Browsing by Author "Roysam, Badrinath"
Now showing 1 - 20 of 31
- Results Per Page
- Sort Options
Item A Visual Query-Driven Search Engine for Brain Tissue Image Analysis(2023-12) Mills, Rachel W; Roysam, Badrinath; Prasad, Saurabh; Mayerich, David; Nguyen, Hien Van; Maric, DraganWe present a versatile multiscale visual search engine for visual query-driven analysis of whole-slide multiplex IHC scans of brain tissue, without the confines, limitations, and programming needs of conventional script-based image analysis. Our unsupervised machine learning-based method adaptively learns the cytoarchitectural characteristics of provided training images, without any human effort or intervention. Then, visual queries can be submitted by indicating individual cell(s), and/or multicellular tissue patch(es) of interest, upon which the search engine retrieves a rank-ordered spatially mapped list of similar other cells or tissue patches based on similar cell morphologies, protein expression, cytoarchitecture, myeloarchitecture, vasculature, etc. Retrievals from multiple queries can be co-analyzed intuitively to generate complex inferences that would otherwise require sophisticated programming. We envision a broad range of uses, e.g., identifying cell populations, discovering cellular/cytoarchitectural similarities and differences across brain regions, delineating brain regions, fitting/refining/building atlases, delineating cortical cell layers, and proofreading automated image analysis results.Item Active and Transfer learning Methods for Computational Histology(2012-12) Padmanabhan, Raghav 1984-; Roysam, Badrinath; Hebert, Thomas J.; Prasad, Saurabh; Labate, Demetrio; Lee, William M.Tissue micro-environments of critical interest like tumors, stem-cell niches, and brain tissue surrounding implanted neuroprosthetic devices are complex in structure and harbor complex processes. Understanding events and perturbations that occur in these micro-environments entails selective molecular imaging of the tissues, delineating cellular structures, and accurate cell classification. The algorithm presented in this thesis advances the state of the art in cell classification in large scale histological studies. The core contribution of this thesis is a novel active machine learning algorithm that leverages the advances made in the fields of optimal experimental design and submodular functions. In large and diverse datasets, manually annotating examples to create a training set is effort intensive and suboptimal due to subjectivity and selection bias introduced by human experts. The proposed algorithm reduces human effort and eliminates subjectivity by actively participating in the learning process to select informative examples for the user to label. The algorithm selects multiple informative examples in a learning iteration reducing the burden of retraining the classifier multiple times. The algorithm relies on the submodularity property of the D-optimal criterion to provide performance guarantees for the examples selected for labeling. The algorithm also obviates the necessity for performing offline analysis for feature selection by using the popular LASSO technique to perform feature selection during training. Our experiments on multiple real world data from clinical studies show that the proposed active learning algorithm outperforms standard learning and other active learning frameworks. Since histological studies involve analysis of similar cells under different conditions, the labeling effort to classify similar or related cells in different tissues or conditions can further be reduced by leveraging knowledge learned from one classification task and using it for a related task. The proposed algorithm is also extended to a transfer learning setting to take advantage of existing labeled data sets even when they are mismatched. When applied in transfer learning mode to endothelial cell classification problems, the algorithm consistently achieves classification accuracies greater than 90% with minimal effort. the The algorithm has been embedded into the open source FARSIGHT toolkit with an intuitive graphical user interface that provides constant feedback about the classification process to the user.Item Active Learning Methods for Computational Delineation and Cellular Profiling of Cortical Layers in Rat Whole Brain Sections using Multiplex Immunofluorescent Imaging and Classification of Histopathological Images using Convolutional Neural Networks(2021-05) Singh, Aditi; Roysam, Badrinath; Prasad, Saurabh; Maric, Dragan; Nguyen, Hien Van; Mayerich, DavidMost machine learning algorithms require an abundance of high-quality training data. Such a requirement creates a major obstacle when using machine learning in the medical image domain, as labeled data collection is difficult. We explore active learning solutions for cortical layer delineation and for training Convolutional Neural Networks with less amount of labeled data. For the former, we develop an objective and automated multiplex imaging-based method for delineation of cortical layers in the whole brain sections. This is an advance over current methods where layers are visually delineated by biologists. We further carryout comprehensive and quantitative profiling of the