Perelman School of Medicine at the University of Pennsylvania

Section for Biomedical Image Analysis (SBIA)

participating with CBICA

Research Themes By Methodology

The Section for Biomedical Image Analysis (SBIA), part of the Center of Biomedical Image Computing and Analytics — CBICA, is devoted to the development of computer-based image analysis methods, and their application to a wide variety of clinical research studies. Image analysis methodologies include functional and structural connectomics, radiomics and radiogenomics, machine learning in imaging, image registration, segmentation, population-based statistical analysis. Clinical research studies span a variety of clinical areas and organs, and are performed within a wide network of collaborations from within and outside Penn. They include brain diseases such as Alzheimer's, schizophrenia, autism, and TBI, evaluation of treatment effects in large clinical trials, and precision diagnostics and predictive modeling in breast and brain cancer.


Diffusion Analytics

The diffusion group is developing methods for diffusion analysis that encompass multi-compartment modeling of new diffusion acquisitions, diffusion measures that characterize the data and harmonization of diffusion data acquired across multiple sites. Details of diffusion based connectomics work can be found at….

Functional Connectivity

Several activities in the lab focus on extracting functional units and networks from resting state fMRI datasets, in order to understand functional connectivity from the local scale of regional functional  coherence, to the  scale of functional networks, and finally to the scale of functional connectomic signatures obtained via machine learning at the individual level.

Current Projects:

 — GraSP: A graph-based parcellation tool originally developed for the functional parcellation of the cortex — GraSP

 — Sparse Connectivity Components of Brain Connectomes —  SCPs

 — Identification of subject-specific brain functional networksFunctional Connectivity Analysis


Machine Learning

Since 2004, the lab has placed significant emphasis on the use of machine learning methods, especially in neuroimaging. This work has lead to indices and patterns related to Alzheimer’s disease, Schizophrenia, brain aging, brain development, Autism, sex differences amongst others. Several methods have been explored, including linear and nonlinear SVMs, non-negative factorization methods, generative-discriminative learning and others.

Machine learning offers great promise as a tool for providing individualized diagnostic and prognostic biomarkers. Although various  imaging measures are affected by diseases in various ways, it is rare to find a single measure that provides sufficient sensitivity and specificity when classifying an individual. Machine learning methods can achieve high classification performance of individuals by leveraging the power of multi-variate pattern analysis. Our group has a long-standing engagement in the development and application of machine learning methods for structural and functional MRI[1-4]. Starting from the conventional supervised learning paradigm, such as finding a discriminative pattern that separates patients form controls, we later moved to semi-supervised learning, in which subtypes of disease are elucidated by different imaging patterns [5-8].

Current Projects:

—  Clustering of Heterogeneous Disease Effects via Distribution Matching of Imaging Patterns —  CHIMERA

— Diagnosis and prognosis using brain scans based on robust regional measures — COMPARE

  • Related software — COMPARE (software page)

—  Ensemble Classification

  • A method to train an ensemble of experts (classifiers) on the given training data, using a multitude of features to produce a less complex model with better performance. Project Page: work in progress

— Generative-Discriminative Basis Learning —  GONDOLA

—  Heterogeneity through Discriminative Analysis — HYDRA

— Grassmann manifold learning for discriminant analysis of functional connectivity patterns — Functional connectivity patterns

— Multivariate inference using discriminatively adaptive smoothing — MIDAS

— NMF-based Decomposition — NMF

— Deriving statistical significance maps from support vector machine (SVM) classifiers, in order to elucidate features that significantly contribute to the classification — SVM Significance

—  Identification of structural subnetworks of the brain — Structural Subnetworks

  • We develop advanced subnetwork detection algorithms that can identify subsystems of the brain network with distinct connectivity patterns or distinct biological or behavioral correlates, using machine learning and probabilistic inference techniques

