SOLICITATION NOTICE
R -- Network Computing for Biomedical Image Processing and Cervigram Segmentation
- Notice Date
- 8/7/2008
- Notice Type
- Presolicitation
- NAICS
- 611310
— Colleges, Universities, and Professional Schools
- Contracting Office
- Department of Health and Human Services, National Institutes of Health, National Library of Medicine, 8600 Rockville Pike, Bethesda, Maryland, 20894
- ZIP Code
- 20894
- Solicitation Number
- 08-174-CYC
- Archive Date
- 9/6/2008
- Point of Contact
- Cara Y Calimano,, Phone: 301 496-6127
- E-Mail Address
-
CalimaC@mail.nlm.nih.gov
- Small Business Set-Aside
- N/A
- Description
- This is a synopsis being conducted under Simplified Acquisition Procedures in accordance with FAR Part 13. This solicitation document incorporates provisions and clauses that are in effect through the June 2008 Federal Acquisition Regulation revision, which includes the consolidation of all Federal Acquisition Circulars through 2005-26. This acquisition is NOT a small business set-aside and the NAICS code is 611310. It is the intent of the National Library of Medicine (NLM) to procure professional services from Dr. Xiaolei Huang with LeHigh University on a sole source basis. Professional services are required over a 12-month period to (1) investigate alternatives for, and to implement a system for network computing for biomedical image processing and (2) refine the development of algorithms for the segmentation of digitized uterine cervix images. Both of these activities will support Content-Based Image Retrieval (CBIR) of biomedical images for the National Library of Medicine (NLM). (1) As part of NLM R&D work to create a Content-Based Image Retrieval System for biomedical images, research and development are required to develop techniques to develop a computing system that will allow exploitation of computing resources at NLM during the hours when they are not used; this system will allow image processing algorithms and data to be distributed among available machines for the computational solution of a larger image processing problem in pipelined fashion. (2) In addition, work is required to segment regions within uterine cervix images in order to index these images by content. Existing NLM algorithms which segment biomedical images by color and/or texture, using methods based on Support Vector Machine technology, will be improved and extended. The focus of the work will be on segmentation of acetowhitened regions in these images, but the methods developed will be analyzed and extended, where feasible, for the segmentation of other significant tissue regions on these images, including squamous epithelium and columnar epithelium. Proposed Work: The scope of the work is (I) investigating and experimentally evaluating the use of multiple PCs for computer-intensive image processing tasks, such as feature extraction from images, and (II) extending existing NLM capabilities for the segmentation of uterine cervix images. For (I), the specific work required is as follows: (1) examine alternative open-source approaches, taking the BOINC software as one possible solution to be examined; compare the technical strengths of the alternative approaches; (2) investigate which computing paradigms* are supported by the alternatives; (3) in consultation with the government, select one approach; (4) implement a computing framework based on this approach; (5) test the framework with a proof-of-concept computing task. *In order of complexity, these are the paradigms of interest to the government: 1. Simple SIMD. In this paradigm, a central node (the “executive” node) distributes different sets of data to N computing nodes. Each node carries out identical instructions and outputs a result set. The executive node gathers the result sets. 2. SIMD with pipeline processing. In this paradigm, the N computing nodes are subdivided into N1, N2,…Nm groups. Each group carries out identical instructions (each node in the group may receive different sets of data, though); each node in a group outputs a result set; the executive gathers the result sets for a group, and may use these results to compute the input data for the next group. 3. Complex algorithm processing. The executive and computing nodes carry out all of the processing and data exchange for an image processing algorithm of the complexity of Active Shape Modeling. (It is expected that this may not be practical within the scope of this contract, or the contractor may conclude that the underlying computing infrastructure is necessarily highly problem-specific for complex algorithms, but the contractor should perform at least a short feasibility analysis with recommendations.) In this respect, a distributed computing project implementation of Active Shape Models (e.g. modeling of the uterine cervix and model-based segmentation) is recommended. Requirements. Below are the desired features of the implementation. Because this is a research and development effort, it is not known a priori whether all of these requirements can be met. The contractor will take these requirements as technical goals, and will advise the government during the course of the work, which requirements can be met, and what technical trade-offs are feasible. The resulting implementation should satisfy the following requirements, subject to the above qualification: (1.) It should operate on both Windows and Linux platforms, on ordinary workstation-class machines. (2.) Computing nodes should not require hosting on server hardware and should not require server software. Server software is hosted by the server (or executive node). The server machine should not require special proprietary server operating system or application software. The server machine may be a ordinary workstation-class hardware. (3.) The implementation should be practical for computational use on workstations during “off-duty” or nighttime hours. (4.) The proof-of-concept should include processing of some standard image processing features used in classification and CBIR work, including color histograms, GLCM matrices, and Gabor-based features. The implementation should be modular and flexible so that new image processing tasks, such as watershed segmentation, Active Shape modeling, can be easily plugged in. (5.) The implementation should provide for consistency checking of data transmitted and received between a computing node and the executive node, for example, by a bi-directional handshake. (6.) The implementation should allow, by empirical measurement, the I/O rate between the executive and a node, and the computational capability of each node, and the executive. This data then can be used to properly apportion the quantity of data distributed to each node to minimize