From quantitative features to imaging signs: agentic AI-driven autonomous discovery in glioblastoma

Abstract The development of semantic imaging signs is time-consuming and relies heavily on clinicians’ experiential knowledge, whereas the abundance of quantitative radiomic features often poses challenges in interpretability and clinical translation. We developed an agentic AI pipeline to autonomously translate quantitative morphological features into semantic signs. The pipeline autonomously performed end-to-end processing—from descriptive statistical profiling and semantic sign translation to visualization. In a development dataset of 106 supratentorial glioblastoma (GBM) cases, the pipeline generated eight candidate signs; the three most consistently proposed signs were the cauliflower, eggplant, and pancake signs. Two radiologists independently validated the cauliflower sign on an external cohort of 50 GBMs and 50 metastases, achieving AUCs of 0.73 (95% CI: 0.64–0.82) and 0.77 (95% CI: 0.69–0.85), respectively (mean 0.75) for tumor differentiation. These findings demonstrate that agentic AI can directly derive semantic imaging signs from quantitative features, offering a systematic and data-driven framework for sign discovery that may enhance the interpretability and clinical translation of radiomics.