Description:
Fast-Track proposals will be accepted.
Direct-to-Phase II proposals will be accepted.
Number of anticipated awards: 2-4
Budget (total costs, per award):
Phase I: up to $400,000 for up to 12 months
Phase II: up to $2,000,000 for up to 2 years
PROPOSALS THAT EXCEED THE BUDGET OR PROJECT DURATION LISTED ABOVE MAY NOT BE FUNDED.
Summary
The use of computer-aided detection and diagnosis systems with endoscopic procedures has the potential to improve the
detection of hard-to-find colonic polyps and esophageal lesions and to differentiate high-risk adenomas and dysplasia from
lower risk lesions. Recent data indicate that machine learning, and artificial intelligence can help improve the detection and
diagnosis of imaged precancerous lesions in the colon, liver, lung, prostate, and other organs. Because most imaging-aided
diagnosis examinations are operator-dependent and thus are limited by operator experience and human error, there is a
demonstrable need for systematic, unbiased quantitative approaches to improve detection and diagnostic decision-making
and during preventative screening and surveillance.
Computer-aided detection and diagnosis systems for endoscopy are designed to capture all abnormalities including flat or
diminutive adenomas and those hidden in poorly visualized areas of the intestine, that are commonly missed with standard
endoscopic instruments. However, AI software diagnostic tools are underrepresented at present or have yet to reach their
Page 91
full potential in the clinical practices. Likewise, the solicited algorithms would detect small foci of dysplasia that often go
unrecognized in heterogenous Barrett’s esophagus lesions. Thus, these algorithms have the potential to identify
characteristics that indicate clinical relevance and cancer risk level of precisely visualized lesions, thereby helping medical
professionals perform more effectively and efficiently.
Some companies are currently developing these areas of artificial intelligence and machine learning however, small
business and academic institutions can provide an impetus for the development of these technologies and provide important
pilot data to determine feasibility that can be further validated either in phase II or independently funded research projects
(R21/R01).
Project Goals
The goal of this topic is to solicit proposals to advance the development and application of artificial intelligence-based
algorithms to improve the visual human-based determination of precancerous lesions examined through visual inspection of
upper and lower endoscopies. The technology should be designed for effective detection and characterization of endoscopic
images to properly help decide clinically relevant next steps and to provide physicians with the diagnostic confidence that
comes with AI-support.
The activities that fall within the scope of this solicitation include the development and application of algorithms for
computer-aided diagnosis of Barrett’s esophagus and dysplasia and colorectal polyps and adenomas. Examples of
appropriate activities include the development of computer-aided algorithms that can distinguish between low-grade and
high-grade dysplasia, precancerous and cancerous lesions of the upper and lower gastrointestinal tract. The offeror may
develop only an upper or lower endoscopy computer aided algorithm. Adequate justification for the appropriateness of
including multiple diseases or organs (upper and lower GI tract) must be provided if the same algorithm is to be used.
Adenoma and dysplasia detection rates are validated quality measures for endoscopy. However, these rates vary based on
several factors, endoscopy indication (surveillance vs screening and anatomical location, distal vs proximal colon). A
successful computer-aided algorithm shall demonstrate statistically significant improvement of detection rates for each
modality compared to standard endoscopy lesion detection rates.
Phase I Activities and Deliverables:
• Establish a multidisciplinary project team with expertise in computer-aided diagnosis, medical imaging software
design, informatics, and gastroenterology or medical oncology to oversee the development of software.
• Develop tools for an artificial intelligence-based system that can analyze cell nuclei, crypt structure, and microvessels in endoscopic images, for the identification of esophageal or colon neoplasms (including polyps,
precancers, dysplasia, and metaplasia).
• Develop an algorithm for evaluating endoscopic images for prediction of progression to more advanced disease
and / or response to cancer interception intervention.
• Develop a system where the primary outcome is accurate differentiation between normal tissue, precancers, and
cancers.
• Design and build a computer-aided diagnosis (CAD) tool as a prototype.
• Evaluate CAD performance via available (retrospective) image data sets.
• Refine CAD tool as needed to improve performance and sensitivity and specificity.
• Perform small scale usability testing (5-10 end users) at multiple sites.
• Finalize discussion with FDA for regulatory requirements to be completed in the SBIR Phase II.
Phase II Activities and Deliverables:
Offerors must propose activities leading to the manufacturing and regulatory approval of the computer-aided diagnosis
(CAD) tool, including but not limited to:
• Validate tools developed in Phase I for an artificial intelligence-based system that can analyze neoplastic and
non-neoplastic lesions including polyps, precancers, dysplasia, and metaplasia.
• Clinical validation of the algorithm/AI system for evaluating endoscopic images for prediction of the progression
to more advanced disease and / or response to cancer preventive intervention.
• Include all requirements from FDA to be completed in Phase II.
• Build the final version of CAD tool and test with 5-10 end users.
• Design a prospective trial to evaluate CAD tool’s ability to distinguish premalignant lesions from high-risk
neoplasia at statistical significance level.
• File regulatory submission with FDA for the CAD product for the specific use.
• Develop and implement a commercialization plan for the CAD tool with customers.