  1. Lao Z, Shen D, Xue Z, Karacali B, Resnick SM, Davatzikos C: Morphological classification of brains via high-dimensional shape transformations and machine learning methods. NeuroImage 2004, 21(1):46-57.
  2. Davatzikos C, Ruparel K, Fan Y, Shen D, Acharyya M, Loughead J, Gur RC, Langleben D: Classifying spatial patterns of brain activity for lie-detection. NeuroImage 2005, 28(3):663-668.
  3. Davatzikos C, Shen DG, Gur RC, Wu XY, Liu DF, Fan Y, Hughett P, Turetsky BI, Gur RE: Whole-brain morphometric study of schizophrenia revealing a spatially complex set of focal abnormalities. Archives of general psychiatry 2005, 62(11):1218-1227.
  4. Davatzikos C, Fan Y, Wu X, Shen D, Resnick SM: Detection of prodromal Alzheimer's disease via pattern classification of magnetic resonance imaging. Neurobiol Aging 2008, 29(4):514-523.
  5. Filipovych R, Davatzikos C: Semi-supervised Pattern Classification of Medical Images: Application to Mild Cognitive Impairment (MCI). NeuroImage 2011, 55(3):1109-1119.
  6. Filipovych R, Resnick S, Davatzikos C: JointMMCC: Joint Maximum-Margin Classification and Clustering of Imaging Data. IEEE transactions on medical imaging 2012, 31(5):1124-1140.
  7. Varol E, Sotiras A, Davatzikos C: HYDRA: Revealing heterogeneity of imaging and genetic patterns through a multiple max-margin discriminative analysis framework. NeuroImage 2016.
  8. Dong A, Honnorat N, Gaonkar B, Davatzikos C: CHIMERA: Clustering of heterogeneous disease effects via distribution matching of imaging patterns. IEEE transactions on medical imaging 2016, 35(2):612-621.



The lab has developed deformable registration methods for normal and abnormal anatomies, the latter incorporating approaches for simultaneously estimating lesions and adapting deformable registration accordingly.

The process of registration of medical images is one of the most fundamental in this field. It arises in various contexts, including when images from different individuals are mapped to a standardized coordinate system or atlas, thereby accounting for inter-individual anatomical variations, or when the scan of a patient is mapped to a later scan at a follow-up examination, in which anatomical change has likely occurred. Our group has a long-standing involvement in deformable registration methods[1-4], with particular emphasis on the use of rich imaging feature vectors as drivers of deformable registration, as well the use of the concept of mutual saliency as a means for weighting registration transformations according to regional confidence in the detected matches.
Current Projects:

— Deformable Registration using Orientation and Intensity — DTI-DROID

— A robust diffusion-based registration method that utilizes local spatial and orientation features.

DTI Warping

— Manifold-based DTI Statistics

DTI-DROID (Software)

— Deformable Registration via Attribute Matching and Mutual-Saliency Weighting —  DRAMMS

— Pre-Operative and post-Recurrence brain Tumor Registration — PORTR

— Spatial Alignment of fMRI data — fMRI

  1. Davatzikos C: Spatial transformation and registration of brain images using elastically deformable models. Computer Vision and Image Understanding 1997, 66(2):207-222.
  2. Shen D, Davatzikos C: HAMMER: hierarchical attribute matching mechanism for elastic registration. IEEE transactions on medical imaging 2002, 21(11):1421-1439.
  3. Ou Y, Sotiras A, Paragios N, Davatzikos C: DRAMMS: Deformable registration via attribute matching and mutual-saliency weighting. Medical image analysis 2011, 15(4):622-639.
  4. Ou Y, Akbari H, Bilello M, Da X, Davatzikos C: Comparative Evaluation of Registration Algorithms in Different Brain Databases with Varying Difficulty: Results and Insights. IEEE transactions on medical imaging 2014.



