SOURCES SOUGHT
R -- Computer Methods for Digitized Histology Image Analysis and Biomedical Classification
- Notice Date
- 8/12/2015
- Notice Type
- Sources Sought
- NAICS
- 541712
— Research and Development in the Physical, Engineering, and Life Sciences (except Biotechnology)
- Contracting Office
- Department of Health and Human Services, National Institutes of Health, National Library of Medicine, 6707 Democracy Blvd., Suite 105, Bethesda, Maryland, 20894, United States
- ZIP Code
- 20894
- Solicitation Number
- NIHLM2015602
- Archive Date
- 8/19/2015
- Point of Contact
- Suet Vu, Phone: 301-496-6546
- E-Mail Address
-
vus@mail.nih.gov
(vus@mail.nih.gov)
- Small Business Set-Aside
- Total Small Business
- Description
- GENERAL INFORMATION INTRODUCTION: This is a Small Business Sources Sought notice. This is NOT a solicitation for proposals, proposal abstracts, or quotations. The purpose of this notice is to obtain information regarding: (1) the availability and capability of qualified small business sources; (2) whether they are small businesses; HUBZone small businesses; service-disabled, veteran-owned small businesses; 8(a) small businesses; veteran-owned small businesses; woman-owned small businesses; or small disadvantaged businesses; and (3) their size classification relative to the North American Industry Classification System (NAICS) code for the proposed acquisition. Your responses to the information requested will assist the Government in determining the appropriate acquisition method, including whether a set-aside is possible. An organization that is not considered a small business under the applicable NAICS code should not submit a response to this notice. The National Institutes of Health (NIH), National Library of Medicine (NLM) is conducting a market survey to help determine the availability and technical capability of qualified small businesses, HUBZone small businesses; service- disabled, veteran-owned small businesses; 8(a) small businesses; veteran-owned small businesses; woman-owned small businesses; or small disadvantaged businesses capable of serving the needs identified below. Background: The proliferation of medical information in the form of digital images has created an opportunity and a challenge for technology to play a role in the analysis and understanding of this data for both research and clinical purposes. In the field of pathology, efforts are underway to use technology to digitally acquire, manage, and analyze histology images with the goals of gaining efficiency and accuracy in interpreting the image data for assessing disease at the tissue level in support of treatment and management programs for health care improvement. The increasing amounts of image data acquired in clinical care centers have created a burden for interpretation by human experts that may reach unsustainable levels. Computer-assisted methods may be able to play a significant role in off-loading demands on experts. These methods must address (1) the acquisition, storage, retrieval, and display of the histology image data, (2) the analysis and interpretation of the histology image data, and (3) the integration of the computer-assisted methods into the clinician (or biomedical researcher) workflow. Previous work [1] done for the National Library of Medicine has produced a set of algorithms which analyzes epithelium images of the uterine cervix and classifies the images into disease categories of Normal, CIN1, CIN2, or CIN3, a standard nomenclature for labeling the epithelium into progressively more diseased categories. (CIN is Cervical Intraepithelial Neoplasia.) A unique characteristic of this previous work is that classifications are produced not only at the level of the entire epithelium segment, but also at sub-segment levels. The epithelium is analyzed and classifications produced in sub-segments which may consist of 1/5 or 1/10 of the original epithelium segment. These previous algorithms have been tested on small sets of digitized uterine cervix histology images from which epithelium regions were manually extracted, boundaries manually segmented (in consultation with expert pathologist), and "ground truth" classifications were obtained from the expert pathologist. Additional background on these algorithms is given below. Methodology of the previously-developed algorithms. (A) Locate and segment epithelium segments within digitized histology images of the uterine cervix (location and segmentation has been manual and/or semi-automated), and also use manually segmented epithelium segments provided by NLM or obtained by the contractor; (B) use algorithms to characterize the geometry and orientation of the epithelium, (C) implement methods to segment, characterize, and/or measure structures within the epithelium that are believed to be of relevance for disease classification, and (D) apply classification algorithms to the data acquired in step (C) to classify subregions within the epithelium into disease categories and to classify the entire epithelium region into a standard disease category. Some elaboration of these steps follows: For (B), since disease is believed to progress in a unidirectional manner, from the basement membrane which separates stroma from epithelium, toward the apical ("surface") side of the epithelium, it is necessary to distinguish and label these two sides of the epithelium; to track disease progression across the epithelium, a natural approach is to trace minimal length paths from the basement membrane to the apical surface, and to analyze epithelium characteristics along these paths; if the epithelium were a rectangle, these paths would be simple perpendiculars between the longer sides of the rectangle; since the epithelium is instead highly irregular, it is first necessary to compute a medial axis which is pointwise equidistant from the two sides of the epithelium; perpendiculars to this axis can then be used as the desired paths across the epithelium. For (C) structures and measurements that are believed to be relevant for disease classification include number of nuclei per unit area, optical