ChatGPT Matches Nuclear Medicine Experts on FDG-PET/CT, But the Real Question Is Clinical Trust

The headline result is striking because FDG-PET/CT interpretation sits at the intersection of pattern recognition, clinical context, and nuanced reporting. If a general-purpose language model can perform at the level of experts in a research setting, it further erodes the old assumption that frontier AI is limited to text generation.

Still, the clinical bar is much higher than benchmark parity. Nuclear medicine decisions often depend on protocol details, prior studies, treatment history, and subtle differences in lesion behavior. A model that appears strong in a study may still struggle with the edge cases and accountability requirements that matter in real practice.

This is why the study is important less as a deployment signal and more as a proof of capability. It suggests that the frontier is moving toward multimodal systems that can reason over imaging findings and medical language together. But capability alone is not enough; reliability, traceability, and governance determine whether such systems can become clinical tools.

For health systems, the lesson is to prepare for a world in which AI can be surprisingly good at specialist tasks, while remaining hard to trust operationally. The next competitive advantage may lie in wrapping these models with the validation, guardrails, and audit trails that hospitals require.