The Breast Radiology Evaluation and Study of Tissues (BREAST) Stamp Project: Image Analysis of Breast Tissue Morphology.

National Cancer Institute (NCI), Division of Cancer Epidemiology and Genetics (DCEG), Hormonal and Reproductive Epidemiology Branch (HREB), plans to procure on a sole source basis services to Image Analysis of Breast Tissue Morphology from The Breast Radiology Evaluation and Study of Tissues (BREAST) Stamp Project from Beth Israel Deaconess Medical Center (BIDMC), 330 Brookline Ave, Boston, MA 02215. This acquisition will be processed in accordance with simplified acquisition procedures as stated in FAR Part 13.106-1(b)(1). The North American Industry Classification System code is 541380 and the business size standard is $14.0 Millions. Only one award will be made as a result of this solicitation. This will be awarded as a firm fixed price type contract. The period of performance is twelve (12) months from date of award. It has been determined there are no opportunities to acquire green products or services for this procurement. NCI / DCEG / HREB, in collaboration with researchers at the University of Vermont center of the Breast Cancer Surveillance Consortium, have completed enrolling women into a study entitled, "Molecular Epidemiology and Biology of Mammographic Density" (lay study name: The Breast Radiology Evaluation and Study of Tissues (BREAST) Stamp Project). The BREAST Stamp Project aims to define the molecular pathology and biology of mammographic density, one of the strongest breast cancer risk factors. Mammographic density reflects the percentage of the breast which is composed of fibroglandular tissue as opposed to fat. Subjects in the BREAST Stamp Project included 1,057 women between the ages of 40 and 65 years participating in the Vermont Breast Cancer Surveillance System who underwent an image-guided biopsy (i.e., ultrasound-guided or stereotactic-guided) to evaluate an abnormality identified on a mammogram at the University of Vermont, Fletcher Allen Health Care from October 1, 2007 through June 30, 2010. Participants provided data about breast cancer risk factors, underwent volumetric breast density assessment using a novel technology, single x-ray absorptiometry, and donated blood and excess tissue for research. Mammographic density was measured as a quantitative volume in the entire right and left breasts and in a localized region surrounding the breast biopsy site. Although elevated mammographic density is related to up to a fivefold increase in breast cancer risk compared with low density, the vast majority of women with dense breasts do not develop cancer and women with low density account for about half of all breast cancers in the U.S. To better understand mammographic density at the tissue level, fixed tissue sections from each BREAST Stamp Project participant's breast biopsy were collected and stained with hematoxylin and eosin (H&E) and are being analyzed for multiple histological parameters, both by visual analysis and using computer-based image analysis. Published data and preliminary results from the BREAST Stamp Project have shown that histological features of breast biopsies are strongly associated with mammographic density, though none has related quantitative histologic features to the volumetric mammographic density measurements collected in the BREAST Stamp Project. This requirement to apply C-Path to the digitized breast tissue sections from all 1,057 BREAST Stamp Project participants will allow for multivariate analysis relating morphologic features to mammographic density in benign and malignant tissues, and to identify features that distinguish benign from malignant tissue in high vs. low mammographic density tissue. The proposed project may contribute to our understanding of the molecular pathology of density, which may help refine risk assessments based on density alone. For those reasons, the BIDMC was chosen because of their unique qualifications to perform this work. The contractor shall use the Computational Pathologist (C-Path), a machine learning-based method which was developed to systematically and comprehensively analyze microscopic breast cancer images to measure quantitative features in breast epithelium and stroma and predict prognosis. The objectives are: 1) To process whole slide digitized images of normal and malignant breast tissue from BREAST Stamp participants (n~1,057) using the Computational Pathologist (C-Path) computer model and provide results for analyzed quantitative features. 2) To analyze multiple target and non-target whole tissue sections from each participant (when multiple sections are available) in order to assess intra-participant reproducibility of measured features, identify potential outliers, and to summarize feature results per participant. 3) To implement quality control (QC) measures for image processing and analysis and provide QC results. 