IBM's World Community Grid provided the Help Defeat Cancer project with more than 2,900 years of computation and enabled CBII to demonstrate the feasibility of using spectral and spatial signatures to characterize expression patterns in imaged cancer specimens. We have been able demonstrate the fact that the resulting library of signatures is not spatially constrained and can therefore be used to perform analysis at multiple levels of granularity, i.e. at the disc, tissue, and tumor level. During the same time, we developed a new region covariance descriptor, which was shown to have superior performance in tumor classification, and a segmentation framework that can reliably delineate multiple classes of imaged cells and tissues, even when they present with different views, illuminations, and scales.
A Special Message from Dr. David Foran
June 12, 2007
I would like to restate my gratitude to all of those individuals throughout the World Community Grid who contributed computer cycles to the "Help Defeat Cancer" project. Like most research pursuits our team began the Project with an innovative idea which had firm scientific grounding, but required systematic investigation to determine its validity. In our case, we envisioned a high-throughput analysis approach for characterizing protein expression patterns in imaged tissue specimens which could reliably classify subtypes and stages of disease progression in breast, colon, and head and neck cancer.
World Community Grid provided the HDC project with more than 2, 900 years of computation and enabled us to demonstrate the feasibility of using spectral and spatial signatures to characterize expression patterns in imaged cancer specimens. Furthermore, we have been able demonstrate the fact that the resulting library of signatures is not spatially constrained and can therefore be used to perform analysis at multiple levels of granularity, i.e. at the disc, tissue, and tumor level. During the same time we have developed a new region covariance descriptor which was shown to have superior performance in tumor classification and a segmentation framework that can reliably delineate multiple classes of imaged cells and tissues even when they present with different views, illuminations, and scales.
Leveraging the experimental results gathered during the course of the "Help Defeat Cancer" project we submitted a new grant proposal to the National Institutes of Health for funding consideration entitled, "Image Mining for Comparative Analysis of Expression Patterns in Tissue Microarrays". The central objective of this proposal is to build a deployable, grid-enabled clinical decision support system to enable researchers and physicians to automatically analyze and classify imaged cancer specimens with improved diagnostic and prognostic accuracy. Proof-of-concept for the proposed system was conducted using the core reference library of expression signatures that was generated as part of the HDC project.
We were recently informed that the research proposal that we submitted to NIH scored in the top 3.9 percentile. To test the performance of the new technologies and computational tools developed during the course of the 4 year project, a Grid-enabled, virtual laboratory will be established among strategic sites located at The Cancer Institute of New Jersey (CINJ), Columbia University (CU), the Ohio State University (OSU), Rutgers University (RU), and the University of Pennsylvania School of Medicine (UPenn). The software and underlying technologies developed as part of this project will be made available to investigators throughout the scientific and cancer communities for use in clinical decision support, investigative research and discovery.
Although this phase of the "Help Defeat Cancer" project has come to a close I remain convinced that this is actually just the start of new beginnings. This has already been borne out as we approach the start of the new NIH project and as we begin a new collaboration with scientists at IBM's T.J. Watson Research Center. The central objective of the new project with IBM is to advance our preliminary work in order to develop a set of multi-modality meta-classifiers which can simultaneously assess the salient genomic, proteomic and image-based profiles of patients in order to provide improved accuracy in detection, treatment and therapy planning.
Latest update from The Cancer Institute of New Jersey
June 12, 2007. The CBII research lab has a number of new publications in press.
- Lin Yang, Peter Meer, David J. Foran. Pixel-Wise Multiple Class Segmentation Using Histogram Descriptors Over Mean-Shift Patches. IEEE Proceedings on Computer Vision and Pattern Recognition. Accepted for publication. To be presented at the International Conference to be held in Minneapolis, Minnesota, June, 2007.
- Lin Yang and David J. Foran. A Variational Framework for Partially Occluded Image Segmentation Using Coarse to Fine Shape Alignment and Semi-parametric Density Approximation. IEEE Proceedings on Image Processing. Accepted for publication. To be presented at the International Conference to be held in San Antonio, Texas, September, 2007.