Multi-atlas, multi-warp methods have been main-stream in the lab, allowing for very accurate automated labeling of brain MRI and other types of images

Segmentation is a fundamental problem in medical image analysis, in which images are labeled by an automated or semi-automated way. Our group has a long-standing activity in this area, with late emphasis on two areas; 1) segmentation of pathologies, such as tumors or lesions, either by building abnormality-specific models [1-2], or by viewing abnormalities as deviations from statistics of normal anatomy [3]; 2) multi-atlas consensus-based labeling, in which multiple transformations, atlases and parameter sets are used to provide an ensemble of estimates of an individual’s labels, and are subsequently integrated via some optimality criterion[4-6].


Current Projects:

Deformable registration of brain tumor images

— GLISTR: GLioma Image SegmenTation and Registration
GLISTRboost: Our proposed method for segmenting low- and high-grade gliomas in multimodal magnetic resonance imaging (MRI) volumes.

Multi Atlas Skull Stripping MASS

— MUlti-atlas region Segmentation utilizing Ensembles of registration algorithms and parameters, and locally optimal atlas selection — MUSE

  1. Gooya A, Pohl KM, Bilello M, Cirillo L, Biros G, Melhem ER, Davatzikos C: GLISTR: Glioma Image Segmentation and Registration. IEEE transactions on medical imaging 2012, 31(10):1941-1954.
  2. Bakas S, Zeng K, Sotiras A, Rathore S, Akbari H, Gaonkar B, Rozycki M, Pati S, Davatzikos C: GLISTRboost: Combining Multimodal MRI Segmentation, Registration, and Biophysical Tumor Growth Modeling with Gradient Boosting Machines for Glioma Segmentation. In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. edn. Edited by Crimi A, Menze B, Maier O, Reyes M, Handels H. Cham: Springer International Publishing; 2016: 144-155.
  3. Zeng K, Erus G, Sotiras A, Shinohara RT, Davatzikos C: Abnormality detection via iterative deformable registration and basis-pursuit decomposition. IEEE transactions on medical imaging 2016, PP(99):1-1.
  4. Doshi J, Erus G, Ou Y, Resnick S, Gur R, Gur R, Satterthwaite T, Furth S, Davatzikos C: MUSE: MUlti-atlas region Segmentation utilizing Ensembles of registration algorithms and parameters, and locally optimal atlas selection. NeuroImage 2016, 127:186-195.
  5. Doshi J, Erus G, Ou Y, Gaonkar B, Davatzikos C: Multi-Atlas Skull-Stripping. Acad Radiol 2013, 20(12):1566-1576.
  6. Doshi J, Erus G, Ou Y, Davatzikos C: Ensemble-based medical image labeling via sampling morphological appearance manifold. In: MICCAI. vol. MICCAI 2013 Workshop on Segmentation:Algorithms, Theory and Applications (SATA). Nagoya, Japan; 2013.  


Structural Connectomics

We study white matter connectivity of the brain by quantifying connections between brain regions and identifying subnetworks (a.k.a modules or systems) of the structural connectome, using diffusion imaging. Our analyses span multiple levels of investigation, ranging from individual regions and connections to subnetworks, and finally to global topology of the entire brain network.

Current Projects

Basic Connectomic Analysis

  • We have established protocols for creating structural connectomes and standard connectomic measures, and their statistical analyses. Established processing pipelines can be made available on request

— Quantitative analysis of structural connectome abnormalities

  • We are developing novel measures to assess connectivity related injury burden. One such measure is Disruption Index of the Structural Connectome (DISC).

Structural subnetwork analysis

  • We develop advanced subnetwork detection algorithms that can identify subsystems of the brain network with distinct connectivity patterns or distinct biological or behavioral correlates, using machine learning and probabilistic inference techniques.



We are developing tractography paradigms that go beyond the state of the art in addressing the problems of partial and displaced tracts, crucial issues in surgical planning when the tracts are disrupted by non enhancing tumor, hidden by edema or displaced due to the mass effect of the tumor.

Current Projects:

— Tumor Imaging and Tractography ANalyzer — TITAN

  • Aims at developing a tractography paradigm that will track through edema, based on a multicompartment model being fitted to the diffusion data.

— Connectivity-based Fiber Extraction and Identification — Confetti