ai.RECIST Project
Artificial Intelligence-Enabled Response Evaluation Criteria in Solid Tumors Project
Quick Links
Share on Social Media
Can AI-enabled imaging tools improve tumor assessment and accelerate clinical trials?
What is the unmet need and why does it matter?
Tumor response assessments rely on standardized and unbiased criteria through the Response Evaluation Criteria in Solid Tumors (RECIST) performed by radiologists (i.e., expert human readers). In clinical trials, RECIST-based assessments provide a systematic approach to objective tumor measurements at defined time points, where representative tumor lesion(s) are monitored over the course of treatment to quantify treatment efficacy. However, the implementation of RECIST faces several challenges, including investigator bias, subjectivity in lesion selection, and variability in measurements across clinical sites and radiologists. Artificial intelligence (AI)-enabled tumor assessment tools have the potential to address these challenges, reducing variability, increasing efficiency, and improving measurement accuracy. However, alignment among AI-tool outputs is necessary.
How are we helping to find solutions?
Friends created a research partnership with AI-enabled tumor assessment tool developers, pharmaceutical companies, academics, patient advocates, and government officials to evaluate how these tools compare with one another and with human-readers when using RECIST to measure tumor changes over time. After aligning on a statistical analysis plan, radiologic images from a clinical trial will be shared with tool developers, who will provide predetermined outputs related to tumor assessments. Understanding the level of agreement among tool developers assessing RECIST-based measurements will also set the stage for future work in exploring novel approaches to assessing tumors.
The key objectives of the project include:
- Assessing AI tool alignment – Do different AI-enabled tools provide consistent tumor assessments?
- Comparing variability among AI tools and human assessments – How well do AI-enabled assessments align with RECIST-based readings by human readers?
How does this impact patients?
Blinded Independent Central Review (BICR) is used in clinical trials to ensure accurate tumor assessments. Regulators often require BICR to minimize bias by blinding human readers to patient and treatment details when evaluating imaging-based endpoints such as progression-free survival and objective response rate for clinical trial assessments. However, BICR is a resource- and time-intensive process, potentially prolonging trial timelines, delaying treatment decisions, and increasing costs. These delays may limit patient access to new therapies and may also require additional imaging or adjudication when discrepancies arise between local and central assessments. AI-enabled tumor assessment tools have the potential to streamline this process by providing consistent, unbiased verification of local assessments, reducing review time, and improving trial efficiency without compromising data integrity. Alternative approaches to RECIST-based tumor assessments may have better association with overall survival and improve efficiency. This project will explore alternative measurement strategies and opportunities to enhance the BICR process. By improving the speed and reliability of tumor assessments, AI could accelerate clinical trial progress and improve patient access to effective treatments.
Project Approach
Project Outcomes
Friends prioritizes sharing findings from our projects with the public to inform policy:
White Paper
Friends Public Meeting
In February 2026, Friends hosted a meeting to review the current landscape and application of evolving early endpoints such as (AI)-enabled tumor assessments and explore how they may improve trial efficiency and inform regulatory decision-making.
Project Partners
Altis Labs, Amgen Inc., AstraZeneca, Bristol Myers Squibb, Clario, Daiichi Sankyo, Inc., Eli Lilly & Co, Foundation for the National Institutes of Health, Friends Advisory Advocates, Genmab, GSK, Guebert, Johnson & Johnson Innovative Medicine, LivAI, MD Anderson Cancer Center, Medidata Solutions, Merck & Co., Inc., MERIT CRO, National Cancer Institute (NCI), Novartis AG, Onc.ai, Pfizer Inc., Picture Health, Project Data Sphere, Quibim, Qure.ai, Radiomics.bio, Raidium, Takeda Pharmaceuticals Inc., Temple University, Tempus AI, University of Pittsburgh Medical Center, Voiant Clinical, and Vysioneer.