Friends of Cancer Research launched the AI-Based Measurement of Response Evaluation Criteria in Solid Tumors, or ai.RECIST, research partnership April 15, 2025. The effort brings together 27 entities across biopharma, tech and academia, though the FDA and other government agencies are conspicuously absent. The first priority: assessing tool capabilities using lung cancer trial data.

Background: Benefits and limitations of tumor-based oncology endpoints

• Overall survival, or OS, is a clinically meaningful measure of both efficacy and safety and is considered the  gold standard for establishing clinical benefit of cancer products. It is defined as the time from trial randomization until death from any cause in the intent-to-treat population. Accurate evaluation of OS often requires lengthy follow-up periods that can be unfavorable for indications with high unmet need. In addition, these studies have potential for confounding variables to distort results.
• Products for serious conditions with unmet medical needs can be approved more quickly through the FDA’s accelerated approval mechanism. These approvals rely on surrogate or intermediate endpoints that the agency considers “reasonably likely to predict the clinical benefit of a drug.” To assure that the product is as safe and effective as was hoped, the FDA requires that companies conduct postapproval studies (also known as postmarketing requirements, or PMRs) to confirm the benefit of their products. The agency issued draft guidance on designing trials to support accelerated approval of oncology products in March 2023. [ Read full AgencyIQ analysis here.]
• The endpoints used to support cancer product efficacy are the focus of a separate guidance finalized in 2018. The determination of an appropriate endpoint in oncology is based on the specific disease and highly depends upon numerous factors. These include effect size, effect duration, depth of response, available therapy, disease setting, location of disease, the clinical consequences of delaying or preventing disease progression or delaying administration of more toxic therapies, and the risk-benefit relationship.
• Tumor-based surrogate and intermediate endpoints such as objective response rate, or ORR, and progression-free survival, or PFS, often serve as the basis of approval for initial regulatory decisions and have been used to support traditional approval in certain circumstances. ORR is the most commonly used surrogate endpoint to support accelerated approval, defined in the 2018 guidance as the proportion of patients with tumor size reduction of a predefined amount and for a minimum time period. The guidance defines PFS as the amount of time from trial randomization to objective tumor progression or death, whichever occurs first. Criteria for response and progression should be predefined in the study protocol, but the FDA notes that “a variety of response criteria have been considered appropriate.”
• The guidance points to the revised response evaluation criteria in solid tumors, or RECIST, guidelines (version 1.1) as an example of a pathology roadmap that has been previously used to measure ORR and PFS for solid tumors. The RECIST framework determines a patient’s level of response to treatment based on the percentage increase or decrease in lesion size as measured in radiographic images. As summarized by FDA regulatory science researchers, “A complete response (CR) is the disappearance of all tumors. A partial response (PR) is at least a 30% decrease in the sum of the longest diameters (LD) of the target lesions, taking the baseline sum LD as reference. A stable disease (SD) is neither sufficient shrinkage nor increase to qualify for either PR or PD. A progressive disease (PD) is at least a 20% increase in the sum of LD of the target lesions, taking as reference the smallest sum of LD recorded since treatment started.”
• Measurements taken under RECIST generally rely on trained reviewers, often with radiology and oncology experience. This introduces some subjectivity regarding the selection and interpretation of lesions. In the 2018 guidance, FDA wrote, “When the primary study endpoint is based on tumor measurements (e.g., PFS or ORR), tumor assessments generally should be verified by central reviewers blinded to study treatments … to ascertain lack of assessment bias.”
• The validity of these endpoints has been called into question in recent years. In early 2022, leadership within FDA’s Oncology Center of Excellence estimated that the accelerated approval program is responsible for access to lifesaving anticancer therapies a median of 3.4 years before they would have been available otherwise. However, in March 2023, OCE leadership published a report that found that some products approved using tumor-based endpoints (ORR or PFS) did not show improvements in overall survival, and some products that did ultimately improve overall survival did not demonstrate improvements in PFS or ORR, or both. [ Read full AgencyIQ analysis here.]

Now, a new Friends of Cancer Research project asks: Can artificial intelligence-based imaging tools improve tumor measurement?

• Quick regulatory context: The FDA’s many efforts related to AI in recent years culminated in the January 2025 release of a foundational draft guidance on the use of AI in regulatory decision-making for drugs and biological products. The document proposes a “risk-based credibility assessment framework” to evaluate the appropriateness of an AI model to address a regulatory question of interest. This includes decision-making regarding safety, efficacy and quality. The credibility assessment comprises a seven-step process: (1) define the question of interest that will be addressed by the AI model, (2) define the context of use, or COU, for the AI model, (3) assess the AI model risk, (4) develop a plan to establish the credibility of AI model output within the COU, (5) execute the plan, (6) document the results of the credibility assessment plan and discuss deviations from the plan, and (7) determine the adequacy of the AI model for the COU. [ Read in-depth AgencyIQ analysis here.]
• FOCR announced the launch of the AI-Based Measurement of Response Evaluation Criteria in Solid Tumors, or ai.RECIST, research partnership April 15. The effort aims to answer this question: Can AI-based imaging tools improve tumor measurement? The project’s webpage outlines bias and subjectivity issues related to RECIST-based assessments. Further, FOCR notes that blinded independent central review is “resource-intensive, potentially prolonging trial timelines, delaying treatment decisions, and increasing costs,” which can have a downstream impact on patient access.
• The announcement says ai.RECIST will build on FOCR’s  Digital PATH Project, which established performance metrics to evaluate AI-driven computational pathology platforms assessing HER2 status in breast cancer. This biomarker is clinically relevant and guides targeted treatment decisions. The results of this project were presented in December 2024. Key findings were that these tools “demonstrate promise,” with a few noted areas of pronounced variability.
• Efficacy enters the chat: The ai.RECIST project raises the stakes, focusing on the criteria used to inform primary efficacy endpoint assessment in clinical trials. The project is set to have two overarching phases, though the amount of time for each is not provided.
• Phase 1 aims to evaluate feasibility of AI tools for supporting RECIST measurements in clinical trials. This will involve assessing AI tool capabilities and establishing common image characteristics and metadata. Then, variability between AI tools and human readers will be compared across a common dataset, which stems from lung cancer trials.
• Phase 2 aims to refine RECIST models and has a broader scope. First, members will consider “alternative approaches for measuring tumor burden,” such as kinetics and metabolomics. At this time, the webpage does not describe further how these other approaches will fit with the imaging tools and whether they will also incorporate AI. AgencyIQ notes that FOCR has led efforts in recent years to advance and harmonize circulating tumor deoxyribonucleic acid, or ctDNA, methods for solid tumor clinical trials. While the FDA has maintained its posture that that ctDNA is not yet a validated surrogate endpoint, FOCR’s work in this area may lend itself to opportunities in this phase of the project. This second phase also intends to “establish a standardized approach for integrating AI-based imaging tools into clinical trials.” Further detail is not provided, but this aspect will likely involve FDA’s seven-step process for establishing credibility of AI.
• The project’s roster includes a consortium of 27 partners spanning industry, academia and the nonprofit sector. Based on AgencyIQ’s categorization in the table below, the project is set to have a high volume of data if all pharmaceutical companies and medical centers contribute to the pool. In addition, the project will have almost a dozen proprietary platforms from which to choose, or possibly integrate, to achieve its goals.