cell layers with respect to their composition (presence of neuronal and glial cell types and sub-types), cell-phenotypic status, and the spatial arrangement of cells. Our method is based on spatial cluster analysis of neuronal features using the Dirichlet Process Mixture Model and refined using active machine learning. It is versatile, modular, and readily amenable to visual inspection and proofreading. The accuracy of the computational cortical layer delineation was validated by comparing it to brain sections that were immunostained with layer-specific molecular markers (NECAB1, FOXP1) and by comparison against manual delineation by biologists. We implement our proposed method on healthy rat brains and rat brains with mild traumatic brain injury (mTBI). Our in-depth cellular profiling of the layers allows us to study the patterns of tissue perturbations in the cortex for mTBI brains. We propose whole cell morphological segmentation methods for five different types of cells which allow an in-depth analysis of the cell state activation and spatial distribution. These are also used in neuronal feature extraction for cortical layer delineation. For the second implementation of active learning, we formulate an active deep learning framework to train CNNs with less amount of labeled data. We implement two parallel active learning criteria for the same. We provide extensive experimental results and in-depth analysis to demonstrate the effectiveness of our algorithm on a breast tumor classification problem. We offer active learning solutions for addressing two different problems encountered in whole brain analysis.Item Advances in Raman and Surface-Enhanced Raman Spectroscopy: Instrumentation, Plasmonic Engineering and Biomolecular Sensing(2014-08) Qi, Ji; Shih, Wei-Chuan; Wolfe, John C.; Han, Zhu; Larin, Kirill V.; Willson, Richard C.; Roysam, BadrinathRaman spectroscopy is a powerful technique for label-free molecular sensing and imaging in various fields. High molecular specificity, non-invasive sampling approach and the need for little or no sample preparation make Raman spectroscopy uniquely advantageous compared to other analytical techniques. However, Raman spectroscopy suffers from the intrinsic limitation of weak signal intensity. Therefore, time-sensitive studies such as diagnosis and clinical applications require improving the throughput of Raman instrumentation. Alternatively, surface-enhanced Raman scattering (SERS) improves the sensitivity by 10^6 to 10^14 times, making the weak Raman intensity no longer a limitation. Nevertheless, it is still a big challenge to engineer plasmonic substrates with high SERS enhancement, good uniformity and reproducibility. This thesis presents advances in: (1) Raman instrumentation towards high-throughput, environmental, biological and biomedical analysis; (2) SERS substrates with high enhancement factor (EF), uniformity and reproducibility; (3) biosensing applications including imaging of cell population and detection of biomolecules towards high time efficiency and sensitivity. In Raman instrumentation, we have built a high-throughput line-scan Raman microscope system and a novel parallel Raman microscope based on multiple-point active-illumination and wide-field hyperspectral data collection. Using the line-scan Raman microscope, we have performed chemical imaging of intact biological cells at the cell population level. We have also demonstrated more flexibility and throughput from the active-illumination Raman microscope in rapid chemical identification and screening of micro and nanoparticles and bacterial spores. Both Raman microscopes have been used to evaluate the large-area SERS uniformity of DC-sputtered gold nanoislands, a low-cost means to fabricate plasmonic substrates. In plasmonic engineering, we have introduced patterned nanoporous gold nanoparticles that feature 3-dimensional mesoporous network with pore size on the order of 10 nm throughput the sub-wavelength nanoparticles. We showed that the plasmonic resonance can be tuned by geometrical engineering of either the external nanoparticle size and shape or the nanoporous network. As an example, we have developed disk-shaped entities, also known as nanoporous gold disks (NPGD) with highly uniform and reproducible SERS EF exceeding 10^8. Label-free, multiplexed molecular sensing and imaging has been demonstrated on NPGD substrates. Using the line-scan Raman microscope and the NPGD substrates, we have successfully developed a label-free DNA hybridization sensor at the single-molecule level in microfluidics. We have observed discrete, individual DNA hybridization events by in situ monitoring the hybridization process using SERS. The advances and promising results presented in this thesis demonstrate potential impact in Raman/SERS imaging and sensing in environmental, biological and biomedical applications.Item Annotation-Free Deep Learning of Large-Scale Nuclear Segmentation and Spatial Neighborhood Analysis on Multiplexed Fluorescence Images(2020-12) LI, Xiaoyang Rebecca; Roysam, Badrinath; Nguyen, Hien Van; Prasad, Saurabh; Eriksen, Jason; Maric, DraganDeep neural networks (DNNs) offer state-of-the-art performance for cell nucleus detection and segmentation. However, they require many manual annotations from skilled biologists for robust algorithm training, which is labor-intensive and not easily scalable. We propose an unsupervised expectation driving pipeline for Brain Cell Analysis using noisy Labels with minimal human input. 