Education & Training

  • Postdoctoral fellow 2015-Present

    Department of Neurosurgery, UCLA

  • Ph.D. 2009-2015

    Ph.D. in Bioengineering

    University of Pennsylvania, Department of Bioengineering

  • M.S.2008

    Master of Science in Mechanical engineering and Applied mechanics

    University of Pennsylvania

  • B.E.2006

    Bachelor of Engineering

    University of Mumbai, Mumbai, India

Filter by type:

Sort by year:

Interpreting support vector machine models for multivariate group wise analysis in neuroimaging [Click to download code]

Bilwaj Gaonkar, Russell T Shinohara, Christos Davatzikos
Journal Paper Medical Image Analysis, July 2015

Abstract

Machine learning based classification algorithms like support vector machines (SVMs) have shown great promise for turning a high dimensional neuroimaging data into clinically useful decision criteria. However, tracing imaging based patterns that contribute significantly to classifier decisions remains an open problem. This is an issue of critical importance in imaging studies seeking to determine which anatomical or physiological imaging features contribute to the classifier’s decision, thereby allowing users to critically evaluate the findings of such machine learning methods and to understand disease mechanisms. The majority of published work addresses the question of statistical inference for support vector classification using permutation tests based on SVM weight vectors. Such permutation testing ignores the SVM margin, which is critical in SVM theory. In this work we emphasize the use of a statistic that explicitly accounts for the SVM margin and show that the null distributions associated with this statistic are asymptotically normal. Further, our experiments show that this statistic is a lot less conservative as compared to weight based permutation tests and yet specific enough to tease out multivariate patterns in the data. Thus, we can better understand the multivariate patterns that the SVM uses for neuroimaging based classification.

Automated Tumor Volumetry Using Computer-Aided Image Segmentation [Click to download code]

Bilwaj Gaonkar, Lukasz Macysyzn, Michel Bilello,Mohammed Salehi Sadaghiani, Hamed Akbari, Mark A. Atiah, Zarina S. Ali, Xiao Da, Yiqang Zhan, Donald O’ Rourke, Sean M. Grady, Christos Davatzikos
Journal Paper Academic Radiology, Volume 22, Issue 5, Pages 653-661

Abstract

Accurate segmentation of brain tumors, and quantification of tumor volume, is important for diagnosis, monitoring, and planning therapeutic intervention. Manual segmentation is not widely used because of time constraints. Previous efforts have mainly produced methods that are tailored to a particular type of tumor or acquisition protocol and have mostly failed to produce a method that functions on different tumor types and is robust to changes in scanning parameters, resolution, and image quality, thereby limiting their clinical value. Herein, we present a semiautomatic method for tumor segmentation that is fast, accurate, and robust to a wide variation in image quality and resolution.

Converting Neuroimaging Big Data To Information: Statistical Frameworks For Interpretation Of Image Driven Biomarkers And Image Driven Disease Subtyping