density of nuclei, nuclei/cytoplasm ratio, nuclei shape characteristics, and quantity of mitotic cells; nuclei segmentation methods are expected to be highly important; for characterizing number of nuclei per unit area, methods such as Delauney triangularization have been used, where the related measurements are average triangle area and/or average length of triangle leg. For (D) the prevailing disease categories for the uterine cervix are Normal, CIN1, CIN2, and CIN3, where CIN refers to Cervical Intraepithelial Neoplasia. It should be noted that the definition of these categories is a subject of ongoing discussion and review within the medical community, and there is particular uncertainty with regard to the definition of the intermediate CIN2 category. Roughly speaking, the prevailing diagnostic method is to examine the epithelium for the presence of abnormalities from the basal membrane to the apical side of the epithelium. If abnormalities are confined to approximately the first 1/3 of this region, it is CIN1; to the first 2/3, CIN2; and, if the abnormalities extend across the entire epithelium, CIN3. The abnormalities may include high cell proliferation (large numbers of nuclei per unit area), abnormal nuclei/cytoplasm ratio, and abnormal nuclei cell characteristics. Since the disease characteristics of epithelium are believed to be heterogeneous, that is, one subregion may be, for example, CIN1, and another subregion, within the same region of epithelim, may be CIN3, it is a goal of this work to devise methods to characterize disease within subregions of the epithelium. A major challenge will be to develop meaningful classification methods, based on the relatively small amount of expert-provided ground truth available, and to progressively improve the classification as more ground truth becomes available. In the currently-proposed work, these previous implementations will be used and extended, and new classification methods will be added, with the goal of achieving improved classification accuracy, as evaluated on larger sets of image data. In addition the algorithms will be implemented, as necessary, in new software language(s) and environment(s), toward the goal of creating a disease classification aid for the clinician or biomedical researcher. Proposed Work The proposed work is to further investigate techniques that promise to be effective in aiding the clinician or biomedical researcher in disease assessment in uterine cervix histology images and to create a prototype implementation of a user interface to these techniques which fits into a normal clinician or research workflow, to allow assessment of interface usability and value of the algorithms to end users in the clinical or biomedical research communities. Specific work requirements are given below: (1) Toward the goal of improving the results of disease classification that have been obtain to date, the contractor will implement classification methods not yet tried on the existing data set. These methods will include classification based on artificial neural networks (ANN) and, specifically, deep learning [2] methods. (2) In order to expand the set of image data for training and testing the algorithms, the contractor will work with an expert pathologist to acquire additional uterine cervix image data, expert annotations of epithelium regions within this image data, and expert "ground truth" disease classifications at both the whole epithelium segment level and at the 1/10 segment level. It is highly desirable that the "ground truth" classifications be acquired from more than one expert pathologist. The acquired data should show examples of Normal, CIN1, CIN2, and CIN3; in addition, it is the goal of the government to obtain histology image data exhibiting reactive changes (such as those caused by inflammation or changes due to the presence of an intrauterine device), in order to train classifiers to discriminate between these "Normal" changes and the presence of CIN. (3) The contractor will develop a user interface to the algorithms that will, at a minimum, provide display of an epithelium image segment, allow execution of the classification algorithms, and display classification results, both at the whole epithelium segment level, and at the sub-segment level segment level. It is highly desirable that the interface provide any additional data that may be produced which may help the user interpret the result or understand how the algorithm arrived at the result. (4) The contractor will develop a concept of how the user interface and/or algorithms can fit into a whole slide imaging workflow and provide any software "hooks" or Application Program Interfaces (APIs) that may be required to achieve this. It is not expected that the contractor will implement such a workflow, but the concept should be a realistic plan which would allow use of the algorithms and/or user interface in a system that takes as input whole slide uterine cervix images, finds epithelium regions, segments those regions, and passes these segmented epithelium regions to the algorithms for disease classification. The following tasks are to be completed. Full detail is given in the Proposed Work section of this Statement of Work. All image analysis and annotation software delivered shall be written in MATLAB or Java or as negotiated with the government. User interface software shall be written in Java or as negotiated with the government. Task description: Task 1: Evaluate algorithms for disease classification of epithelium in digitized uterine cervix images, including artificial neural network (ANN) and deep learning methods. The investigator will conduct R&D for computer-assisted classification of digitized uterine cervix histology images and will extend and refine previously-created NLM algorithms for the feature extraction and diagnostic classification of these epithelium segments, including classification at the epithelium subregion level, and carry out evaluations of the accuracy of these diagnostic classifications, using expert diagnoses as the reference standard. The classification methods