Jagath C. Rajapakse

Statistical approach to segmentation of single-channel cerebral MR images

A statistical model is presented that represents the distributions of major tissue classes in single-channel magnetic resonance (MR) cerebral images. Using the model, cerebral images are segmented into gray matter, white matter, and cerebrospinal fluid (CSF). The model accounts for random noise, magnetic field inhomogeneities, and biological variations of the tissues. Intensity measurements are modeled by a finite Gaussian mixture. Smoothness and piecewise contiguous nature of the tissue regions are modeled by a three-dimensional (3-D) Markov random field (MRF). A segmentation algorithm, based on the statistical model, approximately finds the maximum a posteriori (MAP) estimation of the segmentation and estimates the model parameters from the image data. The proposed scheme for segmentation is based on the iterative conditional modes (ICM) algorithm in which measurement model parameters are estimated using local information at each site, and the prior model parameters are estimated using the segmentation after each cycle of iterations. Application of the algorithm to a sample of clinical MR brain scans, comparisons of the algorithm with other statistical methods, and a validation study with a phantom are presented. The algorithm constitutes a significant step toward a complete data driven unsupervised approach to segmentation of MR images in the presence of the random noise and intensity inhomogeneities.

Approach and applications of constrained ICA

This work presents the technique of constrained independent component analysis (cICA) and demonstrates two applications, less-complete ICA, and ICA with reference (ICA-R). The cICA is proposed as a general framework to incorporate additional requirements and prior information in the form of constraints into the ICA contrast function. The adaptive solutions using the Newton-like learning are proposed to solve the constrained optimization problem. The applications illustrate the versatility of the cICA by separating subspaces of independent components according to density types and extracting a set of desired sources when rough templates are available. The experiments using face images and functional MR images demonstrate the usage and efficacy of the cICA.

Multiple SVM-RFE for gene selection in cancer classification with expression data

This paper proposes a new feature selection method that uses a backward elimination procedure similar to that implemented in support vector machine recursive feature elimination (SVM-RFE). Unlike the SVM-RFE method, at each step, the proposed approach computes the feature ranking score from a statistical analysis of weight vectors of multiple linear SVMs trained on subsamples of the original training data. We tested the proposed method on four gene expression datasets for cancer classification. The results show that the proposed feature selection method selects better gene subsets than the original SVM-RFE and improves the classification accuracy. A Gene Ontology-based similarity assessment indicates that the selected subsets are functionally diverse, further validating our gene selection method. This investigation also suggests that, for gene expression-based cancer classification, average test error from multiple partitions of training and test sets can be recommended as a reference of performance quality.

Segmentation of Clustered Nuclei With Shape Markers and Marking Function

We present a method to separate clustered nuclei from fluorescence microscopy cellular images, using shape markers and marking function in a watershed-like algorithm. Shape markers are extracted using an adaptive H-minima transform. A marking function based on the outer distance transform is introduced to accurately separate clustered nuclei. With synthetic images, we quantitatively demonstrate the performance of our method and provide comparisons with existing approaches. On mouse neuronal and Drosophila cellular images, we achieved 6%-7% improvement of segmentation accuracies over earlier methods.

ICA with Reference

We present the technique of the ICA with Reference (ICA-R) to extract an interesting subset of independent sources from their linear mixtures when some a priori information of the sources are available in the form of rough templates (references). The constrained independent component analysis (cICA) is extended to incorporate the reference signals that carry some information of the sources as additional constraints into the ICA contrast function. A neural algorithm is then proposed using a Newton-like approach to obtain an optimal solution to the constrained optimization problem. Stability of the convergence and selection of parameters in the learning algorithm are analyzed. Experiments with synthetic signals and real fMRI data demonstrate the efficacy and accuracy of the proposed algorithm.

SVM-RFE With MRMR Filter for Gene Selection

We enhance the support vector machine recursive feature elimination (SVM-RFE) method for gene selection by incorporating a minimum-redundancy maximum-relevancy (MRMR) filter. The relevancy of a set of genes are measured by the mutual information among genes and class labels, and the redundancy is given by the mutual information among the genes. The method improved identification of cancer tissues from benign tissues on several benchmark datasets, as it takes into account the redundancy among the genes during their selection. The method selected a less number of genes compared to MRMR or SVM-RFE on most datasets. Gene ontology analyses revealed that the method selected genes that are relevant for distinguishing cancerous samples and have similar functional properties. The method provides a framework for combining filter methods and wrapper methods of gene selection, as illustrated with MRMR and SVM-RFE methods.