Bilwaj Gaonkar
Dissertation

Abstract

Large scale clinical trials and population based research studies collect huge amounts of neuroimaging data. Machine learning classifiers can potentially use these data to train models that diagnose brain related diseases from individual brain scans. In this dissertation we address two distinct challenges that beset a wider adoption of these tools for diagnostic purposes. The First challenge that besets the neuroimaging based disease classification is the lack of a statistical inference machinery for highlighting brain regions that contribute significantly to the classifier decisions. In this dissertation, we address this challenge by developing an analytic framework for interpreting support vector machine (SVM) models used for neuroimaging based diagnosis of psychiatric disease. To do this we first note that permutation testing using SVM model components provides a reliable inference mechanism for model interpretation. Then we derive our analysis framework by showing that under certain assumptions, the permutation based null distributions associated with SVM model components can be approximated analytically using the data themselves. Inference based on these analytic null distributions is validated on real and simulated data. p-Values computed from our analysis can accurately identify anatomical features that differentiate groups used for classifier training. Since the majority of clinical and research communities are trained in understanding statistical p-values rather than machine learning techniques like the SVM, we hope that this work will lead to a better understanding SVM classifiers and motivate a wider adoption of SVM models for image based diagnosis of psychiatric disease. A second deficiency of learning based neuroimaging diagnostics is that they implicitly assume that, `a single homogeneous pattern of brain changes drives population wide phenotypic differences'. In reality it is more likely that multiple patterns of brain deficits drive the complexities observed in the clinical presentation of most diseases. Understanding this heterogeneity may allow us to build better classifiers for identifying such diseases from individual brain scans. However, analytic tools to explore this heterogeneity are missing. With this in view, we present in this dissertation, a framework for exploring disease heterogeneity using population neuroimaging data. The approach we present first computes difference images by comparing matched cases and controls and then clusters these differences. The cluster centers define a set of deficit patterns that differentiates the two groups. By allowing for more than one pattern of difference between two populations, our framework makes a radical departure from traditional tools used for neuroimaging group analyses. We hope that this leads to a better understanding of the processes that lead to disease and also that it ultimately leads to improved image based disease classifiers.

Automated Tumor Volumetry Using Computer-Aided Image Segmentation [Click to download code]

Bilwaj Gaonkar, Lukasz Macysyzn, Michel Bilello,Mohammed Salehi Sadaghiani, Hamed Akbari, Mark A. Atiah, Zarina S. Ali, Xiao Da, Yiqang Zhan, Donald O’ Rourke, Sean M. Grady, Christos Davatzikos
Journal Paper Academic Radiology, Volume 22, Issue 5, Pages 653-661

Abstract

Accurate segmentation of brain tumors, and quantification of tumor volume, is important for diagnosis, monitoring, and planning therapeutic intervention. Manual segmentation is not widely used because of time constraints. Previous efforts have mainly produced methods that are tailored to a particular type of tumor or acquisition protocol and have mostly failed to produce a method that functions on different tumor types and is robust to changes in scanning parameters, resolution, and image quality, thereby limiting their clinical value. Herein, we present a semiautomatic method for tumor segmentation that is fast, accurate, and robust to a wide variation in image quality and resolution.

Breast DCE-MRI Kinetic Heterogeneity Tumor Markers: Preliminary Associations With Neoadjuvant Chemotherapy Response

Ahmed Ashraf, Bilwaj Gaonkar , Carolyn Mies, Angela DeMichele, Mark Rosen, Christos Davatzikos, Despina Kontos
Journal Paper

Abstract

The ability to predict response to neoadjuvant chemotherapy for women diagnosed with breast cancer, either before or early on in treatment, is critical to judicious patient selection and tailoring the treatment regimen. In this paper, we investigate the role of contrast agent kinetic heterogeneity features derived from breast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for predicting treatment response. We propose a set of kinetic statistic descriptors and present preliminary results showing the discriminatory capacity of the proposed descriptors for predicting complete and non-complete responders as assessed from pre-treatment imaging exams. The study population consisted of 15 participants: 8 complete responders and 7 non-complete responders. Using the proposed kinetic features, we trained a leave-one-out logistic regression classifier that performs with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.84 under the ROC. We compare the predictive value of our features against commonly used MRI features including kinetics of the characteristic kinetic curve (CKC), maximum peak enhancement (MPE), hotspot signal enhancement ratio (SER), and longest tumor diameter that give lower AUCs of 0.71, 0.66, 0.64, and 0.54, respectively. Our proposed kinetic statistics thus outperform the conventional kinetic descriptors as well as the classifier using a combination of all the conventional descriptors (i.e., CKC, MPE, SER, and longest diameter), which gives an AUC of 0.74. These findings suggest that heterogeneity-based DCE-MRI kinetic statistics could serve as potential imaging biomarkers for tumor characterization and could be used to improve candidate patient selection even before the start of the neoadjuvant treatment.