developed and evaluated will include artificial neural networks (ANN) and deep learning. The investigator will also carry out collaborative work with a medical expert to acquire 50 additional manually segmented epithelium segments, along with diagnoses. In particular, the investigator will seek expert-classified examples of epithelium with reactive changes, for training classifiers to distinguish these from CIN. Task 2: Create a user interface for allowing feature extraction and disease classification from epithelium segments suitable for use in the normal working environment of the expert pathologist. As a minimum requirement, the contractor will develop and implement a user interface with no dependencies on specialized, commercial software environments such as MATLAB. The interface will display epithelium segments and sub-segments, along with the automatically-computed classifications for each, and such information as may guide the user in understanding how the classification was computed, such as, for example, visually indicating a region of disease. The contractor will work with the government to define reasonable performance requirements and implement them. This task will be accomplished according to the description in the Proposed Work. In addition to these deliverables, the investigator shall provide bimonthly reports providing updates on the progress of the research. All software deliverables shall include documentation and shall be implemented in Java, or as negotiated with the government. The user interface shall be implemented in Java, or as negotiated with the government. All software shall include source code. The contents of the deliverables shall be as follows: Deliverable 1: Semi-annual technical progress report, plus the results of Task 1. Deliverable 2: Final technical report, plus the results of Task 2. For all tasks the deliverables shall consist of the following: A comprehensive report on all work, in summary form and in detail, including a description of algorithms, theoretical and heuristic rationale for the approach taken, conclusions, recommendations and references. This includes research articles published in the literature as a result of this contract. Test results from all experiments, summarized in graphical and tabular form, with detailed results as appropriate to explain both typical and exceptional cases. Well documented software source codes files as well as executables, if any. Program documentation, such as use guide/help data, on the developed software. All data collected from this research, such as segmented shapes, data files, and images. Prototype to demonstrate research results. Training or assistance for integration of the software. 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 The Contractor is 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. ANTICIPATED PERIOD OF PERFORMANCE: It is anticipated that the period of performance shall be for a 12 month base year with two (2) 12 month option periods. An award is anticipated to be made on or around September 2015. It is anticipated that the contract will be a Firmed-Fixed price type. Interested firms responding to this Sources Sought Notice must adhere to the following: (a) Provide a capability statement demonstrating relevant experience, skills, and ability to fulfill the Government's requirement. The capability statement should be complete and contain sufficient detail for the Government to make an informed decision regarding capabilities; however, the statement should not exceed 10 pages. (b) The capability statement must identify the responder's business type and size; DUNS number; NAICS code, and technical and administrative points of contact, including names, titles, addresses, telephone and fax numbers, and e-mail addresses. (c) The National Library of Medicine (NLM) requires proposals to be submitted via eCPS.: 1) Electronic copy via the NLM electronic Contract Proposal Submission (eCPS) website at https://ecps.nih.gov/nlm. All submissions must be submitted by 1:00pm, Local Prevailing Time, on August 18, 2015. For directions on using eCPS, go to https://ecps.nih.gov/nlm/home/howto and click on "How to Submit." NOTE: To submit your electronic proposal using eCPS, all offerors must have a valid NIH External Directory Account, which provides authentication and serves as a vehicle for secure transmission of documents and communication with the NLM. The NIH External Directory Account registration process may take up to 24 hours to become active. Submission of proposals by facsimile or e-mail is not accepted. EMAILS AND FACSIMILES WILL NOT BE ACCEPTABLE. Disclaimer and Important Notes: This notice does not obligate the Government to award a contract or otherwise pay for the information provided in response. The Government reserves the right to use information provided by respondents for any purpose deemed necessary and legally appropriate. Any organization responding to this notice should ensure that its response is complete and sufficiently detailed to allow the Government to determine the organization's qualifications to perform the work. Respondents are advised that the Government is under no obligation to acknowledge receipt of the information received or provide feedback to respondents with respect to any information submitted. After a review of the responses received, a pre-solicitation synopsis and solicitation may be published in Federal Business Opportunities. However, responses to this notice will not be considered adequate responses to a solicitation. Confidentiality. No proprietary, classified, confidential, or sensitive information should be included in your response. The Government reserves the right to use any non-proprietary technical information in any resultant solicitation(s).
- Web Link
-
FBO.gov Permalink
(https://www.fbo.gov/spg/HHS/NIH/OAM/NIHLM2015602/listing.html)
- Place of Performance
- Address: Bethesda, Maryland, 20894, United States
- Zip Code: 20894
- Zip Code: 20894
- Record
- SN03835257-W 20150814/150812235524-3e7b2cdfc6e287999a39ad67d934aec7 (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 |