Neuro semantic verification improves trust in glioblastoma AI findings
Evidence-Bound Reasoning: Neuro-Semantic Verification of Biomedical AI in Glioblastoma Radiogenomics
Computation and Language
Summary
AI can create believable explanations in cancer imaging but often doesn’t check if those explanations match the patient’s actual data. The authors created a system that turns brain tumor image measurements into evidence that a computer can verify logically. This makes sure AI claims about a patient’s tumor are accurate and backed by real evidence, even when AI’s ability to predict outcomes is not perfect. Their approach works consistently across different hospitals and can work alongside large language models to structure explanations while verifying them with strict checks.
What this means in practice
- •For medical imaging engineers: Implement patient-specific evidence verification in brain tumor imaging AI to improve reliability of automated diagnostic explanations.
- •For clinical ai deployment teams: Integrate deterministic verification layers in biomedical AI pipelines to ensure claims about tumor characteristics match source patient data across multiple centers.
Authors
Mariya Miteva, Maria Nisheva-Pavlova
Abstract
Background: Biomedical AI can generate plausible explanations without reliably verifying whether each statement is supported by patient-specific evidence. We developed a neuro-semantic verification framework that converts radiomic measurements into addressable evidence records and machine-checkable claims. Methods: UPenn-GBM radiomics were aligned with de novo CaPTk extraction from standardized MRI and expert-validated segmentations in an independent multicenter cohort. The shared space comprised 1,728 features from T1, T1GD, T2, and FLAIR MRI across three tumor regions. Reference-defined semantic states were derived from 611 UPenn cases. We evaluated cross-cohort transportability, model-linked provenance, deterministic verification, controlled predictive degradation, and an LLM claim-extraction pilot; MGMT prediction served only as a transport stress test. Results: Median semantic-state agreement was 0.786 (weighted kappa 0.709), ranging from 0.918 for morphologic to 0.252 for intensity features. The external evidence ledger contained 1,655 model-linked records for 331 patients. The verifier achieved 100% exact-set accuracy in a 6,620-claim corruption benchmark. In a 24-case pilot, GPT-5.6 Sol reproduced 72/72 prespecified atomic claims, and the frozen verifier recovered 24/24 expected conditions. During controlled degradation, ROC AUC declined from 0.899 to 0.500 while verification accuracy remained 1.000. External MGMT discrimination was weak (ROC AUC 0.543). Conclusions: Verifiability can be engineered and evaluated independently of predictive performance. LLMs may structure explanations, while final evidence-consistency checking remains deterministic.