Deformable registration for quantifying longitudinal tumor changes during neoadjuvant chemotherapy

Yangming Ou, Susan P Weinstein, Emily F Conant, Sarah Englander, Xiao Da, Bilwaj Gaonkar, Meng‐Kang Hsieh, Mark Rosen, Angela DeMichele, Christos Davatzikos, Despina Kontos
Journal Paper Magnetic Resonance in Medicine, July 2014

Abstract

The purpose of the study was to evaluate DRAMMS, an attribute-based deformable registration algorithm, compared to other intensity-based algorithms, for longitudinal breast MRI registration, and to show its applicability in quantifying tumor changes over the course of neoadjuvant chemotherapy. I was involved in data preprocessing and interacting with the primary clinician on the project.

Analytic estimation of statistical significance maps for support vector machine based multi-variate image analysis and classification [Click to download code]

Bilwaj Gaonkar, Christos Davatzikos
Journal Paper Neuroimage, Volume 78, September 2013, Pages 270-283

Abstract

Multivariate pattern analysis (MVPA) methods such as support vector machines (SVMs) have been increasingly applied to fMRI and sMRI analyses, enabling the detection of distinctive imaging patterns. However, identifying brain regions that significantly contribute to the classification/group separation requires computationally expensive permutation testing. In this paper we show that the results of SVM-permutation testing can be analytically approximated. This approximation leads to more than a thousandfold speedup of the permutation testing procedure, thereby rendering it feasible to perform such tests on standard computers. The speedup achieved makes SVM based group difference analysis competitive with standard univariate group difference analysis methods.

Multi-Atlas Skull-Stripping

Jimit Doshi, Guray Erus, Yangming Ou,Bilwaj Gaonkar, Christos Davatzikos
Journal Paper Academic Radiology, Volume 20, September 2013, Pages 1566-1576

Abstract

A software package was created that addresses the fundamental need for skull removal as a pre requisite for downstream image processing tasks. We implemented a multi atlas registration based tool for addressing this challenge.

Identifying Multivariate Imaging Patterns: Supervised, Semi-Supervised, and Unsupervised Learning Perspectives

Roman Filipovych, Bilwaj Gaonkar, Christos Davatzikos
Book Chapter Academic Press Library in Signal Processing: Image, Video Processing and Analysis, Hardware, Audio, Acoustic and Speech Processing
Perspectives on image based heterogeneity analysis were provided in this book chapter. Some of our preliminary work on heterogeneity analysis has been published here.

Deriving statistical significance maps for SVM based image classification and group comparisons [Click to download Code]

Bilwaj Gaonkar Christos Davatzikos
Conference Papers Medical Image Computing and Computer-Assisted Intervention– MICCAI 2012, Oral presentation

Abstract

Population based pattern analysis and classification for quantifying structural and functional differences between diverse groups has been shown to be a powerful tool for the study of a number of diseases, and is quite commonly used especially in neuroimaging. The alternative to these pattern analysis methods, namely mass univariate methods such as voxel based analysis and all related methods, cannot detect multivariate patterns associated with group differences, and are not particularly suitable for developing individual-based diagnostic and prognostic biomarkers. A commonly used pattern analysis tool is the support vector machine (SVM). Unlike univariate statistical frameworks for morphometry, analytical tools for statistical inference are unavailable for the SVM. In this paper, we show that null distributions ordinarily obtained by permutation tests using SVMs can be analytically approximated from the data. The analytical computation takes a small fraction of the time it takes to do an actual permutation test, thereby rendering it possible to quickly create statistical significance maps derived from SVMs. Such maps are critical for understanding imaging patterns of group differences and interpreting which anatomical regions are important in determining the classifier's decision.

Pattern based morphometry

Bilwaj Gaonkar,Kilian Pohl, Christos Davatzikos
Conference Papers Medical Image Computing and Computer-Assisted Intervention– MICCAI 2011

Abstract

Voxel based morphometry (VBM) is widely used in the neuroimaging community to infer group differences in brain morphology. VBM is effective in quantifying group differences highly localized in space. However it is not equally effective when group differences might be based on interactions between multiple brain networks. We address this by proposing a new framework called pattern based morphometry (PBM). PBM is a data driven technique. It uses a dictionary learning algorithm to extract global patterns that characterize group differences. We test this approach on simulated and real data obtained from ADNI . In both cases PBM is able to uncover complex global patterns effectively.