4) To utilize measured quantitative features in the breast epithelium, stroma and their organizational relationships in order to construct statistical models relating features to both mammographic density and pathologic diagnosis. The contractor shall: 1. Process whole slide digitized images of normal and malignant breast tissue from BREAST Stamp participants (n~1,057) using the Computational Pathologist (C-Path) computer model and provide results for analyzed quantitative features. a. To train a classifier to identify epithelium, collagenous stroma, fibro-fatty stroma and blood in invasive breast cancer and benign breast parenchyma; b. To provide results for analyzed features as follows: i. Basic region-based features (# features = 17): proportion, number of super-pixels, and area of epithelium, collagenous stroma, and fibro-fatty stroma, and blood. ii. Basic epithelial nuclear features (# features = 8): 1. Number, proportion, and percent of nuclei (small, medium, large); 2. "Histological Index" score that summarizes nuclear counts of small, medium, large nuclei into a single score. iii. Contextual/experimental epithelial and stromal features (# epithelial features ~800; # stromal features ~800): 1. Raw measures of size, shape, texture, intensity, variability, relationships to neighbors, characteristics of nuclei in regions; 2. Raw epithelial and stromal object measures summarized by min, max, mean, standard deviation per image. 2) The Contractor shall analyze multiple target and non-target whole tissue sections from each participant (when multiple sections are available) in order to assess intra-participant reproducibility of measured features, identify potential outliers, and to summarize feature results per participant. 3) The Contractor shall implement quality control (QC) measures for image processing and analysis and provide QC results for the following steps: a. Manually reviewing images prior to image processing to ensure that each slide contains representative breast parenchyma; b. Manually reviewing C-Path labeled images to flag images where image processing fails; c. Examining intra-replicate correlations to identify outliers with substantially lower correlations. Identified outliers shall be manually reviewed. 4) The Contractor shall utilize measured quantitative features in the breast epithelium, stroma and their organizational relationships in order to construct statistical models relating features to both mammographic density and pathologic diagnosis. a. Divide the study dataset of~ 1,057 participants into independent training and testing sets. b. Perform univariate statistical analyses to identify morphologic features associated with mammographic density in benign and malignant tissue, and to identify features that distinguish benign from malignant tissue in high vs. low mammographic density tissue. The contractor shall use permutation testing with the SAM procedure to account for multiple hypothesis testing and generate false discovery rates (3). Significant associations identified on the training dataset shall be validated on the held-out (i.e., testing) dataset. c. Build image feature-based multivariate models to classify low vs. high mammographic density in benign and malignant tissues. In addition, the Contractor shall build image feature-based models to classify benign vs. malignant tissue in low and high mammographic density settings. Statistical models will be trained using logistic regression and implemented with the glmnet package in R (4). Model performance on the training dataset will be evaluated by cross-validation. In addition, the Contractor shall identify the most robust features using bootstrap methods (5). The best-performing models and features will then be fixed and evaluated on a held-out data set. d. To gain insights into the most significant features, the Contractor shall identify the morphologic features with strongest univariate associations with mammographic density and pathological diagnosis, and identify the features contributing the most weight to the multivariate models and showing the most robust performance in the bootstrap analyses. The Contractor shall review histologic images characteristic of low, medium, and high values of the individual features (and of the model predictions) and attempt to relate the quantitative features to our qualitative understanding of the morphological features of the breast tissue samples. This notice is not a request for competitive quotation. However, if any interested party, especially small business believes it can meet the above requirement, it may submit a proposal or quotation. The response and any other information furnished must be in writing and must contain material in sufficient detail to allow NCI to determine if the party can perform the requirement. Responses must be received in the contracting office by 11:00 AM EST, on September 12, 2013. All responses and questions must be in writing and faxed +1 (555) 000-0000 or emailed to Reyes Rodriguez, Contracting Specialist via electronic mail at contact@example.gov. A determination by the Government not to compete this proposed requirement based upon responses to this notice is solely within the discretion of the Government. Information received will be considered solely for the purpose of determining whether to conduct a competitive procurement. No collect calls will be accepted. In order to receive an award, contractors must be registered and have valid certification in the Central Contractor Registration (CCR) and the Online Representations and Certifications Applications (ORCA) through sam.gov. Reference: NCI-130145-RR on all correspondence.