Meanwhile, in Amsterdam, a team of researchers looked at pleural mesothelioma on CT scans and apparently said: what if we stopped pretending this cancer grows like a polite little marble?

That is the central elegance of ARTIMES, the new AI-assisted volumetric response system from Groot Lipman and colleagues in The Lancet Oncology (DOI: 10.1016/S1470-2045(26)00084-7). Pleural mesothelioma grows along the lining of the lung, often in a crescent shape, like a shadow wrapping around a lantern. Traditional tumor measurements prefer tidy diameters. Mesothelioma offers a rind, a curve, a stubbornly irregular presence. Very wabi-sabi, medically inconvenient.

The Ruler Was Doing Its Best

Cancer trials often use RECIST-style rules to decide whether tumors shrink, stay put, or grow. For mesothelioma, doctors use modified RECIST, or mRECIST, which measures tumor thickness at selected points. It is practical. It is standardized. It is also a bit like judging a spilled cup of tea by measuring three droplets.

ARTIMES takes a different path. Instead of asking, "How thick is this tumor here, here, and here?" it asks, "How much tumor is there in total?" The system uses deep learning to segment tumor on CT scans, calculates three-dimensional volume, and tracks how that volume changes over time. In image segmentation terms, the AI draws the boundary around the thing of interest. Imagine a radiology intern with infinite patience and no need for coffee, though sadly also no taste in music.

This matters because mesothelioma response assessment has long had a problem of ma, the Japanese idea of meaningful space. The important information sits not only in the visible mass, but in the spaces between slices, readers, scans, and timepoints. ARTIMES tries to honor the whole shape.

A Large Study, Not a Napkin Sketch

The researchers used 10,926 CT scans from 2,080 patients across 14 cohorts. A training subset included 1,176 annotated scans plus 100 negative scans, labeled by 12 radiologists and one pulmonologist. That is a lot of careful outlining. If you have ever manually traced anything on medical imaging, you know this is where enthusiasm goes to find a quiet room.

The model performed strongly in internal testing, with Dice similarity coefficients around 94-95%. Dice measures overlap between two segmentations; 100% would mean perfect agreement, which in medicine is about as common as a printer that works before the meeting. External testing was tougher, with Dice scores around 71-80% against manual segmentations, though surface-distance scores and volume correlations remained useful. That gap matters. It reminds us that AI does not escape messy reality. Different hospitals, scanners, readers, and disease appearances all tug at the model.

Then came the real question: does this help predict outcomes better than mRECIST?

In eight clinical trials covering 4,674 scans from 943 patients, ARTIMES had better patient-level prognostic performance than mRECIST: concordance index 0.83 versus 0.73, with a statistically significant difference. It also detected progression earlier - median 124 days versus 162 days, about five weeks sooner (ASCO Post summary).

Five weeks is not a small brushstroke. For a patient on a treatment that is no longer working, that time can mean fewer side effects, faster treatment changes, and less waiting in the fog.

Why Volume Feels More Honest

The deeper idea here is simple: biology happens in volumes, not just lines. A tumor does not care which three places a human chose to measure. It grows, recedes, thickens, thins, and spreads according to its own unhappy geometry.

This is where ARTIMES has a quiet kind of ikigai - a purpose. It gives clinical trials a response measure that may line up better with overall survival. The study reported that ARTIMES-based progression-free survival correlated much more strongly with overall survival at the trial level than mRECIST-based progression-free survival. That could make future mesothelioma trials cleaner and more sensitive, assuming prospective validation confirms the result.

And yes, "assuming prospective validation" is doing real work here. This was retrospective. Large, multicohort, and serious, but still retrospective. Before doctors use ARTIMES to make routine treatment decisions, it needs prospective testing, regulatory review, workflow integration, and continued checks across hospitals. AI in medicine should not be the flashy guest who shows up, grabs the microphone, and starts diagnosing things after one good karaoke performance.

The Wider AI Imaging Moment

ARTIMES also fits a broader trend. Recent reviews of AI in pleural disease and lung cancer describe the same promise and the same friction: segmentation, diagnosis, prognosis, workflow support, but also data quality, privacy, interpretability, and external validation (Karatas and Dikensoy, 2026; Marchi et al., 2025; Current and future applications of AI in lung cancer). A 2023-2024 mesothelioma segmentation study also showed how model thresholds can change tumor volume estimates, a useful reminder that the "AI answer" often depends on knobs humans still set (arXiv:2312.00223).

The best version of this future is not AI replacing radiologists. It is AI handling the tedious geometry so specialists can spend more attention on judgment, context, and patients. Less ruler. More seeing.

There is elegance in that. Not every measurement needs to be loud. Sometimes the better tool is the one that respects the shape in front of it.

References

1. Groot Lipman KBW, Wittenberg R, de Oliveira Taveira M, et al. Development and validation of artificial intelligence-assisted volumetric response criteria in pleural mesothelioma (ARTIMES): a retrospective, multicohort, multicentre study. The Lancet Oncology. 2026. DOI: 10.1016/S1470-2045(26)00084-7

2. ASCO Post Staff. AI-Assisted Tumor Volume Response Criteria Outperform Physician Assessments and RECIST Criteria in Pleural Mesothelioma. 2026. Link

3. Karatas F, Dikensoy O. Artificial Intelligence in Pleural Diseases: Current Applications and Next Steps. Thoracic Research and Practice. 2026;27(1):57-67. DOI: 10.4274/ThoracResPract.2025.2025-6-2

4. Marchi G, Mercier M, Cefalo J, et al. Advanced imaging techniques and artificial intelligence in pleural diseases: a narrative review. European Respiratory Review. 2025;34(176):240263. PubMed

5. Shenouda M, Gudmundsson E, Li F, et al. Convolutional Neural Networks for Segmentation of Malignant Pleural Mesothelioma: Analysis of Probability Map Thresholds (CALGB 30901, Alliance). 2023. arXiv:2312.00223

6. Current and future applications of artificial intelligence in lung cancer. PubMed. 2025. PMID: 40541405

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.