total computation time for a problem. (7.) The implementation should work, if possible, in a heterogeneous environment with a mix of Windows or Linux machines. For (II) the work required is as follows: (1) Investigate interactive segmentation methods in terms of speed, accuracy, usability, scalability, and reproducibility (i.e. whether an algorithm produces consistent results given different seed-point initializations). Interactivity may include such actions as marking points or small regions on the image to be segmented as “cues” for the segmentation algorithm, interactively selecting image weights from the training set of images, and improved editing of the segmented image (to include features such as undo/redo capability for deleting/restoring connected regions). Investigate the concept of using relevance feedback to improve the segmented results. Create an interactive segmentation implementation with a Web interface that is built upon existing NLM uterine cervix SVM segmentation algorithms. (2) Investigate and make recommendations for future tasks to possibly incorporate new algorithms, new interactive functionality, and possibly new features (beyond color) for the segmentation of all anatomy and features of interest in the uterine cervix: cervix ROI, acetowhite tissue, squamous epithelium, columnar epithelium, os, squamous metaplasia, blood, mucus, cysts in an interactive segmentation system. The result of this part of the work is a technical analysis and recommendations. Experimental results to support the analysis are highly desirable, but not a requirement. REQUIRED TASKS: The work described in the Proposed Work section shall be completed. Below, that work is divided into three required subtasks and presented in summary form. All software delivered shall be coded in MATLAB, except the Web interface, which shall use standard Web technology, (such as HTML, CSS, Javascript, Java, or PHP), the networking computing capability (which may use C/C++ or other standard programming languages, with the goal of being hardware-platform independent), or except as negotiated with the government. All software delivered shall include source code and documentation. Task description: Subtask 1: Investigate alternatives for solutions for network computing for image processing and implement a solution which operates at a preliminary level. Extend segmentation capability for segmenting acetowhite regions on uterine cervix cervicography images using Support Vector Machine algorithms by implementing a preliminary version of an interactive segmentation system. Carry out the subtask, according to the requirements in the Proposed Work. Subtask 2: For network computing for image processing, implement a solution which achieves simple SIMD and SIMD with pipelined processing, and satisfies as many of the specified requirements as possible. Provide analysis of tradeoffs and factors affecting the use of the implementation for a complex image processing algorithm, such as Active Shape Modeling. Implement a network-computing version of ASM, if feasible. Extend segmentation capability for segmenting acetowhite regions on uterine cervix cervicography images using Support Vector Machine algorithms by implementing a final version of an interactive segmentation system. Provide a technical analysis and recommendations for future tasks toward the creation of an interactive segmentation system for all anatomy and tissue regions of interest on the uterine cervix. Carry out the subtask, according to the requirements in the Proposed Work. For subtasks 1-2, the Contractor shall use cervicography images made available by the government. Notice of Government Unlimited Rights to Work First Produced Under This Contract: Government rights to work first produced under this contract are established by Federal law including, but not limited to, this specific reference: FAR 42.227-14, Rights in Data – General, (b) (1). Requirement to Notify Government of Proprietary Work Dependencies: All algorithms and code developed under this contract are U.S. government property and are for public use and benefit, including, but not limited to, internal use by government institutions, use by researchers or other parties collaborating in government work, and public domain release, at the option of the government. Offerors are required to notify the Government in writing of any dependencies of the deliverables under this contract on proprietary, copyrighted, or patented work that potentially inhibits, restricts, or requires permission for the dissemination of the deliverables to the public, other governmental agencies or research groups, or to any other parties whatsoever. Lehigh University is uniquely qualified for this work because of the expertise and experience of the principal investigator below (Dr. Huang) in the specific area of image segmentation by Support Vector Machines combined with her educational background and research experience in advanced image processing and numerous other technical publications, and the support of the facilities and staff of the Image Data Emulation and Analysis (idea) laboratory. Dr. Huang has provided preliminary results of Support Vector Machine segmentation for the uterine cervix images and has presented additional technical work to the Lister Hill Center at the National Library of Medicine. This notice of intent is not a request for competitive quotations however, all responses received, within 15 days from the date of publication of this synopsis will be considered by NLM. EMAILED OR FAXED PROPOSALS WILL NOT BE ACCEPTED. The quoter shall include all information which documents and/or supports the qualification criteria in one clearly marked section of its quotation. The contractor shall comply with all applicable Federal, State, and local laws, executive orders, rules and regulations applicable to its performance under this order. Full text of clauses and provisions are available at Federal Acquisition Regulation (FAR): http://www.arnet.gov/far/index.html. The following clauses and provisions apply to this acquisition and may be obtained from the web site: FAR 52.213-4, Terms and Conditions—Simplified Acquisitions (Other Than Commercial Items) (February 2008).
- Web Link
-
FedBizOpps Complete View
(https://www.fbo.gov/?s=opportunity&mode=form&id=7adde6710cd1372c23a05528ce4f89b9&tab=core&_cview=1)
- Place of Performance
- Address: 6707 Democracy Blvd., Suite 105, Bethesda, Maryland, 20892, United States
- Zip Code: 20892
- Zip Code: 20892
- Record
- SN01633787-W 20080809/080807222453-7adde6710cd1372c23a05528ce4f89b9 (fbodaily.com)
- Source
-
FedBizOpps Link to This Notice
(may not be valid after Archive Date)
| FSG Index | This Issue's Index | Today's FBO Daily Index Page |