1910MO - ARTIMES: Automated response evaluation to treatment in mesothelioma

Background

Pleural mesothelioma (PM) grows in a crescent shape along the thoracic wall, complicating diameter-based response evaluation. Manual total tumor volume (TTV) monitoring, while promising, is labor-intensive and thus infeasible. Therefore, we developed an Artificial Intelligence (AI) model for automatic TTV quantification and devised novel volumetric response evaluation criteria (ARTIMES), aiming to improve upon the existing standard mRECIST.

Methods

We included 9203 CT scans from 1723 patients with PM over 113 hospitals in 27 countries (3532 CT scans from phase II-III clinical trials: NVALT19, NivoMes, INITIATE, PEMMELA, SAKK17/18, MIST1, and LUME-MESO; 5671 CT scans standard care). Partial response (PR) was defined as a decrease larger than the measurement error between the AI and radiologists. Progressive disease (PD) criteria were defined on the premise that ‘no treatment-effect’ should equate to untreated tumor growth, based on pre-treatment CT scans and untreated control group of NVALT19.

Results

Annotated data from 1102 full CT scans by 13 radiologists were used for AI training and validation. On external validation across 18 independent hospitals (100 CT scans), the AI achieved 95% (DICE) overlap in pixel labeling with radiologists’ annotations and R2=99% in TTV. AI delineation was preferred by two independent radiologists in 70% of the cases compared to manual radiologist annotations in 44 CT scans of SAKK17/18. PR equals the measurement error at -35 mL tumor decrease. PD was defined as TTV growth of 40% and +35 mL, or an absolute growth of 100 mL. ARTIMES based on AI-derived TTV only outperformed mRECIST on 7 clinical trials combined (3532 CT scans), reaching a concordance index of 0.76 (p<0.0001) compared to 0.72 (p<0.0001) for mRECIST in a time-varying Cox Proportional Hazards model. ARTIMES identified PD on the median 7 weeks earlier than mRECIST (p<0.0001). We observed a significant prognostic value of the AI-derived baseline TTV (651 CT scans, log-rank, p<0.0001).

Conclusions

We crafted the first volumetric response evaluation criteria based on accurate AI-delineated tumor volume. ARTIMES outperformed mRECIST by reaching a higher concordance index than mRECIST and detecting PD earlier. Prospective validation and clinical AI implementation are ongoing.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

K.B.W. Groot Lipman.

Funding

NVALT - Dutch Society of Pulmonology and Tuberculosis.

Disclosure

T. Boellaard: Financial Interests, Institutional, Research Grant: Pfizer; Financial Interests, Personal, Speaker, Consultant, Advisor: MSD, Pfizer. D.A. Fennell: Financial Interests, Institutional, Research Grant: Aldeyra, Boehringer Ingelheim, Astex Therapeutics, Bayer, BMS, GSK, Eli Lilly, MSD, Clovis, RS Oncology; Financial Interests, Personal, Speaker, Consultant, Advisor: Atara, BMS, Boehringer Ingelheim, Cambridge Clinical Laboratories, Targovax, Roche, RS Oncology. A. Curioni-Fontecedro: Financial Interests, Personal, Advisory Board: AstraZeneca, Bristol Meyer Squibb, Boehringer Ingelheim, MSD, Novartis, Amgen, Roche, Takeda, Janssen; Financial Interests, Institutional, Advisory Board: Daiichi Sankyo; Non-Financial Interests, Leadership Role: Swiss Academy for Clinical Cancer Research (SAKK); Non-Financial Interests, Principal Investigator, of clinical trials: Roche; Non-Financial Interests, Principal Investigator, Clinical Trials: Takeda, MSD, Bristol Meyer Squibb, Amgen, AstraZeneca; Non-Financial Interests, Principal Investigator, Clinical trials: iTEOS therapeutics. P. Baas: Financial Interests, Institutional, Advisory Board: BMS, MSD; Financial Interests, Institutional, Research Grant: MSD, BMS. S. Burgers: Financial Interests, Institutional, Sponsor/Funding, MSD sponsored the study financially and provided the medication: MSD; Financial Interests, Institutional, Speaker, Consultant, Advisor: Bristol Myers Squibb; Financial Interests, Institutional, Other, Honoraria: MSD. All other authors have declared no conflicts of interest.