Notice of Special Interest (NOSI): Leveraging Real-World Imaging Data for Artificial Intelligence-based Modeling and Early Detection of Abdominal Cancers
Notice Number:

Key Dates

Release Date:

January 25, 2021

First Available Due Date:
June 05, 2021
Expiration Date:
January 08, 2024

Related Announcements

PA-20-185 - NIH Research Project Grant (Parent R01 Clinical Trial Not Allowed)

Issued by

National Cancer Institute (NCI)


The purpose of this Notice of Special Interest (NOSI) is to solicit applications to support the secondary use of real-world data for Artificial Intelligence (AI)-based predictive modeling with the ultimate goal of improving early detection and risk assessment for abdominal cancers. This Notice encourages applications proposing multi-institutional collaborative AI development approaches such as federated learning, which distributes the models to data-owners and aggregates the results without sharing the actual data.


Many abdominal cancers are diagnosed too late due to a lack of effective prevention and early detection strategies, even for high-risk groups. Retrospective studies show that abnormal imaging findings can be detected one or two years before diagnosis. This offers a window of opportunity for improving early detection by leveraging subtle changes in radiological scans relevant to the early and premalignant stages of these cancers. Other opportunities involve further stratification of high-risk populations to be tested for these cancers based on combined imaging and clinical predictors.

Next-generation predictive AI systems may enhance early detection and risk assessment for abdominal cancers, reduce overdiagnosis and excessive surveillance in high risk-populations, and enable population-based opportunistic screening. For cancers with population-wide screening programs, such systems have been shown to surpass traditional risk models and computer aided detection tools. These next-generation systems excel at identification and segmentation of incidental and screen-detected lesions, detection and classification of early-stage abdominal cancers, risk assessment for asymptomatic high-risk patients, or further stratification of high-risk groups to guide screening or surveillance. Yet, the use of AI for abdominal cancer screening and diagnostics is currently severely limited by scarcity of data for model training. To match human performance, AI algorithms need to be trained on large, multi-institutional datasets with thousands of examples per category. For early detection applications, such systems also need access to longitudinal datasets representing a wide range of abnormalities. In the absence of population-wide screening programs and extensive cohort studies, obtaining such datasets for abdominal cancers present a challenge.

To overcome the sample size challenge, researchers are increasingly turning to Real-World Data (RWD) to build AI-ready retrospective cohorts. Many health systems, academic centers, and global data networks have access to thousands of radiological images and millions of patient records. For example, Americans undergo 22 million abdominal CT scans every year for reasons mostly unrelated to cancer. Information embedded within these scans can be used to stratify risk for future adverse events in asymptomatic patients more accurately than traditional clinical parameters. In addition to diagnostic outcomes, RWD images are often linked to other valuable information, including diagnoses, risk scores assigned by expert radiologists, patient demographics, comorbidities, lab results, biomarkers, scanners, and image acquisition parameters. These linkages can be used to create clinical predictors or image labels, reduce potential biases, and further improve model performance.

Presently, aggregation of multi-site RWD data outside of research centers is not feasible for many ethical, legal, technical, and practical reasons. New collaborative AI development methods, such as federated learning, bypass the need for direct data access and aggregation while making code and models widely available. The advantages of federated learning have been demonstrated in many medical imaging studies, including whole-brain segmentation, tumor segmentation, and classification to find disease biomarkers. By linking healthcare institutions and not just research centers, federated learning could help overcome data scarcity, which hampers the reproducibility of AI applications in early detection and risk assessment. Federated learning may also represent a sustainable and cost-effective alternative to centralized repositories because data deidentification, storage, and computing costs are minimized or shared with data owners. This approach obviates the need for public cloud hosting, which can be costly for extensive imaging datasets. Lastly, federated learning can be combined with AI Challenges to enable independent benchmarking and integration of AI models.

Research Objectives

This NOSI will serve as a first critical step towards developing next-generation predictive AI systems to aid in early detection and risk prediction for abdominal cancers, including but not limited to pancreatic, liver, and kidney cancers. Through this NOSI, NCI is particularly interested in research that advances multi-institutional AI model development based on federated learning. The NOSI will support, but is not limited to, the following research topics:

  • Develop AI-ready retrospective cohorts such as pre-diagnostic, early-stage, neoplasm or similar cohorts by harmonizing RWD data and adding granular, use-case-specific annotations
  • Develop baseline AI-based algorithms and biomarkers to improve efficiency and effectiveness of (i) screening for high-risk populations; (ii) opportunistic screening in pre-symptomatic populations; (iii) individual risk assessment and stratification of high-risk groups to guide screening and surveillance decision making and practices; or (iv) identification, detection, segmentation, classification, and risk assessment of early-stage cancers, clinically meaningful lesions, and benign abnormalities in pre-symptomatic populations.
  • Identify and characterize common sources of bias, such as image acquisition parameters, scanners, imaging modalities, comorbidities, patient characteristics, and regional differences.
  • Apply federated learning approaches to enhance secondary use of multi-institutional private datasets; expedite access to datasets that have not been previously disseminated for any ethical, legal, technical, or practical reasons.


  • Proposed data collections should have permission to use the data for federated learning and AI development.
  • Initial patient data can remain private but resulting datasets should be made available for annotation and AI development through a federated model of data sharing.
  • The inclusion of additional public datasets with permissions for broad data sharing is encouraged.
  • The code and models should be made available to the research community on ModelHub.AI or similar container-based platforms for structured dissemination of AI models. Applicants are reminded hyperlinks/URL to such platforms may not be included in the application (NOT-OD-20-174).


Application and Submission Information

This notice applies to due dates on or after June 5, 2021, and subsequent receipt dates through January 8, 2024.

Submit applications for this initiative using the following funding opportunity announcement (FOA) or any reissues of this FOA through the expiration date of this NOSI.

  • PA-20-185 NIH Research Project Grant (Parent R01 Clinical Trial Not Allowed)

All instructions in the SF424 (R&R) Application Guide and the funding opportunity announcement used for submission must be followed, with the following additions:

  • For funding consideration, applicants must include “NOT-CA-21-028” (without quotation marks) in the Agency Routing Identifier field (Box 4b) of the SF424 R&R form. Applications without this information in Box 4b will not be considered for this initiative.

Applications nonresponsive to terms of this NOSI will not be considered for the NOSI initiative.


Please direct all inquiries to the contacts in Section VII of the listed funding opportunity announcements with the following additions/substitutions:

Natalie Abrams, PhD
National Cancer Institute (NCI)
Telephone: (240) 474-7336

Lalitha Shankar, M.D., PhD
National Cancer Institute (NCI)
Telephone: 240-276-6510

Sudhir Srivastava, PhD
National Cancer Institute (NCI)
Telephone: (240) 276-7040

Weekly TOC for this Announcement
NIH Funding Opportunities and Notices