Yue Joseph Wang

Bio: Yue Joseph Wang is an academic researcher from Virginia Tech. The author has contributed to research in topics: Image registration & Visualization. The author has an hindex of 16, co-authored 47 publications receiving 599 citations. Previous affiliations of Yue Joseph Wang include Children's National Medical Center & The Catholic University of America.

A blockwise relaxation labeling scheme and its application to edge detection in cardiac mr image sequences

Abstract: We present a segmentation scheme for magnetic resonance (MR) image sequences based on vector quantization of a block-partitioned image followed by a relaxation labeling procedure. By first searching a coarse segmentation, the algorithm yields very fast and effective performance on images that are inherently noisy, and can effectively use the correlation in a sequence of images for robust performance and efficient implementation. The algorithm defines feature vectors by the local histogram on a block-partioned image and approximates the local histograms by normal distributions. The relative entropy is chosen as the meaningful distance measure between the feature vectors and the templates. After initial computation of the normal distribution parameters, a blockwise classification maximization algorithm classifies blocks in the block-partitioned image by minimizing their relative entropy distance for a coarse-resolution segmentation; and finally, finer resolution is obtained by contextual Bayesian relaxation labeling in which label update is performed pixelwise by incorporating neighborhood information. Sequence processing is then performed to segment all images in the sequence. The scheme is applied to left ventricular boundary detection in short-axis MR image sequences, and results are presented to show that the algorithm successfully extracts the endocardial contours and that sequence processing significantly improves edge detection performance and can avoid local minima problems. © 1998 John Wiley & Sons, Inc. Int J Imaging Syst Technol, 9, 340–350, 1998

Multimodality medical image fusion: probabilistic quantification, segmentation, and registration

Abstract: Multimodality medical image fusion is becoming increasingly important in clinical applications, which involves information processing, registration and visualization of interventional and/or diagnostic images obtained from different modalities This work is to develop a multimodality medical image fusion technique through probabilistic quantification, segmentation, and registration, based on statistical data mapping, multiple feature correlation, and probabilistic mean ergodic theorems The goal of image fusion is to geometrically align two or more image areas/volumes so that pixels/voxels representing the same underlying anatomical structure can be superimposed meaningfully Three steps are involved To accurately extract the regions of interest, we developed the model supported Bayesian relaxation labeling, and edge detection and region growing integrated algorithms to segment the images into objects After identifying the shift-invariant features (ie, edge and region information), we provided an accurate and robust registration technique which is based on matching multiple binary feature images through a site model based image re-projection The image was initially segmented into specified number of regions A rough contour can be obtained by delineating and merging some of the segmented regions We applied region growing and morphological filtering to extract the contour and get rid of some disconnected residual pixels after segmentation The matching algorithm is implemented as follows: (1) the centroids of PET/CT and MR images are computed and then translated to the center of both images (2) preliminary registration is performed first to determine an initial range of scaling factors and rotations, and the MR image is then resampled according to the specified parameters (3) the total binary difference of the corresponding binary maps in both images is calculated for the selected registration parameters, and the final registration is achieved when the minimum number of mismatch pixels gives the optimal scaling factor and rotation angle Cross-modality quantification is then performed by incorporating the probabilistic pixel memberships from one modality (eg, MRI) and the functional activities from another modality (eg, PET or CT) within any given region of interests The consistent nature of the comparative clinical studies show that our matching technique result in a robust and accurate image fusion of MRI and PET/CT scans

Patient site model supported change detection

Abstract: This paper reports the development of a non-rigid registration technique to bring into alignment a sequence of a patient's single-view mammograms acquired at different times. This technique is applied in a patient site model supported change detection algorithm with a clinical goal of lesion detection and tracking. The algorithm flow contains four steps: preprocessing, image alignment, change detection, and site model updating. The preprocessing step includes segmentation, using standard finite normal mixture and Markov random field models, morphological processing, monotony operators, and Gaussian filtering. The site model in this research is composed of object boundaries, previous change, potential control points, and raw/segmented images. In the alignment step, the current mammogram is aligned to the site model using a two step process consisting of principle axis of the skin line followed by thin-plate spline using matched points from the potential control point pool. With the assumption of minimal global change, subtraction and thresholding will be used to create the change map that highlights significant changes. Finally, the change information will be used to update the site model. This two-step registration process facilitates change detection by aligning corresponding regions of mammograms so local change analysis can be performed in a coherent manner. The result of the change detection algorithm will be a local change and a patient specific site model showing past and present conditions.