- Lin Yang, Peter Meer, Lauri A. Goodell, Michael D. Feldman, and David J. Foran. High-Throughput Breast Cancer Analysis on the Grid. Proceedings on Medical Image Computing and Computer-Assisted Intervention. Accepted for publication. To be presentated at the International Conference to be held in Brisbane, Australia, October, 2007.
- Bonnie H. Hall, Wenjin Chen, and David J. Foran. A Clinically Motivated 2-Fold Framework for Quantifying and Classifying of Immunohistochemically Stained Specimens.Proceedings on Medical Image Computing and Computer-Assisted Intervention. Accepted for publication. To be presented at the International Conference to be held in Brisbane, Australia, October, 2007.
- Gabriela Niculescu, John L. Nosher, David J. Foran. Non-rigid Registration of the Liver in Consecutive CT Studies for Assessment of Tumor Response to Radiofrequency Ablation. IEEE Proceedings of the Engineering in Medicine and Biology Society. Accepted for publication. To be presentated at the International Conference to be held in Lyon, France, August, 2007.
Apr 25, 2007. "Help Defeat Cancer" researchers at The Cancer Institute of New Jersey continue to perform advanced statistical methods to search for correlations among morphological changes and staining characteristics of the immunostained cancerous tissues and the clinical presentation and prognosis of patients. At the same time they are working with investigators from the University of Pennsylvania and Ohio State University to establish a means for combining emerging tissue microarray data exchange standards with the set of new image-based feature measurements that have been generated as part of the "Help Defeat Cancer" project. This capability will make it possible to reliably document, track, and aggregate results in a standard manner across multiple institutions.
Feb 1, 2007. Using the library of expression signatures generated through the "Help Defeat Cancer" project, the research team at The Cancer Institute of New Jersey has begun to develop and evaluate a prototype system for performing data-mining and staging disease progression.
Jan. 10, 2007. Feasibility studies which have recently moved from breast cancer to cancer of the colon, head and neck have consistently shown the utility and reproducibility of using spectral and spatial signatures generated through the "Help Defeat Cancer" project to classify expression patterns in tissue microarrays. We will continue to expand the scale of these studies and have begun to undertake a set of systematic man-machine comparative performance studies.
Dec. 13, 2006. Based upon the success of performance studies using the spectral and spatial signatures to classify expression patterns in breast cancer tissues, we have begun systematic experiments to investigate their usefulness in colon and head and neck cancer.
Click Here for the latest statistics from the World Community Grid
A relatively new investigative tool called tissue microarrays (TMA) holds great promise in helping doctors in selecting proper treatment strategies and providing accurate prognosis for cancer patients. Although TMA is not currently being used by doctors to render primary diagnoses, it does make it possible for researchers to determine the specific type and stage of cancer present and systematically investigate which therapies or combinations of treatments are most likely to be effective for each kind of cancer based upon the known outcomes of individual patients. Specific courses of treatment can then be prescribed for actual cancer patients based on whether a specific set of antigens is present or not.
Much of the difficulty in rendering consistent evaluation of expression patterns in cancer tissue microarrays arises from subjective impressions of observers. It has been shown that when characterizations are based upon computer-aided analysis, objectivity, reproducibility and sensitivity improve considerably. Professor David J. Foran's laboratory at The Cancer Institute of New Jersey, UMDNJ - Robert Wood Johnson Medical School, leads a collaborating project with investigators at Rutgers University and the University of Pennsylvania which has developed a web-based, robotic prototype for automatically imaging, analyzing, archiving and sharing digitized tissue microarrays. Utilizing a combination of sophisticated image processing and pattern recognition strategies, the system can automatically analyze and characterize expression patterns in cancer tissue microarrays. Through funding from the National Institutes of Health, contracts 5R01LM007455-03 from the National Library of Medicineand 1R01EB003587-01A2 from the National Institute of Biomedical Imaging and Bioengineering, these researchers have begun analyzing breast cancer and will soon proceed to evaluate protein and molecular expression patterns in head and neck cancers.