Biomedical imaging, signal processing, and computational biology
Novel biomedical imaging techniques have enabled the acquisition of quantitative information from cells, tissues, and organs with unprecedented accuracy and specificity. Combined with the availability of vast computational resources, quantitative biomedical imaging pipelines have the potential to accelerate scientific discovery and improve clinical practice. An important engineering problem in this area relates to extracting quantitative information related to the form (shape and texture) of cells, tissues, and organs. In a concerted effort with the mission of our Center for Bioimage Informatics, the Lane Center for Computational Biology, and the Department of Biomedical Engineering at CMU, the focus of my research efforts is to work in close collaboration with biologists, physicians, as well as other experimental scientists, engineers, and mathematicians to identify, formulate, and implement solutions to important problems related to information extraction from biomedical signals and images. In particular, current projects include the development of general purpose segmentation methods for biomedical image data, visualization of high-throughput screening data, filament (actin, microtubules) extraction, and others. Funding for our research has been provided by the National Institutes of Health, USA (R01GM090033,R21GM088816), the state of Pennsylvania, USA (PITA, tobacco settlement funds, life sciences greenhouse initiative), and Carnegie Mellon's Berkman faculty fund.
Image-based cell morphometry, pathology, and high throughput screening:
We are developing new methods for high throughput screening of microscopy images based on cellular and subcellular morphometry, with specific focus on visualization of statistically meaningful information. Our work combines the benefit of automated image segmentation (see below) with new tools for visualizing statistically significant discriminating information between large groups of cells. Traditional methods utilize numerical features chosen a priori for summarizing and discriminating between sets of cells. Our methods are based on comparing morphological exemplars through deformation and transportation approaches, yielding not only a method for quantitatively comparing two or more cells without the need for a priori features, but also a method for traversing and visualizing the dataset. An example showing the application of the technique to large groups of nuclear chromatin pattern obtained from histopathology images is shown on the right. The technique has also been applied successfully to other types of signals obtained from microscopy images (including fluorescence) of cells.
- W. Wang, J. A. Ozolek, D. Slepcev, A. B. Lee, C. Chen, G. K. Rohde, An optimal transportation approach for nuclear structure-based pathology, IEEE Transactions on Medical Imaging, 30, pp. 621-631, 2011. (pdf)
- W. Wang, Y. Mo, J. A. Ozolek, G. K. Rohde, Penalized Fisher discriminant analysis and its application to image-based morphometry. Pattern Recognition Letters, 32(15): 2128-2135, 2011. (pdf)
- S. Choi, W. Wang, A. J. S. Ribeiro, A. Kalinowski, S. Q. Gregg, P. L. Opreski, L. J. Niedemhofer, G. K. Rohde, K. N. Dahl, Computational image analysis of nuclear morphology associated with various nuclear specific aging disorders. Accepted in Nucleus, 2011. (link)
- W. Wang, J. A. Ozolek, G.K. Rohde, Detection and classification of thyroid follicular lesions based on nuclear structure from histopathology images. Cytometry 77A, pp. 485-494, 2010. (reprint)
- G.K. Rohde, A.J.S. Ribeiro, K.N. Dahl, R.F. Murphy, Deformation-based nuclear morphometry: Capturing nuclear shape variation in HeLa cells. Cytometry, vol 73A, pp. 341-350, 2008. (pdf)
Automated image segmentation
Before different structures (subcellular organelles, cells, tissues, organs) can be quantified these often must be accurately segmented from a variety of images. We are developing methods for segmenting biomedical structures from image data based on a supervised learning strategy. The idea is to utilize a few sample expert annotated images to automatically train a program to repeat the same task on other images. With our software, a researcher interested in segmenting structures from novel imaging modalities can, with little effort, design a finely tuned segmentation method to automatically segment any quantity of image data. Our method fuses elements related to pixel classification, region optimization, and front evolution methods. The method has been used to segment tissues from histopathology images, nuclei from fluorescence and histopathology images, yeast cells from microscopy images, brain tissues from MRI images, and others.
- C. Chen, J. A. Ozolek, W. Wang, G. K. Rohde, A general system for automatic biomedical image segmentation using intensity neighborhoods, International Journal of Biomedical Imaging, Volume 2011, Article ID 606857. (link)
- C. Chen, J. A. Ozolek, W. Wang, G.K. Rohde, A pixel classification system for segmenting biomedical images intensity neighborhoods and dimension reduction, Proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI), 2011, p. 1649-1652. (pdf)
Automated tissue identification & classification from histopathology images:
Tissues in histopathology images can be automatically identified and classified using the segmentation software described above. The idea is to use a few expert annotated images to train a pixel classifier capable of repeating the task in any input image. The software is generic in that it does not utilize user selected features to perform classification, but rather, utilizes the entire pixel information to identify the class of each pixel.
- C. Chen, J. A. Ozolek, W. Wang, G. K. Rohde, A general system for automatic biomedical image segmentation using intensity neighborhoods, International Journal of Biomedical Imaging, Volume 2011, Article ID 606857. (link)
Cellular filament modeling and estimation
Filamentous structures (actin, microtubules) play important roles in intracellular transport, mechanical properties, cell division, and other cellular processes. We are developing a computer program capable of measuring the filament distribution (both of actin and microtubules) of 2D and 3D microscopy images. We believe the program may be useful for scientists who wish to study the properties of filamentous distributions in cells under different experimental conditions.
- A. Shariff, R. F. Murphy, G. K. Rohde, Automated estimation of microtubule model parameters from 3-D live cell microscopy images, Proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI), 2011, p. 1330-1333. (pdf)
- A. Shariff, R. F. Murphy, G. K. Rohde, A generative model of microtubule distributions, and indirect estimation of its parameters from fluorescence microscopy images. Cytometry 77A, pp. 457-466, 2010. (reprint)
- A. Shariff, G. K. Rohde, and R. F. Murphy, Indirect learning of generative models for microtubule distribution from fluorescence microscope images. Proceedings of the ICML-UAI-COLT 2009 Workshop on Automated Interpretation and Modeling of Cell Images (Cell-Image Learning 2009), 2009 (pdf)
Image registration, motion correction, and interpolation:
We have developed a series of algorithms for image registration for quantitative applications in biomedicine. Our multimodal nonrigid registration methods have been used for template-based image segmentation, population analysis, motion correction, distortion correction, and other applications. The methods work well with multi-modal data, since they are based on mutual information optimization. We have also investigated interpolation artifacts, and their effect on image registration itself, as well as their effect on quantitative estimates produced based on registered data. In the papers below we offer several precautions that may be taken to minimize such artifacts.
