Ge Cong

Model-based segmentation of nuclei

Abstract A new approach for the segmentation of nuclei observed with an epi-fluorescence microscope is presented. The proposed technique is model based and uses local feature activities in the form of step-edge segments, roof-edge segments, and concave corners to construct a set of initial hypotheses. These local feature activities are extracted using either local or global operators and corresponding hypotheses are expressed as hyperquadrics. A neighborhood function is defined over these features to initiate the grouping process. The search space is expressed as an assignment matrix with an appropriate cost function to ensure local and neighborhood consistency. Each possible configuration of nucleus defines a path and the path with least overall error is selected for final segmentation. The system is interactive to allow rapid localization of large numbers of nuclei. The operator then eliminates a small number of false alarms and errors in the segmentation process.

Morphometic analysis of TCGA glioblastoma multiforme

BackgroundOur goals are to develop a computational histopathology pipeline for characterizing tumor types that are being generated by The Cancer Genome Atlas (TCGA) for genomic association. TCGA is a national collaborative program where different tumor types are being collected, and each tumor is being characterized using a variety of genome-wide platforms. Here, we have developed a tumor-centric analytical pipeline to process tissue sections stained with hematoxylin and eosin (H&E) for visualization and cell-by-cell quantitative analysis. Thus far, analysis is limited to Glioblastoma Multiforme (GBM) and kidney renal clear cell carcinoma tissue sections. The final results are being distributed for subtyping and linking the histology sections to the genomic data.ResultsA computational pipeline has been designed to continuously update a local image database, with limited clinical information, from an NIH repository. Each image is partitioned into blocks, where each cell in the block is characterized through a multidimensional representation (e.g., nuclear size, cellularity). A subset of morphometric indices, representing potential underlying biological processes, can then be selected for subtyping and genomic association. Simultaneously, these subtypes can also be predictive of the outcome as a result of clinical treatments. Using the cellularity index and nuclear size, the computational pipeline has revealed five subtypes, and one subtype, corresponding to the extreme high cellularity, has shown to be a predictor of survival as a result of a more aggressive therapeutic regime. Further association of this subtype with the corresponding gene expression data has identified enrichment of (i) the immune response and AP-1 signaling pathways, and (ii) IFNG, TGFB1, PKC, Cytokine, and MAPK14 hubs.ConclusionWhile subtyping is often performed with genome-wide molecular data, we have shown that it can also be applied to categorizing histology sections. Accordingly, we have identified a subtype that is a predictor of the outcome as a result of a therapeutic regime. Computed representation has become publicly available through our Web site.

An algebraic solution to surface recovery from cross-sectional contours

Abstract A new approach for reconstruction of 3D surfaces from 2D cross-sectional contours is presented. By using the so-called “equal importance criterion,” we reconstruct the surface based on the assumption that every point in the region contributes equally to the surface reconstruction process. In this context, the problem is formulated in terms of a partial differential equation, and we show that the solution for dense contours (contours in close proximity) can be efficiently derived from the distance transform. In the case of sparse contours, we add a regularization term to ensure smoothness in surface recovery. The approach is also generalized to other types of cross-sectional contours, where the spine may not be a straight line. The proposed technique allows for surface recovery at any desired resolution. The main advantages of our method is that inherent problems due to correspondence, tiling, and branching are avoided. In contrast to existing implicit methods, we find an optimal field function and develop an interpolation method that does not generate any artificial surfaces. We will demonstrate that the computed high-resolution surface is well represented for subsequent geometric analysis. We present results on both synthetic and real data.

Model based segmentation of nuclei

A new approach for segmentation of nuclei observed with an epi-fluorescence microscope is presented. The technique is model based and uses local feature activities such as step-edge segments, roof-edge segments, and concave corners to construct a set of initial hypotheses. These local feature activities are extracted using either local or global operators to form a possible set of hypotheses. Each hypothesis is expressed as a hyperquadric for better stability, compactness, and error handling. The search space is expressed as an assignment matrix with an appropriate cost function to ensure local adjacency, and global consistency. Each possible configuration of a set of nuclei defines a path, and the path with the least error corresponds to best representation. This result is then presented to an operator who verifies and eliminates a small number of errors.