1) It uses a parametric method to generate noisy labels for cell nuclei and refines through an iterative training process. 2) We introduce a background recovery technique to enhance the detection and estimation of segmentation accuracy, especially in densely packed brain regions. 3) A novel sparse decomposition method is used to identify anomalous cell detections and automatically correct them to improve the accuracy further. We also provide extensive experiments evaluated on both supervised and unsupervised measurements to demonstrate our method's high effectiveness. The results of segmentation can be further used for phenotyping and cell localization. Besides, we proposed a spatial model to analyze the neuron-glia cells neighborhoods by cumulative influence of all neuron-glial pairs from the same circular surrounding area centered at the neuronal nuclei. A fast co-location analysis is applied to profile cell spatial neighborhoods in the healthy brain efficiently. LASSO-based feature selection methods are adopted to reveal their changes in different tissue conditions. Finally, we developed an accurate, fast-speed, and scalable method to align large-scale images from multi-round to pixel-level accuracy.Item Automated Extraction of Brain Cell Layers using Cell Networks and Active Machine Learning(2012-12) Hulikal Somasundar, Vinay 1988-; Roysam, Badrinath; Hebert, Thomas J.; Shah, Shishir KiritThe cortex of the brain consists of layers of cells which differ in morphology and their connectivity to the inner parts of the brain. Different areas of the cortex have different laminar structures based on the functionality of their layers. Hence, delineation of the cortical layers is a major step in the analysis of the cortical activity. The methods proposed in this thesis advance the state-of-the-art in extracting the structural features of the layers and delineating them. The thesis contributes methods and tools to quantify the spatial distributional properties of the neurons such as their density and the distance from the cortical surface using efficient graphical data structures. It also presents the application of an actively trained logistic regression classifier to the problem of delineating the layers by classifying the neurons as belonging different layers based on the morphological and spatial properties. The proposed approach is applied to diverse datasets acquired from different cortical areas and it consistently achieves an accuracy of around 90% in delineating the laminar boundaries. It also proves to perform better than unsupervised learning methods and to be more efficient and accurate compared to unaided manual training of the classifier. The proposed methods are integrated into the FARSIGHT toolkit and can be readily used by experts to analyse the cortical layers.Item Combining Quantum Dot Nanoparticles with Expansion Microscopy Enables Three-Dimensional Wide-Field Super-Resolved Imaging(2022-08-15) Gunawardhana, Loku Kuruppu Arachchige Dona D.; Mayerich, David; Roysam, Badrinath; Reddy, Rohith K.; Ruchhoeft, Paul; Eriksen, JasonBiological tissue is inherently a three-dimensional structure. Therefore, detailed biological investigations require three-dimensional imaging of cells and tissue microstructures. Given the limitations in traditional light microscopy and a lack of photostable labeling, multiplexed imaging of biological molecules with high precision is difficult to achieve. While recent advances in super-resolution microscopy (SRM) address some of the limitations associated with conventional optical imaging, they impose severe limitations on acquisition speed and are technically challenging. Expansion microscopy (ExM) is a recent development in SRM that enables nanoscale imaging on conventional microscopes. ExM is a relatively simple tissue processing technique that relies on embedding a tissue specimen within a swellable polyelectrolyte hydrogel. When the hydrogel expands, the corresponding molecular labels smoothly expand within a diffraction-limited region, significantly improving spatial resolution. While increasing the tissue volume by 4x-20x dramatically improves resolution, confocal imaging becomes impractical for large samples. This is because label de-crowding dilutes fluorescence signal and prolonged exposure results in photobleaching of adjacent fluorophores. This dissertation focuses on overcoming these limitations by developing a fast tissue imaging methodology capable of multiplex three-dimensional imaging at super-resolution. This method combines widefield imaging with quantum dot nanoparticles (Qdots) in ExM labeling and deconvolution for contrast enhancement. Three major contributions leveraging the advancement of ExM in fluorescence imaging are discussed to this end. ExM compatible Qdot labeling for improved photostability, widefield imaging for improved signal-to-noise ratio with fast imaging speed, and deconvolution for further contrast, resolution improvement in three-dimensional volume rendering. The whole research opens the door to fast