Back to all papers

Artificial intelligence-enabled early intracranial volumetric response predicts systemic progression-free survival in the phase III CROWN study.

September 7, 2026pubmed logopapers

Authors

Lu SL,Chang YC,Liang CH,Chiang PL,Maresca K,Pollom E,Wilner K,Lu JT,Toffalorio F,Giaccone G,Duan C

Affiliations (8)

  • Department of Radiation Oncology, National Taiwan University Cancer Center, Taipei, Taiwan.
  • Graduate Institute of Oncology, National Taiwan University College of Medicine, Taipei, Taiwan.
  • Vysioneer Inc., Cambridge, Massachusetts, USA.
  • Pfizer, Cambridge, Massachusetts, USA.
  • Department of Radiation Oncology, Stanford University, Palo Alto, California, USA.
  • Pfizer, San Diego, California, USA.
  • Pfizer, Milan, Italy.
  • Meyer Cancer Center, Weill Cornell Medicine, New York, New York, USA.

Abstract

Manual endpoint assessment is labor-intensive and variable. This study evaluated artificial intelligence (AI)-powered intracranial response assessment and explored imaging prognostic factors in the phase III CROWN trial of advanced treatment-naive anaplastic lymphoma kinase (ALK)-positive non-small cell lung cancer (NSCLC). CROWN randomized patients to lorlatinib or crizotinib. Seventy-two with baseline brain metastases (BMs) had brain MRI every 8 weeks. We evaluated an FDA-cleared AI platform, VBrain, for automated lesion-level tracking of BMs and early tumor response (ETR) in relation to systemic progression-free survival (PFS), the primary endpoint of the trial. Progression-free survival was assessed by blinded central review, independent of AI. VBrain assessments were compared with 2 neuroradiologists (R1, R2). Early tumor response at the first on-treatment scan was derived from modified response evaluation criteria in solid tumors (mRECIST) target-lesion diameters and from AI-based all-lesion volumes (AI-ETR). Treatment-neutral Cox models related ETR to systemic PFS; discrimination was assessed using Uno's C and time-dependent AUC(<i>t</i>) with bootstrap uncertainty. Per mRECIST v1.1, VBrain, R1, and R2 measured 68, 32, and 51 targets, with high concordance ( <math xmlns="http://www.w3.org/1998/Math/MathML"><mi>r</mi></math>  = 0.94). Across 666 time-points, pairwise assessment discordance and time to intracranial progression did not differ between AI and readers. Beyond mRECIST, VBrain tracked 322 tumors. AI-ETR significantly stratified PFS (multivariable-adjusted hazard ratio [HR] 0.44, 95% CI, 0.23-0.86), whereas target-diameter ETR was not significant (adjusted HR 0.64, 95% CI, 0.29-1.41). Adding AI-ETR improved prediction vs a clinical base model (Δ<i>c</i>-index +0.07; AUC(<i>t</i>) 0.71 vs 0.55). The AI and human readers reached a high agreement in endpoint evaluations. Volumetric all-BMs ETR outperforming target-only mRECIST offers early prognostic information in advanced ALK-positive NSCLC. ClinicalTrials.gov identifier: NCT03052608.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.