Novel AI-assisted volumetric criteria for assessing progression in pleural mesothelioma can improve the accuracy of survival predictions, a new study has shown.
In trial data, progression-free survival calculated using the CT-based ARTIMES (AI-assisted response to treatment evaluation in mesothelioma) criteria showed a stronger correlation with overall survival than standard criteria.
ARTIMES, which still requires radiologist oversight, also detected progression more than five weeks earlier than the modified version of Response Evaluation Criteria in Solid Tumors (mRECIST) when applied to individual patient scans.
“Pleural mesothelioma has a distinct, crescent-shaped growth pattern that limits the validity and reproducibility of diameter-based response criteria such as Response Evaluation Criteria in Solid Tumors (RECIST) and modified RECIST (mRECIST),” noted the international team of authors.
“Our findings show that volumetric response assessment can potentially serve as a biologically and clinically meaningful alternative to mRECIST or other response evaluation methods used in routine care,” they said.
In the study, published in The Lancet Oncology [link here], researchers trained the deep learning model on a set of 1,176 annotated CT scans from routine care of pleural mesothelioma in the Netherlands, as well as 100 negative CT scans.
Based on radiologist variability, the minimal detectable change was set at 35mL and 12.4%, forming the basis of ARTIMES thresholds.
Partial response was defined as greater than 35mL or 15% reduction from maximum tumour volume since the start of treatment, or a 75% or more reduction from peak volume.
Progressive disease was defined as an increase in tumour volume greater than 40% and greater than 35 mL from nadir, or a greater than 70 mL increase, or a new extrapleural lesion.
In a validation cohort of 943 trial participants with 4,674 CT scans, ARTIMES and mRECIST identified disease progression in a similar number of patients (635 and 629 patients, respectively), but ARTIMES detected progression 38 days earlier (124 days vs 162 days, p<0.0001).
“Earlier detection of progression could enable timely treatment adjustments and potentially reduce both financial and treatment toxicity,” suggested the authors.
“ARTIMES-derived progression-free survival could reduce the number of negative phase 3 trials by better capturing treatment response at phase 2 and reducing the sample size, while expanding eligibility criteria because there would be no measurable disease requirement,” they added.
An analysis of surrogate endpoints showed that ARTIMES was more prognostic than mRECIST (concordance index 0.83 vs 0.73, p=0.023), and correlated more strongly with overall survival (R2 88% vs 6%).
“AI-assisted volumetry represents a promising direction for response evaluation in mesothelioma and potentially other tumours,” the authors noted. However, they stressed that “prospective validation and regulatory clearance are needed before widespread clinical adoption”.
Proposed workflow for adoption in clinical trials
In an accompanying editorial [link here], Professor Kevin Blyth, Honorary Consultant in Respiratory Medicine at Queen Elizabeth University Hospital, UK, called the study “important” and wrote that “the challenge now is for trialists and regulators to decide what further assurance is needed before adoption.”
He emphasised that ARTIMES involves AI assistance, not AI reporting, and that adoption in clinical trials “seems reasonable” if a human radiologist is presented with volumetric AI tumour calculations as a first read for masked central independent review.
“If designed purposefully, this workflow could enable live AI optimisation, with additional training mediated by any corrections needed, while simultaneously addressing the shortcomings of mRECIST-derived trial endpoints,” he wrote.
“Trials that choose to use AI in this manner should also ensure that contractual arrangements permit data sharing to support AI regulatory approval, paving the way for near-time routine clinical deployment post licensing,” he concluded.