and relatively simple 3D nanoscale imaging for applications in biology and medicine at low-cost settings.Item Composite Kernel Dimension Reduction for Multi-Source Remote Sensing Image Analysis(2016-12) Yan, Lifeng; Prasad, Saurabh; Roysam, Badrinath; Hebert, Thomas J.Multi-source remote sensing data has the potential to enable robust image analysis. These sources can be images taken from different sensors on the same region of interest, or multiple different types of diverse features extracted from the same sensor. Either way, composite kernel methods can effectively use information from different sources for classification tasks. In recent work, Angular Discriminant Analysis (ADA) was developed as a technique for effective supervised dimensionality reduction of hyperspectral images. In this thesis, we present a composite-kernel variant of locality preserving ADA (CKLADA) for multi-source remote sensing image analysis. Experiments using a dataset that contains hyperspectral imagery (HSI), light detection and ranging (LiDAR) data and extended multi-attribute profile (EMAP) features, and a single-sensor dataset that contains HSI and EMAP features were conducted whose results show that proposed method provides very effective feature reduction for multi-source remotely sensed data.Item Computation of Breast Ptosis from 3D Scans of Torso(2013-08) Li, Danni 1988-; Merchant, Fatima Aziz; Roysam, Badrinath; Ogmen, Haluk; Reece, Gregory P.Ptosis is an important morphological parameter for characterizing breast aesthetics and is frequently used for assessing the outcome of breast surgery. It refers to the extent to which the nipple is lower than the terminus of the inframammary fold (the contour along which the inferior part of the breast attaches to the chest wall). Current clinical assessment of ptosis involves qualitative visualization by observers which is subject to inter- and intra-observer variability. Alternatively, ptosis can be measured anthropometrically from the patient or from clinical photographs, but these methods are error prone. As stereophotography is now finding its niche in clinical breast surgery, in this study we investigated and evaluated the utility of three-dimensional (3D) features such as surface curvature, coronal projection and surface normal for the assessment of breast ptosis using 3D scans of the torso. Experimental results suggest that 3D features are successful for objectively categorizing breast ptosis with high accuracy and precision.Item Computational Analysis of Tissue Remodeling in the Rat Brain after Mild Traumatic Injury(2017-12) Grama, Kedar Balaji; Roysam, Badrinath; Mayerich, David; Ziburkus, Jokubas; Dash, Pramod K.; Shih, Wei-ChuanAnalysis of pathology in the brain at the cellular scale can yield rich insight into the tissue pathology effects of commonly used drug therapies for brain injury. We propose methods to computationally reconstruct whole rat brains slices and analyze cell populations from 2D fluorescence microscopy images. The proposed methods are applied to a group of rats with mild traumatic brain injury and a profile of cell population alterations are presented.Item Computational Image Analysis of Glial Morphology Following Binge-induced Damage and Exercise-driven Recovery(2014-08) Barton, Emily Avalon; Leasure, J. Leigh; Hernandez, Arturo E.; Roysam, BadrinathNeuronal health is dependent upon proper functioning of glial cells. When this support system fails, neurons cannot function properly. Therefore, a more complete understanding of the role of glia in brain health and pathology is vital. Exercise augments the supportive capabilities of glia, which may account for the overall beneficial effect of exercise on brain health. Conversely, binge drinking damages vulnerable corticolimbic structures and causes cognitive impairments. In the present study, we used computational image analysis to examine the effects of binge alcohol consumption and exercise on the glia in the medial prefrontal cortex (mPFC). Twenty-four female Long Evans rats were exposed to a four-day binge before exercising voluntarily for four weeks. Rats were sacrificed 35 days after their last dose of alcohol. The tissue was stained for microglia (Iba1), neurons (NeuN), astrocytes (GFAP) and cell nuclei (DAPI). Fluorescent confocal microscope images were analyzed using FARSIGHT, a computational image analysis toolkit. We found that exercise increased the number of microglia and the amount of GFAP signal surrounding microglia. However, exercise following binge exposure resulted in a reduced number of microglia and stopped the increase of GFAP surrounding microglia. Additionally, the microglia arbors in binged animals did not fan out in all directions, but instead stayed closer together and extended out further; suggesting binge exposure caused a lasting change in microglia reactivity. Together, our results demonstrate enduring changes of binge alcohol consumption in the frontal cortex 35 days after a single binge episode. Furthermore, previous binge exposure appears to reduce the available plastic response of microglia and astrocytes.Item Computational Modeling of Breast Shape Using Spherical Harmonics(2018-05) Cheong, Audrey; Merchant, Fatima Aziz; Roysam, Badrinath; Mayerich, David; Shah, Shishir