Albert Montillo PhD, University of Pennsylvania   Research Associate Computational Biomedicine, Imaging and Modeling Center Phone: (267) 257-5094

Short bio Research Publications Lectures Professional Service Research Links

 Links to research areas:

• Large data analytics and machine learning
• Medical image analysis
• Computer vision and image processing
• Fast marching methods for implicit modeling

• Large data analytics and machine learning

  Entanglement

  < Coming soon! >

   

  Automatic segmentation of multiple anatomical structures in seconds through discriminative learning

  Real world applications of machine learning in medical imaging require modeling intra- / inter subject variability, computational and memory efficiency to learn from several 100's of scans and achieve prediction on test data measured in seconds not hours. In this work, we address these issues through the construction of a general purpose, highly accurate decision forest which fuses powerful, computationally efficient, long-range, features for extremely fast segmentation (a few seconds or less per volume) of many structures simultaneously from novel, never seen before scan. Applications include structure quantification and abnormality detection, single-click navigation, image guided interventions and structured image search. The approach succeeds were others fail: even organs only partially visible in the scanner's field of view are segmented accurately at the voxel level. While demonstrated on MDCT for organ segmentation, the method is quite general and readily adapted to other modalities and anatomical scales through a simple and rapid retraining.

  

  Feature selection and dimensionality reduction

  I am interested in optimal pairing of feature selection methods with state-of-the-art regression machines in ways that maximize performance and are insensitive to their parameterization. In this research, I study four effects of pairing including: learning time, model selection time, prediction time, and prediction accuracy. Five feature selection methods are tested, including a novel one based on a combination of forward selection and correlation relevancy metrics. The proposed method yields hundreds of times performance improvement for both support vector regression and Random Forest regression over a wide parameter range.

  

   

  Discriminative models for image-based regression

  Highly accurate discriminative models that drive image-based regression and provide near instantaneous prediction on novel test images are essential in many applications. In one project, I propose a method to automatically predict the age of a person given only a single image of a person's face. In this research, I propose new feature construction methods, including a new, efficient, region descriptor called the integral Gabor histogram and use the signals from this descriptor together with the Random Forest regressor to learn a discriminative model of aging. Tests of this method for accuracy on publicly available face data, reveal a prediction accuracy of 3.1 years, which compares favorably to human prediction on the faces. The method also provides a model which is readily interpretable, includes importance measures for each feature, and finds features pertinent for facial development without using a prior anthropometric model.

.

• Medical image analysis

Biomechanical models for tissue segmentation and tracking

Most of the clinically relevant parameters that characterize heart function can be readily derived from a dynamic segmentation of the heart in 3D+time images. Obtaining the dynamic segmentation is challenging due to a wide variety of heart shape, complex motion, and relatively low data quality compared to images of non-moving organs. In this research, I propose a method that uses a volumetric deformable model with regularization constraints derived from biomechanics to dynamically segment the left and right ventricles of the human heart. While most other methods start from a manual segmentation, our method automatically constructs the patient specific model using adaptive statistics-driven, image processing and shape based interpolation. By progressively incorporating elastic tissue properties, the method achieves high segmentation accuracy during the heartbeat for both normal and diseased hearts.

Tissue tracking using a Gabor filter bank

I am interested in the precise tracking of complex tissue motion. In these I experiments I propose a novel approach for the extraction of tissue deformation based on the responses of a bank of Gabor filters. Additionally I propose an interpolation method to recover all deformation information at a finer resolution than the filter bank parameters. The method is fully automated, requiring no user supervision. Tests of the method on synthetic images of tissue undergoing an isovolumic contraction, for which I have exact ground truth, showed very rapid computation (a few seconds) and sub-pixel tracking accuracy.

(a-b) Example of a test case consisting of a synthetic isovolumic contraction sequence, with just the ED and ES short axis images shown (in this case I am simulating tagged MRI). On the right, in the color images are (c) the automatically recovered Dx displacement component where red indicates movement from left to right, blue indicates movement from right to left, and (d) the displacement error compared to ground truth is small throughout the myocardium.

Generative models for segmentation and registration

Spatially parameterized segmentation of the whole brain

In this research, I develop a supervised learning algorithm and use it to build a 3D statistical brain model in MRI. Recognizing that the imaged intensities of each structure vary spatially even within the same tissue, I construct encode spatial information globally using a probabilistic atlas and regionally using an adaptive Markov random field. Compared to most brain segmentation methods which label only 3 structures (CSF, white matter, and gray matter), this method accurately labels over 35 structures throughout the brain. When my collaborators evaluated the method with a team of clinical experts, it showed that 15 subcortical structures can be used to detect changes that presage early Alzheimer's disease (AD).

