Sayan Ghosal’s paper accepted to SPIE Medical Imaging

Title: A Generative-Predictive Framework to Capture Altered Brain Activity in fMRI and its Association with Genetic Risk: Application to Schizophrenia

Abstract: We present a generative-predictive framework that captures the differences in regional brain activity between a neurotypical cohort and a clinical population, as guided by patient-specific genetic risk. Our model assumes that the functional activations in the neurotypical subjects are distributed around a population mean, and that the altered brain activity in neuropsychiatric patients is dened via deviations from this neurotypical mean. We employ group sparsity to identify a set of brain regions that simultaneously explain the salient functional differences and specify a set of basis vectors, which span the low dimensional data subspace. The patient-specific projections onto this subspace are used as feature vectors to identify multivariate associations with genetic risk. We have evaluated our model on a task-based fMRI dataset from a population study of schizophrenia. We compare our model with two baseline methods, LASSO regression and Random Forest (RF) regression, where we establish a direct association between the brain activity during a working memory task and a schizophrenia polygenetic risk score. Our model demonstrates greater consistency and robustness across bootstrapping experiments than the machine learning baselines. Moreover, the differential activation in the set of brain regions implicated by our model underlie the well documented executive cognitive deficits in schizophrenia.

Naresh Nandakumar’s paper accepted to CNI!

Title: Defining Patient Specific Functional Parcellations in Lesional Cohorts via Markov Random Fields

Abstract: We propose a hierarchical Bayesian model that refines a population-based atlas using resting-state fMRI (rs-fMRI) coherence. Our method starts from an initial parcellation and then iteratively reassigns the voxel memberships at the subject level. Our algorithm uses a maximum a posteriori inference strategy based on the neighboring voxel assignments and the Pearson correlation coefficients between the voxel time series and the parcel reference signals. Our method is generalizable to different initial atlases, ensures spatial and temporal contiguity in the final network organization, and can handle subjects with brain lesions, whose rs-fMRI data varies tremendously from that of a healthy cohort. We validate our method by comparing the intra-network cohesion and the motor network identification against two baselines: a standard functional parcellation with no reassignment and a recently published method with a purely data-driven reassignment procedure. Our method outperforms the original functional parcellation in intra-network cohesion and both methods in motor network identification.

Niharika Shimona D’Souza and Jeff Craley have papers accepted to MICCAI!

Title: A Generative-Discriminative Basis Learning Framework to Predict Clinical Severity from Resting State Functional MRI Data

Abstract: We propose a matrix factorization technique that decomposes the resting state fMRI (rs-fMRI) correlation matrices for a patient population into a sparse set of representative subnetworks, as modeled by rank one outer products. The subnetworks are combined using patient specific non-negative coefficients; these coefficients are also used to model, and subsequently predict the clinical severity of a given patient via a linear regression. Our generative-discriminative framework is able to exploit the structure of rs-fMRI correlation matrices to capture group level effects, while simultaneously accounting for patient variability. We employ ten fold cross validation to demonstrate the predictive power of our model on a cohort of fifty eight patients diagnosed with Autism Spectrum Disorder. Our method outperforms classical semi-supervised frameworks, which perform dimensionality reduction on the correlation features followed by non-linear regression to predict the clinical scores.

 

Title: A Novel Method for Epileptic Seizure Detection Using Coupled Hidden Markov Models 

Abstract: We propose a novel Coupled Hidden Markov Model to detect epileptic seizures in multichannel electroencephalography (EEG) data. Our model defines a network of seizure propagation paths to capture both the temporal and spatial evolution of epileptic activity. To address the intractability introduced by the coupled interactions, we derive a variational inference procedure to efficiently infer the seizure evolution from spectral patterns in the EEG data. We validate our model on EEG acquired under clinical conditions in the Epilepsy Monitoring Unit of the Johns Hopkins Hospital. Using 5-fold cross validation, we demonstrate that our model outperforms three baseline approaches which rely on a classical detection framework. Our model also demonstrates the potential to localize seizure onset zones in focal epilepsy.

NSA Lab hosts Prof. Paul Thompson for the William B. Kouwenhoven Memorial Lecture

Please join the Whiting School of Engineering and the Johns Hopkins Department of Electrical and Computer Engineering  for the William B. Kouwenhoven Memorial Lecture, titled “The ENIGMA Consortium: Mapping Human Brain Diseases with Imaging and Genomics in 50,000 Individuals from 35 Countries,” presented by Dr. Paul Thompson, Director of the ENIGMA Center for Worldwide Medicine, Imaging & Genomics.

Archana Venkataraman installed as the John C. Malone Assistant Professor

April 14, 2017 @ the Johns Hopkins Club, Homewood Campus

Dr. Venkataraman’s research lies at the intersection of multimodal integration, network modeling, and clinical neuroscience. Her goal is to develop a comprehensive and system-level understanding of the brain by strategically combining analytical tools, such as matrix factorization, signal processing, and probabilistic inference, with application-driven hypotheses. This approach promises to yield novel insights into debilitating neurological disorders, with the long-term goal of improving patient care.

Archana Venkataraman to speak at the IEEE JCM in Rochester, NY

April 5, 2017 @ 5:30pm, Rochester Institute of Technology

Title: An Adaptable Framework to Extract Abnormal Brain Networks

Abstract: There is increasing evidence that complex neurological disorders reflect distributed impairments across multiple brain systems. These findings underscore the importance of network-based approaches for functional data. However, network analyses in clinical neuroimaging is largely limited to aggregate measures, which do not pinpoint a concrete etiological mechanism. In contrast, I will present a novel Bayesian framework that captures the underlying topology of the altered functional pathways. I will also highlight some exciting future directions for our methodology that revolve around clinical understanding and interventions.

Archana Venkataraman to speak at the ICM Distinguished Seminar Series

March 7, 2017 @ 11am in Clark Hall 110

Title: An Adaptable Framework to Extract Abnormal Brain Networks

Abstract: There is increasing evidence that complex neurological disorders reflect distributed impairments across multiple brain systems. These findings underscore the importance of network-based approaches for functional data. However, network analyses in clinical neuroimaging is largely limited to aggregate measures, which do not pinpoint a concrete etiological mechanism. In contrast, I will present a novel Bayesian framework that captures the underlying topology of the altered functional pathways. I will also highlight some exciting future directions for our methodology that revolve around clinical understanding and interventions.