Kirit; Reece, Gregory P.As the number of cosmetic and reconstructive breast surgeries performed have been steadily increasing over the years, there is a greater need for improved technologies, such as developing a computational three-dimensional breast model. The breast model will be an invaluable tool for surgeons in surgical planning and during clinical consultations with patients in shared decision making. In this dissertation, a breast model using spherical harmonics is presented. A 3D breast surface image is converted to a spherical harmonic (SPHARM) description, which is represented with three sets of coefficients and can be used to reconstruct a computational model of the breast. Our modeling results demonstrate significantly moderate to strong correlations between specific SPHARM coefficients and breast shape descriptors, such as height, width, projection, and ptosis. We employ these correlations to evaluate breast shape across several subjects and to interactively modify breast shape through these coefficients. We tested the robustness of our method to convert breast image data to SPHARM models and performed classification using the SPHARM coefficients on two types of breast reconstructions: transverse rectus abdominis myocutaneous (TRAM) flap reconstruction and implant reconstruction. Additionally, results were shown on predicting reconstructed breasts from SPHARM models based on preoperative breasts. Contributions of this research: A parametric breast model was developed that can (1) accurately and compactly represent breast data with a set of coefficients (shape descriptors), (2) easily adjust breast shape through modeled coefficients, (3) employ modeling approach to evaluate and differentiate different breast shapes, (4) generate template shape models representative of specific breast shape types, (5) perform classification using the SPHARM description, and (6) perform predictive modeling.Item Data-Driven, Label Consistent, Dictionary Learning Methods for Analysis of Biological Datasets(2016-08) Megjhani, Murad; Roysam, Badrinath; Contreras-Vidal, Jose L.; Shih, Wei-Chuan; Mayerich, David; Leasure, J. Leigh; Burks, JaredThe goal of this thesis is to develop a data-driven, label consistent, and dictionary learning based framework that can be applied on a variety of signal analysis problems. Current methods based on analytical models do not adequately take the variability within and across datasets into consideration when designing signal analysis algorithms. This variability can be added as a morphological constraint to improve the signal analysis algorithms. In particular, this work focuses on three different applications: 1) we present a method for large-scale automated three-dimensional (3-D) reconstruction and profiling of microglia populations in extended regions of brain tissue for quantifying arbor morphology, sensing activation states, and analyzing the spatial distributions of cell activation patterns in tissue; this work provided an opportunity to profile the distribution of microglia in the controlled and device implanted brain. 2) we present a novel morphological constrained spectral unmixing (MCSU) algorithm that combines the spectral and morphological cues in the multispectral image data cube to improve the unmixing quality, this work provided an opportunity to identify new therapeutic opportunities for pancreatic ductal adenocarcinoma (PDAC) from the images collected from humans; and finally, 3) we developed a framework to analyze neuronal response from electroencephalography (EEG) datasets acquired from the infants ranging from 6-24 months. We demonstrated that combining different frequency bands from different spatial locations, yields better classification results, instead of the traditional approach where either one or two frequency bands are used. Using an adaptation of Tibshirani’s Sparse Group LASSO algorithm, we uncovered different spatial and bio markers for understanding a human infant’s brain. These bio-markers can be used for developmental stages of infants and further analysis is required to study the clinical aspects of infant’s social and cognitive development. This work establishes the fundamental mathematical basis for the next generation of algorithms that can leverage the morphological cues from the biological datasets. The algorithm has been embedded into the open source FARSIGHT toolkit with an intuitive graphical user interface.Item Deep Learning for Multi-Channel Image Analysis with Applications to Remote Sensing and Biomedicine(2022-05-13) Foroozandeh Shahraki, Farideh; Prasad, Saurabh; Roysam, Badrinath; Mayerich, David; Reddy, Rohith K.; Maric, DraganDeep neural networks are emerging as a popular choice for multi-channel image analysis – compared with other machine learning approaches, they have been shown to be more effective for a variety of applications in hyperspectral (HSI) and multispectral imaging. We focus on application specific nuances and design choices with respect to deploying such networks for robust analysis of hyperspectral and multispectral images. We provide quantitative and qualitative results with a variety of deep learning architectures in remote sensing, biomedical FTIR and multiplex rat brain images. In this work, not