classification. Thus in this study, besides clarifying how to apply the Chebyshev kernel functions on vector inputs, we also increase the generalization capability of the previously proposed Chebyshev kernels and show how to derive new kernel functions by using the generalized Chebyshev polynomials. The proposed set of kernel functions provides competitive performance when compared to all other common kernel functions on average for the simulation datasets. The results indicate that they can be used as a good alternative to other common kernel functions for SVM classification in order to obtain better accuracy. Moreover, test results show that the generalized Chebyshev kernel approaches to the minimum support vector number for classification in general.

Prostate cancer localization with multispectral MRI based on Relevance Vector Machines

Prostate cancer is one of the leading causes of cancer death for men. However, early detection before cancer spreads beyond the prostate can reduce the mortality. Therefore, invivo imaging techniques play an important role to localize the prostate cancer for treatment. Although Magnetic Resonance Imaging (MRI) has been proposed to localize prostate cancer, the studies on automated localization with multispectral MRI have been limited. In this study we propose combining the pharmacokinetic parameters derived from DCE MRI with T2 MRI and DWI. We also propose to use Relevance Vector Machines (RVM) for automatic prostate cancer localization, compare its performance to Support Vector Machines (SVM) and show that RVM can produce more accurate and more efficient segmentation results than SVM for automated prostate cancer localization with multispectral MRI.

Activity Detection in Scientific Visualization

For large-scale simulations, the data sets are so massive that it is sometimes not feasible to view the data with basic visualization methods, let alone explore all time steps in detail. Automated tools are necessary for knowledge discovery, i.e., to help sift through the data and isolate specific time steps that can then be further explored. Scientists study patterns and interactions and want to know when and where interesting things happen. Activity detection, the detection of specific interactions of objects which span a limited duration of time, has been an active research area in the computer vision community. In this paper, we introduce activity detection to scientific simulations and show how it can be utilized in scientific visualization. We show how activity detection allows a scientist to model an activity and can then validate their hypothesis on the underlying processes. Three case studies are presented.

Group dynamics in scientific visualization

The ability to visually extract and track features is appealing to scientists in many simulations including flow fields. However, as the resolution of the simulation becomes higher, the number of features to track increases and so does the cost in large-scale simulations. Since many of these features act in groups, it seems more cost-effective to follow groups of features rather than individual ones. Very little work has been done for tracking groups of features. In this paper, we present the first full group tracking framework in which we track groups (clusters) of features in time-varying 3D fluid flow simulations. Our framework uses a clustering algorithm to group interacting features. We demonstrate the use of our framework on data output from a 3D simulation of wall bounded turbulent flow.

Generalized Chebyshev Kernels for Support Vector Classification

In this paper, a method to generalize previously proposed Chebyshev kernel function is presented for support vector classification in order to obtain more robust and higher classification accuracy. By introducing the generalized Chebyshev polynomials for vector inputs, we increase the performance of this kernel function. The simulation results show that the proposed generalized Chebyshev kernel has better performance than the previously proposed kernel for support vector classification. Early simulation results show that the proposed kernel function yields the best classification results for a breast cancer dataset.

Activity Detection for scientific visualization

Understanding the science behind ultra-scale simulations requires extracting meaning from data sets of hundreds of terabytes or more. At extreme scales, the data sets are so huge, there is not even enough time to view the data, let alone explore it with basic visualization methods. Automated tools are necessary for knowledge discovery to help sift through the information and isolate characteristic patterns, thereby enabling the scientist to study local interactions, the origin of features, and their evolution, i.e. activity detection in large volumes of 3D data. Defining and modelling such activities in 3D scientific data sets remains an open research problem, though it has been widely studied in the computer vision community. In this work we demonstrate how utilizing activity detection can help us model and detect complex events (activities) in large 3D scientific data sets. We employ Petri nets which support distributed and discrete graphical modelling of spatio-temporal patterns to model activities in time-varying 3D scientific data sets. We demonstrate the use of Petri nets on three different data sets.

Image classification by multi-kernel dictionary learning

Recent studies have indicated the efficacy of selecting and combining the salient features from a pool of feature types in image retrieval and classification applications. In contrast to previous work, in this paper, we approach this problem as a selection and combination of the salient feature type(s) from a pool of feature types rather than selecting an individual feature. Our approach utilizes multiple kernels within the dictionary-learning framework where a combination of dictionary atoms represents individual categories. The category specific feature combination parameters or weights for kernel combination are determined by the mutual information techniques. The method is compared to a meta-algorithm for feature nomination. The multi-kernel dictionary learning method yields, on average, a 10% increase in classification accuracy with respect to the meta-algorithm in our preliminary experiments.

An interactive method for activity detection visualization

Visualizing each time step in an activity from a scientific dataset can aid in understanding the data and phenomena. In this work, we present a Graphical User Interface (GUI) that allows scientists to first graphically model an activity, then detect any activities that match the model, and finally visualize the detected activities in time varying scientific data sets. As a graphical and state based interactive approach, an activity detection framework is implemented by our GUI as a tool for modelling, hypothesis-testing and searching for interested activities from the phenomena evolution of the data set. We demonstrate here some features of our GUI: a histogram is used to visualize the number of activities detected as a function of time and to allow the user to focus on a moment in time; a table is used to give details about the activities and the features participating in them; and finally the user is given the ability to click on the screen to bring up 3D images of the overall activity sequence, single time steps of an activity, or individual feature in an activity. We present examples from applications to two different data sets.

Support vector regression for surveillance purposes

This paper addresses the problem of applying powerful statistical pattern classification algorithm based on kernel functions to target tracking on surveillance systems. Rather than directly adapting a recognizer, we develop a localizer directly using the regression form of the Support Vector Machines (SVM). The proposed approach considers to use dynamic model together as feature vectors and makes the hyperplane and the support vectors follow the changes in these features. The performance of the tracker is demonstrated in a sensor network scenario with a constant velocity moving target on a plane for surveillance purpose.