Detection of mammographic masses using sector features with a multiple-circular-path neural network

Abstract: In the clinical course of detecting masses, mammographers usually evaluate the surrounding background of a radiodense when breast cancer is suspected. In this study, we adapted this fundamental concept and computed features of the suspicious region in radial sections. These features were then arranged by circular convolution processes within a neural network, which led to an improvement in detecting mammographic masses. In this experiment, randomly selected mammograms were processed by morphological enhancement techniques. Radiodense areas were isolated and were delineated using the region growing algorithm with a valley blocking technique. The boundary of each region of interest was then divided into 36 sectors using 36 equi-angle dividers radiated from the center of the area. Four features at each section were computed: (1) the radius, (2) the normal angle of the boundary, (3) the average gradient along the radial direction, and (4) the gray value difference (i.e., contrast) along the radial direction. Hence, 144 computed features (i.e., 4 features per sector for 36 sectors) were used as input values for the newly designed multiple circular path neural network (MCPNN). The neural network is constructed to emphasize on the correlation information associated with the feature interactions within the angle and between adjacent angles. We have tested this approach on our research database consisting of 91 mammograms. The over-all performance in the detection of masses was 0.78-0.80 for the areas (Az) under the ROC curves using the conventional neural network. However, the performance was improved to Az values of 0.84-0.89 using the multiple circular path neural network.

Applications of Support Vector Machine (SVM) Learning in Cancer Genomics.

Abstract: Machine learning with maximization (support) of separating margin (vector), called support vector machine (SVM) learning, is a powerful classification tool that has been used for cancer genomic classification or subtyping. Today, as advancements in high-throughput technologies lead to production of large amounts of genomic and epigenomic data, the classification feature of SVMs is expanding its use in cancer genomics, leading to the discovery of new biomarkers, new drug targets, and a better understanding of cancer driver genes. Herein we reviewed the recent progress of SVMs in cancer genomic studies. We intend to comprehend the strength of the SVM learning and its future perspective in cancer genomic applications.

Copy number analysis indicates monoclonal origin of lethal metastatic prostate cancer

Abstract: Many studies have shown that primary prostate cancers are multifocal and are composed of multiple genetically distinct cancer cell clones. Whether or not multiclonal primary prostate cancers typically give rise to multiclonal or monoclonal prostate cancer metastases is largely unknown, although studies at single chromosomal loci are consistent with the latter case. Here we show through a high-resolution genome-wide single nucleotide polymorphism and copy number survey that most, if not all, metastatic prostate cancers have monoclonal origins and maintain a unique signature copy number pattern of the parent cancer cell while also accumulating a variable number of separate subclonally sustained changes. We find no relationship between anatomic site of metastasis and genomic copy number change pattern. Taken together with past animal and cytogenetic studies of metastasis and recent single-locus genetic data in prostate and other metastatic cancers, these data indicate that despite common genomic heterogeneity in primary cancers, most metastatic cancers arise from a single precursor cancer cell. This study establishes that genomic archeology of multiple anatomically separate metastatic cancers in individuals can be used to define the salient genomic features of a parent cancer clone of proven lethal metastatic phenotype.

Integrative approaches for finding modular structure in biological networks

Abstract: A central goal of systems biology is to elucidate the structural and functional architecture of the cell. To this end, large and complex networks of molecular interactions are being rapidly generated for humans and model organisms. A recent focus of bioinformatics research has been to integrate these networks with each other and with diverse molecular profiles to identify sets of molecules and interactions that participate in a common biological function - that is, 'modules'. Here, we classify such integrative approaches into four broad categories, describe their bioinformatic principles and review their applications.

Bibliography of Self-Organizing Map SOM) Papers: 1998-2001 Addendum

Abstract: The Self-Organizing Map (SOM) algorithm has attracted a great deal of interest among researches and practitioners in a wide variety of fields. The SOM has been analyzed extensively, a number of variants have been developed and, perhaps most notably, it has been applied extensively within fields ranging from engineering sciences to medicine, biology, and economics. We have collected a comprehensive list of 5384 scientific papers that use the algorithms, have benefited from them, or contain analyses of them. The list is intended to serve as a source for literature surveys. The present addendum contains 2092 new articles, mainly from the years 1998-2002. We have provided a keyword index to help finding articles of interest, and additionally a modern automatically constructed variant of a thematic index: a WEBSOM interface to the whole article collection of years 1981-2000. The SOM of SOMs is available at http://websom.hut.fi/websom/somref/search.cgi for browsing and searching the collection.