only are traditional deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Convolutional-Recurrent Neural Networks (CRNNs) investigated, we also design and develop Graph based convolutional neural networks (GCNs) with applications to hyperspectral images. To optimize GCNs for HSI data, we proposed a new method of adjacency matrix construction (a semi-supervised adjacency matrix) which leverages class specific and cluster specific properties of the underlying imagery data. Finally, we designed a semi-automatic pipeline that utilizes registration and deep semantic segmentation for aligning (fitting) a brain atlas on multiplex rat brain images.Item Fluorescence Signal Correction and Deep Cell Population Profiling Algorithms for Analyzing Multiplex Images of Whole Rat Brain Slices(2019-12) Jahanipour, Jahandar; Roysam, Badrinath; Mayerich, David; Nguyen, Hien Van; Prasad, Saurabh; Maric, DraganThe altered brain needs our urgent attention. While international brain mapping initiatives remain focused on the structure and working of the neuronal networks, conditions like concussion, stroke, Alzheimer's and experimental drug treatments inflict complex and multi-scale brain cellular alterations that deserve to be mapped in a comprehensive manner. At the cellular level, these conditions initiates a complex web of pathological alterations in all the types of brain cells, ranging from individual cells to multi-cellular functional units. These alterations represent a mixture of changes associated with the primary injury and secondary injuries. Many of these alterations can be subtle and/or latent, only discernible by sensing changes in cell morphology and/or the expression and/or intra-cellular distribution of specific molecular markers. Current immunohistochemistry (IHC) methods based on Hematoxylin and Eosin staining (H\&E) or 3–5 channels of fluorescence immunostaining reveal only a fraction of these alterations at a time, miss the many other alterations and side effects that are occurring concurrently, and do not provide quantitative readouts. Having the goal of advancing brain histology for pre-clinical studies, we propose a novel and comprehensive approach to seamlessly integrate highly multiplexed imaging (10-50 channels) with image processing and machine learning techniques. We describe a combination of signal reconstruction, deep cell detection, profiling and high-dimensional data analysis approaches to generate quantitative readouts of cellular alterations at multiple scales ranging from individual cells to multi-cellular units for comparative analysis. The acquired multiplex images are analyzed computationally to extract the specific fluorescent signals of interest while rejecting non-specific intra-channel and inter-channel signals. The reconstructed image data are analyzed using deep neural networks that detect each cell in the montage, and generate the information of cell location, cell type and cell functional status based on uniquely meaningful combinations of the molecular markers. The cellular measurements are exported for visualization and statistical profiling using other software tools. The proposed pipeline has applications in clinical studies and brain system biology. These data can be used for testing hypotheses, screening individual drugs and combination therapies, and initiating system-level studies.Item Functional Tests of Atypical Connectivity in the Autistic Brain using Magnetoencephalography(2013-05) Coskun, Mehmet A. 1985-; Sheth, Bhavin R.; Ogmen, Haluk; Roysam, Badrinath; Francis, David J.; Breitmeyer, Bruno G.Emerging evidence for differences between individuals with autism and neurotypical controls in tactile processing suggests the somatosensory cortex as a promising substrate for identifying differences in functional brain connectivity. Identifying a neural biomarker of ASD can spur novel biological and pharmacological treatments. We tested four candidate neural biomarkers of autism using magnetoencephalography (MEG) to examine the cortical response to passive tactile stimulation of the thumb and index finger of the dominant hand and lip of individuals with and without autism. The following observations can be made: 1) Individuals with autism have bigger brains early in development. We found the distance between the cortical representations of thumb and lip was significantly larger in the autism group, but not for the representations of the index finger and lip. The inhomogeneity in cortical topography is not a straightforward consequence of a bigger brain but does demonstrate abnormality in sensory organization of the autistic brain. 2) Autistic brains have noisy synapses. The hypothesis predicts increased variability in the response to touch. We did not find higher variability in the evoked response to tactile stimulation in autism, arguing against the noisy synapse hypothesis. 3) Lower level of inhibition in brain circuits of autism is a leading biomarker candidate. The amplitude of cortical response to the stimulation of adjacent fingers is governed by the level of cortical inhibition, and is a physiological test. A comparison of the two groups found a difference in the direction opposite that of prediction. We also examined neural adaptation to prolonged stimulation as cortical inhibition levels, at least in part, controls its extent. Contrary to prediction, the somatosensory cortex of individuals with autism adapts to touch to the same extent as control. Thus, our findings do not support reduced cortical inhibition in autism. 