Shantanu Singh

Spatio-temporal models of mental processes from fMRI

Understanding the highly complex, spatially distributed and temporally organized phenomena entailed by mental processes using functional MRI is an important research problem in cognitive and clinical neuroscience. Conventional analysis methods focus on the spatial dimension of the data discarding the information about brain function contained in the temporal dimension. This paper presents a fully spatio-temporal multivariate analysis method using a state-space model (SSM) for brain function that yields not only spatial maps of activity but also its temporal structure along with spatially varying estimates of the hemodynamic response. Efficient algorithms for estimating the parameters along with quantitative validations are given. A novel low-dimensional feature-space for representing the data, based on a formal definition of functional similarity, is derived. Quantitative validation of the model and the estimation algorithms is provided with a simulation study. Using a real fMRI study for mental arithmetic, the ability of this neurophysiologically inspired model to represent the spatio-temporal information corresponding to mental processes is demonstrated. Moreover, by comparing the models across multiple subjects, natural patterns in mental processes organized according to different mental abilities are revealed.

Activity Analysis Using Spatio-Temporal Trajectory Volumes in Surveillance Applications

In this paper, we present a system to analyze activities and detect anomalies in a surveillance application, which exploits the intuition and experience of security and surveillance experts through an easy- to-use visual feedback loop. The multi-scale and location specific nature of behavior patterns in space and time is captured using a wavelet-based feature descriptor. The system learns the fundamental descriptions of the behavior patterns in a semi-supervised fashion by the higher order singular value decomposition of the space described by the training data. This training process is guided and refined by the users in an intuitive fashion. Anomalies are detected by projecting the test data into this multi-linear space and are visualized by the system to direct the attention of the user to potential problem spots. We tested our system on real-world surveillance data, and it satisfied the security concerns of the environment.

Non parametric cell nuclei segmentation based on a tracking over depth from 3D fluorescence confocal images

3D cell nuclei segmentation from fluorescence microscopy images is a key application in many biological studies. We propose a new, fully automated and non parametric method that takes advantage of the resolution anisotropy in fluorescence microscopy. The cell nuclei are first detected in 2D at each image plane and then tracked over depth through a graph based decision to recover their 3D profiles. As the tracking fails to separate very close cell nuclei along depth, we also propose a corrective step based on an intensity projection criterion. Experimental results on real data demonstrate the efficacy of the proposed method.

Non-parametric population analysis of cellular phenotypes

Methods to quantify cellular-level phenotypic differences between genetic groups are a key tool in genomics research. In disease processes such as cancer, phenotypic changes at the cellular level frequently manifest in the modification of cell population profiles. These changes are hard to detect due the ambiguity in identifying distinct cell phenotypes within a population. We present a methodology which enables the detection of such changes by generating a phenotypic signature of cell populations in a data-derived feature-space. Further, this signature is used to estimate a model for the redistribution of phenotypes that was induced by the genetic change. Results are presented on an experiment involving deletion of a tumor-suppressor gene dominant in breast cancer, where the methodology is used to detect changes in nuclear morphology between control and knockout groups.

State-space models of mental processes from fMRI

In addition to functional localization and integration, the problem of determining whether the data encode some information about the mental state of the subject, and if so, how this information is represented has become an important research agenda in functional neuroimaging. Multivariate classifiers, commonly used for brain state decoding, are restricted to simple experimental paradigms with a fixed number of alternatives and are limited in their representation of the temporal dimension of the task. Moreover, they learn a mapping from the data to experimental conditions and therefore do not explain the intrinsic patterns in the data. In this paper, we present a data-driven approach to building a spatio-temporal representation of mental processes using a state-space formalism, without reference to experimental conditions. Efficient Monte Carlo algorithms for estimating the parameters of the model along with a method for model-size selection are developed. The advantages of such a model in determining the mental-state of the subject over pattern classifiers are demonstrated using an fMRI study of mental arithmetic.

Identifying nuclear phenotypes using semi-supervised metric learning

In systems-based approaches for studying processes such as cancer and development, identifying and characterizing individual cells within a tissue is the first step towards understanding the large-scale effects that emerge from the interactions between cells. To this end, nuclear morphology is an important phenotype to characterize the physiological and differentiated state of a cell. This study focuses on using nuclear morphology to identify cellular phenotypes in thick tissue sections imaged using 3D fluorescence microscopy. The limited label information, heterogeneous feature set describing a nucleus, and existence of subpopulations within cell-types makes this a difficult learning problem. To address these issues, a technique is presented to learn a distance metric from labeled data which is locally adaptive to account for heterogeneity in the data. Additionally, a label propagation technique is used to improve the quality of the learned metric by expanding the training set using unlabeled data. Results are presented on images of tumor stroma in breast cancer, where the framework is used to identify fibroblasts, macrophages and endothelial cells--three major stromal cells involved in carcinogenesis.

Analysis of spatial variation of nuclear morphology in tissue microenvironments

We present a study of the spatial variation of nuclear morphology of stromal and cancer-associated fibroblasts in the mouse mammary gland. The work is part of a framework being developed for the analysis of the tumor microenvironment in breast cancer. Recent research has uncovered the role of stromal cells in promoting tumor growth and progression. In specific, studies have indicated that stromal fibroblasts - formerly considered to be passive entities in the extra-cellular matrix - play an active role in the progression of tumor in mammary tissue. We have focused on the analysis of the nuclear morphology of fibroblasts, which several studies have shown to be a critical phenotype in cancer. An essential component of our approach is that the nuclear morphology is studied within the 3D spatial context of the tissue, thus enabling us to pose questions about how the locus of a cell relates to its morphology, and possibly to its function. In order to make quantitative comparisons between nuclear populations, we build statistical shape models of cell populations and infer differences between the populations through these models. We present our observation on both normal and tumor tissues from the mouse mammary gland.

Microstructure preserving synthesis of biomedical images

We present an approach for synthesizing biological tissue textures from existing tissue samples. At microscopic resolution, tissues are characterized by a spatial arrangement of nuclei, cytoplasm, red-blood cells (RBCs) and adipose components etc. We employ 2-point correlation functions (2-pcfs) to encode the geometrical aspects of component arrangements in the synthesis process. The 2-pcfs belong to a class of neighborhood density estimators and were recently introduced in microscopic image analysis. We provide examples of their application toward synthesis of histology-based tissue textures. We show that our methods retain properties such as component volume fractions, sizes and density in comparison to standard approaches. Our methods are also shown to improve the performance of segmentation algorithms by automatically generating labeled texture classes.