Automated segmentation of brain lesions by combining intensity and spatial information

Bilwaj Gaonkar,Erus Guray, Nick Bryan, Christos Davatzikos
Conference Papers IEEE International symposium on biomedical imaging 2010

Abstract

Quantitative analysis of brain lesions in large clinical trials is becoming more and more important. We present a new automated method, that combines intensity based lesion segmentation with a false positive elimination method based on the spatial distribution of lesions. A Support Vector Regressor (SVR) is trained on expert-defined lesion masks using image histograms as features, in order to obtain an initial lesion segmentation. A lesion probability map that represents the spatial distribution of true and false positives on the intensity based segmentation is constructed using the segmented lesions and manual masks. A k-Nearest Neighbor (kNN) classifier based on the lesion probability map is applied to refine the segmentation.

Automated segmentation of cortical necrosis using awavelet based abnormality detection system

Bilwaj Gaonkar, Güray Erus, Kilian M Pohl, Manoj Tanwar, Stefan Margiewicz, R Nick Bryan, Christos Davatzikos
Conference Papers IEEE International symposium on biomedical imaging 2011

Abstract

We propose an automated method to segment cortical necrosis from brain FLAIR-MR Images. Cortical necrosis are regions of dead brain tissue in the cortex caused by cerebrovascular disease (CVD). The accurate segmentation of these regions is difficult as their intensity patterns are similar to the adjoining cerebrospinal fluid (CSF). We generate a model of normal variation using MR scans of healthy controls. The model is based on the Jacobians of warps obtained by registering scans of normal subjects to a common coordinate system. For each patient scan a Jacobian is obtained by warping it to the same coordinate system. Large deviations between the model and subject-specific Jacobians are flagged as `abnormalities'. Abnormalities are segmented as cortical necrosis if they are in the cortex and have the intensity profile of CSF. We evaluate our method by using a set of 72 healthy subjects to model cortical variation. We use this model to successfully detect and segment cortical necrosis in a set of 37 patients with CVD. A comparison of the results with segmentations from two independent human experts shows that the overlap between our approach and either of the human experts is in the range of the overlap between the two human experts themselves.

Classifying medical images using morphological appearance manifold

Erdem Varol, Bilwaj Gaonkar, Christos Davatzikos
Conference Papers IEEE International symposium on biomedical imaging 2013

Abstract

Input features for medical image classification algorithms are extracted from raw images using a series of pre processing steps. One common preprocessing step in computational neuroanatomy and functional brain mapping is the nonlinear registration of raw images to a common template space. Typically, the registration methods used are parametric and their output varies greatly with changes in parameters. Most results reported previously perform registration using a fixed parameter setting and use the results as input to the subsequent classification step. The variation in registration results due to choice of parameters thus translates to variation of performance of the classifiers that depend on the registration step for input. Analogous issues have been investigated in the computer vision literature, where image appearance varies with pose and illumination, thereby making classification vulnerable to these confounding parameters. The proposed methodology addresses this issue by sampling image appearances as registration parameters vary, and shows that better classification accuracies can be obtained this way, compared to the conventional approach.

Deriving Statistical Significance Maps for Support Vector Regression Using Medical Imaging Data

Bilwaj Gaonkar, Aristeidis Sotiras, Christos Davatzikos
Conference Papers Pattern Recognition in Neuroimaging (PRNI), 2013

Abstract

Regression analysis involves predicting a continuos variable using imaging data. The Support Vector Regression (SVR) algorithm has previously been used in addressing regression analysis in neuroimaging. However, identifying the regions of the image that the SVR uses to model the dependence of a target variable remains an open problem. It is an important issue when one wants to biologically interpret the meaning of a pattern that predicts the variable(s) of interest, and therefore to understand normal or pathological process. One possible approach to the identification of these regions is the use of permutation testing. Permutation testing involves 1) generation of a large set of 'null SVR models' using randomly permuted sets of target variables, and 2) comparison of the SVR model trained using the original labels to the set of null models. These permutation tests often require prohibitively long computational time. Recent work in support vector classification shows that it is possible to analytically approximate the results of permutation testing in medical image analysis. We propose an analogous approach to approximate permutation testing based analysis for support vector regression with medical imaging data. In this paper we present 1) the theory behind our approximation, and 2) experimental results using two real datasets.