4) Another leading candidate biomarker is local overconnectivity. When a digit is stimulated, i.e., touched, its representation in cortex is directly activated; local intracortical connections indirectly activate non-primary cortical representations corresponding to adjacent digits. Local overconnectivity in autism implies higher nonprimary/primary response ratios, which we did not observe. The results were more consistent with local underconnectivity instead.Item High-Throughput Single-Cell Functional and Molecular Profiling of Immune Cells in Cancer Immunotherapy(2015-08) Liadi, Ivan; Varadarajan, Navin; Willson, Richard C.; Donnelly, Vincent M.; Roysam, Badrinath; Zhang, Xiaoliu ShaunImmunotherapy has revolutionized the treatment of cancer and newer approaches including the adoptive transfer of genetically modified T cells reprogrammed to target tumor antigens have shown remarkable responses. Despite their promise, the efficacy of adoptive immunotherapy remains unpredictable due to the heterogeneous nature of the infusion products, patients’ characteristics, treatment regimens, and tumor burdens. Specifically with regards to the T-cell infusion product, there is a need to develop methodologies that allow for definition of potencies to understand the phenotypic, molecular, and functional contribution of infusion products at single-cell level. In the first part of this dissertation, we implemented Timelapse Imaging Microscopy in Nanowell Grids (TIMING) to demonstrate that while CD4+CAR+ (CAR4) cells killed at slower rate, most likely due to lower granzyme B content, they benefited from apoptosis resistance compared to CD8+CAR+ (CAR8) cells. These findings suggest that overall potency of multi-killing should be evaluated together in their context to resist apoptosis. In the second part of this dissertation, we developed single-cell multiplexed platforms comprising beads biosensors for detecting protein secretion, TIMING to monitor motility and cell-cell interactions, and microfluidic qPCR for transcriptional profiling. Analysis of thousands of single-cell interactions for over 5 hours revealed that the integrated behavior of polyfunctional T cells that kill and secrete IFN-γ was similar to those without IFN-γ secretion, suggesting cytolysis to be the dominant determinant of the interaction behavior and that killing enables faster synapse termination. In addition, tracking the speed of these cells by TIMING indicated that serial killer T cells may be identified based on their high out-of-contact basal motility. Transcriptional profiling of these single-cells confirmed that the motile cells expressed increased amounts of perforin and displayed an activated phenotype. In summary, these results highlight the heterogeneity of immune cells and thus, the need for definition of potency prior to infusion. We propose that single-cell platforms as demonstrated here are suitable to uncover the diversity and to help identify optimal functional and molecular biomarkers for applications in the clinic.Item How the Alcohol-Damaged Brain Responds to Exercise(2017-12) Barton, Emily Avalon; Leasure, J. Leigh; Eriksen, Jason; Kosten, Therese A.; Roysam, BadrinathAlcohol use disorders (AUDs) are a public health concern associated with damage to corticolimbic brain regions and cognitive impairments. As the prevalence of AUDs increases, it is necessary to find effective treatments. Exercise is a low-cost adjunctive treatment option that has been investigated. While it provides numerous health and cognitive benefits, the interactive effects of alcohol and exercise on the brain remain largely unexplored. The work presented in this dissertation examines how the timing of exercise following binge alcohol consumption alters exercise-driven plasticity, as well as the interactive effects of binge alcohol and exercise on microglia, behavioral performance, and neural efficiency. Microglia, the immune cells of the brain, were examined because they help heal the damaged brain; however, their activation is also implicated in the neurotoxic effects of chronic alcohol consumption. Exercise beginning seven days after the binge was found to increase the number of microglia displaying a partially activated morphology in the mPFC, an effect that was not seen in animals waiting 35 days to begin exercise. This suggests binge alcohol was priming microglia to be more responsive to exercise. Due to increased microglial priming and susceptibility to alcohol-induced damage in females, sex differences in response to binge alcohol and exercise were also assessed. Binged females had an increase in microglia and had more microglia with a partially activated morphology compared to binged males. Moreover, females had a higher baseline expression of MHC II+ microglia that significantly increased when females both binged and exercised. This increased reactivity of microglia to alcohol in females may contribute to the increased susceptibility of females to alcohol-induced neural damage. Finally, neural