AI Measures Whole Tumors, Challenging Old Rules for Assessing Mesothelioma Response

For patients with pleural mesothelioma, whether treatment is working is often not an easy question to answer from imaging. This cancer often spreads like a thin membrane along the lung and chest wall, and does not necessarily form a clearly bounded mass whose diameter can be measured with a ruler; when the shape of the lesion itself is irregular, physicians’ interpretation of whether it has “grown” or “shrunk” can also easily fall into a gray area.

Researchers at the Netherlands Cancer Institute said on June 17 that the AI model they developed, ARTIMES, performed better than physician interpretation and current tumor measurement standards represented by RECIST in assessing treatment response in pleural mesothelioma. The related study was published in The Lancet Oncology, and was designed as a retrospective, multi-cohort, multicenter analysis.

The core use of ARTIMES is not to diagnose cancer in patients, but to estimate the overall tumor volume in CT images before and after treatment and compare how it changes over time. This is especially critical for pleural mesothelioma: traditional RECIST uses tumor diameter as a common language, facilitating communication in clinical trials and among medical teams, but for lesions that spread along the pleura, one or a few linear measurements may not fully represent the true tumor burden.

According to information released by the research team, model training and testing included 121 hospitals worldwide, more than 2,000 patients, and over 11,000 CT images, and its performance was evaluated using data from 8 clinical trials. The research team said ARTIMES’ response criteria based on volume change better reflected information related to patient survival than current standards, and may also reduce noise in clinical trials caused by unstable measurement.

The practical significance of this advance is that treatment decisions may receive clearer signals earlier. If imaging shows that the tumor is in fact not responding, physicians may be able to consider stopping ineffective treatment and switching to other options more quickly, allowing patients to endure fewer side effects and less uncertainty. However, the research team also emphasized that AI measurement results are still reviewed by physicians; at present, it is more like a tool that strengthens image interpretation than a system that automatically replaces clinical judgment.

The limitations also cannot be overlooked. Although this study was large and multicenter in origin, it was still a retrospective analysis; the model’s prospective performance across different hospital workflows, scanning parameters, and patient populations still requires longer-term validation. The press release also did not provide enough detail for external readers to fully assess all sources of bias, so its clinical promise should be understood in the context of the paper’s methods and subsequent independent validation.

Regulation is also the next hurdle. ARTIMES can currently be used at the Netherlands Cancer Institute under the European Union’s current in-house exemption rules, but if it is to be expanded to other hospitals, it will still need to obtain the corresponding medical device approval. The research team has made the model publicly available for research use; what will truly determine whether it can change pleural mesothelioma care will not be just algorithm scores, but whether it can improve decision-making steadily and transparently in everyday medical care and clinical trials.