NEW: The method forms the basis of NeuroQuant, a product sold by Cortechs Labs, Inc which has (1) generated SBIR funding for clinical trial evaluation, (2) received 510(k) FDA clearance as a medical device for the detection of mild cognitive impairment and early AD, and (3) was selected as a Finalist in 2007 Most Innovative New Products Awards.

Sample 3D brain labeling result, and detail showing labeling of subcortical region:

 

Building a probabilistic model of 3D registration

Accurate registration of complex 3D structures often involves a very high dimensional nonlinear transformation requiring hundreds of thousands of parameters. Elastic matching regularizes registration, typically combining the finite element method and constraints from the theory elasticity of continuum mechanics. Recognizing that most deformable shapes change along a low number of deformation modes, I use an elastic matching to learn a probabilistic deformation model, and investigate run-time performance gains by deforming first along eigen-deformation modes.

---

• Computer vision and image processing

Recovering the geometric description of an object from an image

In this research, I propose a method to rapidly and accurately estimate the orientation of an N-dimensional object by processing the patterns in its spatial frequency transform using a non-linear scaling and Hough-line finding. I then refine a geometric description of an object using a hierarchical estimation of object feature dimensions and projected intensity profile processing. US patents #6813377 and #6526165 have been issued for these methods, which are now deployed in successful industrial inspection applications for Cognex, Inc.

 

Segmenting regions of similar texture

Some classes of objects have at least a portion of the object which has a characteristic pattern or texture. In order to rapidly recover the pose of such objects, I propose a method that employs spatial frequency analysis to construct a feature vector for every pixel, including: frequency spectrum, power spectrum, and the angles of dominant powers. I integrate this information to segment regions of similar texture and validate the method on a variety of industrial inspection applications. US patent #6647132 has been issued for this method.

 

Adaptive, non-stationary image denoising

Clinical imaging can be highly variable between imaging sites due to protocol and hardware variations. In magnetic resonance (MR), images can be affected by additional variations including the number and placement of surface coils and several sources of image noise. To normalize such image variation, I propose a new, non-stationary noise suppression method that adaptively regulates anisotropic diffusion using the strength of an estimated intensity inhomogeneity field. I test this method on thousands of images including (1) synthetic images created by simulating the imaging physics of surface coils using the Biot-Savart electromagnetics law, and (2) real MR images. Quantitative assessment of the test results show that the edge-strength of critical tissue boundaries is better preserved using this adaptive denoising, with a high level of statistical significance. This method received a first place award in an international medical imaging conference, and has a wide variety of applications, including the brain, heart, and lungs.

 

 

Optimal correction of intensity inhomogeneity

The intensity of a tissue often varies spatially across an image. Correction of this inhomogeneity is an important step towards automating image analysis. Such inhomogeneity is particularly problematic in newer MR hardware, where higher magnetic fields provide increased image resolution but tend to come with increased intensity inhomogeneity. In 3D the inhomogeneity usually varies in all three dimensions. In 3D+time images, the intensity can also vary over time as magnetized blood moves into or out of the imaged region. In this investigation, I propose an iterative method to progressively segment large single-tissue regions and estimate & suppress the inhomogeneity on 3D+time images. Additionally I investigate the effect of the ordering of denoising and inhomogeneity correction and validate the method on tests of over one thousand images and suggest an optimum ordering of the image preparation operations.

.

---

• Fast marching methods for implicit modeling

Optimal time Eikonal solution on artifact-laden triangulated manifolds

Solving the Eikonal equation can facilitate many types of image analyses and simulations. Most solutions require special processing to handle artifacts such as holes and obtuse triangles. In this research, I present an optimal-time solution for triangulated surfaces that extends the Fast Marching method to handle all mesh artifacts uniformly and provides exceptionally low parameter sensitivity. Tests on a wide range of examples from computer graphics and medical imaging demonstrate the high accuracy and suitability for geodesic measurements and mesh construction.

Implicit 3D curvature as a new speed term for level sets

For this investigation, I explore the use of local shape properties to constrain level set-based shape segmentation. Certain anatomic shapes, such as vasculature, contain consistent local shape properties. In particular these shapes have high first principal curvature while low second principal curvature, while the ratio of these curvatures tends to be relatively constant. I compute the principle curvatures, k1 and k2, directly from the gaussian curvature, K, and mean curvature, H, which in turn are computed from 3D image derivatives.

Comparison of shape representation methods

I am interested in methods that efficiently represent complex 3D shapes. In this research, I study and compare four competing methods including: (1) two coupled surface propagation methods with applications to brain cortex segmentation, (2) one method for geodesic active contours constrained by a generative statistical shape model with applications to subcortical structure segmentation, and (3) a medial core shape representation method. The advantages and disadvantages are detailed and promising future directions are discussed.