Learning effective brain connectivity with dynamic Bayesian networks.

We propose to use dynamic Bayesian networks (DBN) to learn the structure of effective brain connectivity from functional MRI data in an exploratory manner. In our previous work, we used Bayesian networks (BN) to learn the functional structure of the brain (Zheng, X., Rajapakse, J.C., 2006. Learning functional structure from fMR images. NeuroImage 31 (4), 1601-1613). However, BN provides a single snapshot of effective connectivity of the entire experiment and therefore is unable to accurately capture the temporal characteristics of connectivity. Dynamic Bayesian networks (DBN) use a Markov chain to model fMRI time-series and thereby determine temporal relationships of interactions among brain regions. Experiments on synthetic fMRI data demonstrate that the performance of DBN is comparable to Granger causality mapping (GCM) in determining the structure of linearly connected networks. Dynamic Bayesian networks render more accurate and informative brain connectivity than earlier methods as connectivity is described in complete statistical sense and temporal characteristics of time-series are explicitly taken into account. The functional structures inferred on two real fMRI datasets are consistent with the previous literature and more accurate than those discovered by BN. Furthermore, we study the effects of hemodynamic noise, scanner noise, inter-scan interval, and the variability of hemodynamic parameters on the derived connectivity.

Prediction of protein relative solvent accessibility with a two-stage SVM approach.

Information on relative solvent accessibility (RSA) of amino acid residues in proteins provides valuable clues to the prediction of protein structure and function. A two‐stage approach with support vector machines (SVMs) is proposed, where an SVM predictor is introduced to the output of the single‐stage SVM approach to take into account the contextual relationships among solvent accessibilities for the prediction. By using the position‐specific scoring matrices (PSSMs) generated by PSI‐BLAST, the two‐stage SVM approach achieves accuracies up to 90.4% and 90.2% on the Manesh data set of 215 protein structures and the RS126 data set of 126 nonhomologous globular proteins, respectively, which are better than the highest published scores on both data sets to date. A Web server for protein RSA prediction using a two‐stage SVM method has been developed and is available (http://birc.ntu.edu.sg/∼pas0186457/rsa.html). Proteins 2005.

Segmentation of subcortical brain structures using fuzzy templates.

We propose a novel method to automatically segment subcortical structures of human brain in magnetic resonance images by using fuzzy templates. A set of fuzzy templates of the structures based on features such as intensity, spatial location, and relative spatial relationship among structures are first created from a set of training images by defining the fuzzy membership functions and by fusing the information of features. Segmentation is performed by registering the fuzzy templates of the structures on the test image and then by fusing them with the tissue maps of the test image. The final decision is taken in order to optimize the certainty in the intensity, location, relative position, and tissue content of the structure. Our method does not require specific expert definition of each structure or manual interactions during segmentation process. The technique is demonstrated with the segmentation of five structures: thalamus, putamen, caudate, hippocampus, and amygdala; the performance of the present method is comparable with previous techniques.

Two-stage support vector regression approach for predicting accessible surface areas of amino acids

We address the problem of predicting solvent accessible surface area (ASA) of amino acid residues in protein sequences, without classifying them into buried and exposed types. A two‐stage support vector regression (SVR) approach is proposed to predict real values of ASA from the position‐specific scoring matrices generated from PSI‐BLAST profiles. By adding SVR as the second stage to capture the influences on the ASA value of a residue by those of its neighbors, the two‐stage SVR approach achieves improvements of mean absolute errors up to 3.3%, and correlation coefficients of 0.66, 0.68, and 0.67 on the Manesh dataset of 215 proteins, the Barton dataset of 502 nonhomologous proteins, and the Carugo dataset of 338 proteins, respectively, which are better than the scores published earlier on these datasets. A Web server for protein ASA prediction by using a two‐stage SVR method has been developed and is available (http://birc.ntu.edu.sg/∼pas0186457/asa.html). Proteins 2006.