efficiency during a spatial learning task was assessed. Binge alcohol reduced neural efficiency in the mPFC of animals on the last day of water maze testing, without impairing spatial learning. Exercise following binge alcohol resulted in perseverative-like behavior during the probe trial and altered patterns of neural activation in the mPFC and DG. We conclude that exercise has a differential effect on microglia reactivity and neural processing while alcohol is still influencing the brain.Item Individual motile CD4+ T cells can participate in efficient multikilling through conjugation to multiple tumor cells(Cancer Immunology Research, 2016-05) Liadi, Ivan; Singh, Harjeet; Romain, Gabrielle; Rey-Villamizar, Nicolas; Merouane, Amine; Adolacion, Jay R.T.; Kebriaei, Partow; Huls, Helen; Qiu, Peng; Roysam, Badrinath; Cooper, Laurence J.N.; Varadarajan, NavinT cells genetically modified to express a CD19-specific chimeric antigen receptor (CAR) for the investigational treatment of B-cell malignancies comprise a heterogeneous population, and their ability to persist and participate in serial killing of tumor cells is a predictor of therapeutic success. We implemented Timelapse Imaging Microscopy in Nanowell Grids (TIMING) to provide direct evidence that CD4+CAR+ T cells (CAR4 cells) can engage in multikilling via simultaneous conjugation to multiple tumor cells. Comparisons of the CAR4 cells and CD8+CAR+ T cells (CAR8 cells) demonstrate that, although CAR4 cells can participate in killing and multikilling, they do so at slower rates, likely due to the lower granzyme B content. Significantly, in both sets of T cells, a minor subpopulation of individual T cells identified by their high motility demonstrated efficient killing of single tumor cells. A comparison of the multikiller and single-killer CAR+ T cells revealed that the propensity and kinetics of T-cell apoptosis were modulated by the number of functional conjugations. T cells underwent rapid apoptosis, and at higher frequencies, when conjugated to single tumor cells in isolation, and this effect was more pronounced on CAR8 cells. Our results suggest that the ability of CAR+ T cells to participate in multikilling should be evaluated in the context of their ability to resist activation-induced cell death. We anticipate that TIMING may be used to rapidly determine the potency of T-cell populations and may facilitate the design and manufacture of next-generation CAR+ T cells with improved efficacy. Cancer Immunol Res; 3(5); 473–82. ©2015 AACR.Item Information Fusion for Multi-Source Data Classification(2015-12) Zhang, Yuhang; Prasad, Saurabh; Contreras-Vidal, Jose L.; Roysam, Badrinath; Labate, Demetrio; Crawford, Melba M.Multi-source data, either from different sensors or disparate features extracted from the same sensor, are often valuable for data analysis due to their potential for providing complementary information. Effective fusion of information from such multi-source data is critical to enhanced and robust interpretation about the underlying classification problem. Nevertheless, multi-source data also bring unique challenges for data processing, e.g., high-dimensional features, lack of compact representation, and insufficient quantity of labeled data. To make the most use of multi-source data and to address the above challenges, in this research, we develop and validate data fusion algorithms on multiple datasets in two active research areas: remote sensing and brain machine interface (BMI). We develop a mixture-of-kernels approach for data fusion, and demonstrate its efficacy at fusion of multi-source data in the kernel space. In the proposed approach, each source of data is represented by a dedicated kernel -- one can then learn a classifier (or an ``optimal'' feature subspace) by optimizing the kernel parameters for maximum discriminative potential. A direct related benefit is that this learning framework provides a natural and automated mechanism to learn weight distributions in the weighted mixture of kernels, that are strongly indicative of strengths and weaknesses of various sources in the underlying multi-source data analysis problem. We illustrate the benefit of this property and apply it to infer the relative importance of different sources of information in a BMI application. Additionally, to save the labor of labeling a large quantity of samples in real world remote sensing applications, an ensemble based multiple kernel active learning framework is proposed to effectively select important unlabeled samples from multi-source data for classification. We also propose a multi-source feature extraction method based on a composite kernel mapping, to project the multi-source data to a lower dimensional subspace for effective feature fusion. Finally, to effectively represent multi-source data in a compact and robust manner, we propose a joint sparse representation model with adaptive locality weights for classification. By adapting the penalty on individual atoms in the dictionary, we show that one can achieve better signal representation and reduce estimation errors. Further, we also develop a kernel variant of the proposed fusion framework, which is conceptually consistent and aligned with the mixture-of-kernels approach developed previously.