A Composite Multivariate Polygenic and Neuroimaging Score for Prediction of Conversion to Alzheimer's Disease

Bilwaj Gaonkar, Aristeidis Sotiras, Christos Davatzikos
Conference Papers Pattern Recognition in Neuroimaging (PRNI), 2012

Abstract

Alzheimer's disease (AD) and Mild Cognitive Impairment (MCI) are characterized by widespread pathological changes in the brain. At the same time, Alzheimer's disease is heritable with complex genetic underpinnings that may influence the timing of the related pathological changes in the brain and can affect the progression from MCI to AD. In this paper, we present a multivariate imaging genetics approach for prediction of conversion to Alzheimer's disease in patients with mild cognitive impairment. We employ multivariate pattern recognition approaches to obtain neuroimaging and polygenic discriminators between the healthy individuals and AD patients. We then design, in a linear manner, a composite imaging-genetic score for prediction of conversion to Alzheimer's disease in patients with mild cognitive impairment. We apply our approach within the Alzheimer's Disease Neuroimaging Initiative and show that the integration of polygenic and neuroimaging information improves prediction of conversion to AD.

Adaptive geodesic transform for segmentation of vertebrae on CT images

Bilwaj Gaonkar, Liao Shu, Gerardo Hermosillo, Yiqiang Zhan
Conference Papers SPIE Medical imaging, 2014 (Oral)

Abstract

Vertebral segmentation is a critical first step in any quantitative evaluation of vertebral pathology using CT images. This is especially challenging because bone marrow tissue has the same intensity profile as the muscle surrounding the bone. Thus simple methods such as thresholding or adaptive k-means fail to accurately segment vertebrae. While several other algorithms such as level sets may be used for segmentation any algorithm that is clinically deployable has to work in under a few seconds. To address these dual challenges we present here, a new algorithm based on the geodesic distance transform that is capable of segmenting the spinal vertebrae in under one second. To achieve this we extend the theory of the geodesic distance transforms proposed in1 to incorporate high level anatomical knowledge through adaptive weighting of image gradients. Such knowledge may be provided by the user directly or may be automatically generated by another algorithm. We incorporate information 'learnt' using a previously published machine learning algorithm2 to segment the L1 to L5 vertebrae. While we present a particular application here, the adaptive geodesic transform is a generic concept which can be applied to segmentation of other organs as well. © (2014) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE).

Deep learning in the small sample size setting: cascaded feed forward neural networks for medical image segmentation

Bilwaj Gaonkar,David Hovda, Neil Martin, Luke Macyszyn
Conference Papers SPIE Medical imaging, 2016

Abstract

The task of automatic detection and segmentation of organs or regions of interest is fundamental to medical image analysis. Several machine learning techniques have been proposed for addressing these tasks. However, a key drawback of these methods has been the need of large quantities of training datasets. Generating this training data often requires a substantial amount of effort and time on the part of an experienced clinician. Thus, generating a large amount of such data comes at an incredibly high cost. In stark contrast to machine learning based diagnostic systems, the human visual system works perfectly well with an incredibly limited quantity of data. For instance, first year medical students may look at lungs on a single CT scan, synthesize this information and use it to identify lungs on the very next scan they see. This implies the existence of an extremely fast visual learning system which can learn with extremely limited data. In this paper, we introduced a training algorithm for cascaded deep neural networks which mimic this capacity.

At My Office

You can find me at my office located at University of California at Los Angeles, Department of Neurosurgery, 5th Floor of Wasserman building