Robust and efficient surface reconstruction from contours

We propose a new approach for surface recovery from planar sectional contours. The surface is reconstructed based on the so-called “equal importance criterion,” which suggests that every point in the region contributes equally to the reconstruction process. The problem is then formulated in terms of a partial differential equation, and the solution is efficiently calculated from distance transformation. To make the algorithm valid for different application purposes, both the isosurface and the primitive representations of the object surface are derived. The isosurface is constructed by means of a partial differential equation, which can be solved iteratively. The traditional distance interpolating method, which was used by several researchers for surface reconstruction, is an approximate solution of the equation. The primitive representations are approximated by Voronoi diagram transformation of the surface space. Isosurfaces have the advantage that subsequent geometric analysis of the object can be easily carried out while primitive representation is easy to visualize. The proposed technique allows for surface recovery at any desired resolution, thus avoiding the inherent problems of correspondence, tiling, and branching.

Shape metamorphism using p-Laplacian equation

We present a new approach for shape metamorphism, which is a process of gradually changing a source shape (known) through intermediate shapes (unknown) into a target shape (known). The problem, when represented with implicit scalar function, is under-constrained, and regularization is needed. Using the p-Laplacian equation (PLE), we generalize a series of regularization terms based on the gradient of the implicit function, and we show that the present methods lack additional constraints for a more stable solution. The novelty of our approach is in the deployment of a new regularization term when p /spl rarr/ /spl infin/ which leads to the infinite Laplacian equation (ILE). We show that ILE minimizes the supremum of the gradient and prove that it is optimal for metamorphism since intermediate solutions are equally distributed along their normal direction. Applications of the proposed algorithm for 2D and 3D objects are demonstrated.

A new regularized approach for contour morphing

In this paper, we propose a new approach for interpolating curves (contour morphing) in time, which is a process of gradually changing a source curve (known) through intermediate curves (unknown) into a target curve (known). The novelty of our approach is in the deployment of a new regularization term and the corresponding Euler equation. Our method is applicable to implicit curve representation and it establishes a relationship between curve interpolation and a two dimensional function. This is achieved by minimizing the supremum of the gradient, which leads to the infinite Laplacian equation (ILE). ILE is optimal in the sense that interpolated curves are equally distributed along their normal direction. We point out that the existing distance field manipulation (DFM) methods are only an approximation to the proposed optimal solution and that the relationship between ILE and DFM is not local as it has been asserted before. The proposed interpolation can also be used to construct multiscale curve representation.

BioSig: a bioinformatic system for studying the mechanism of inter-cell signaling

Mapping inter-cell signaling pathways requires an integrated view of experimental and informatic protocols. BioSig provides the foundation of cataloging inter-cell responses as a function of particular conditioning, treatment, staining, etc. for either in vivo or in vitro experiments. This paper outlines the system architecture, a functional data model for representing experimental protocols, algorithms for image analysis, and the required statistical analysis. The architecture provides remote shared operation of an inverted optical microscope, and couples instrument operation with images acquisition and annotation. The information is stored in an object-oriented database. The algorithms extract structural information such as morphology and organization, and map it to functional information such as inter-cellular responses. An example of usage of this system is included.

Shape from equal thickness contours

A unique imaging modality based on Equal Thickness Contours (ETC) has introduced a new opportunity for 3D shape reconstruction from multiple views. We present a computational framework for representing each view of an object in terms of its object thickness, and then integrating these representations into a 3D surface by algebraic reconstruction. The object thickness is inferred by grouping curve segments that correspond to points of second derivative maxima. At each step of the process, we use some form of regularization to ensure closeness to the original features, as well as neighborhood continuity. We apply our approach to images of a sub-micron crystal structure obtained through a holographic process.

Curve evolution for corner enhancement

A curve evolution approach is proposed for corner enhancement. The constraint is that a new corner should not be generated. Starting from the general geometric heat flow (GGHF), we study under what conditions the GGHF satisfies the scale space causality criteria. The criteria of corner enhancement are also proposed. Then, a new curve evolution scheme which can enhance strong corners, suppress noise and satisfies the scale space criteria is presented.