Knowledge graph representation of the mappings between seizure semiology and epileptogenic zones

Precise epileptogenic zone (EZ) localization remains challenging for epilepsy surgery planning. While seizure semiology provides valuable localization information, subjective interpretation and inter-observer variability limit clinical utility. We developed a computational framework utilizing knowledge graph architectures to analyze ictal semiology-EZ relationships systematically. We constructed a semiology-EZ knowledge graph from 852 clinical cases extracted from peer-reviewed literature. GPT-4o facilitated automated extraction and standardization of semiological terminology. Statistical modeling, including Gaussian mixture modeling and Bayesian inference, quantified association strengths between semiological features and anatomical regions. SeeKr, a query platform, was developed to generate EZ localization predictions from patient symptoms with confidence measures. Expert epileptologists evaluated key semiology-to-brain region mappings using a four-point assessment scale. The framework achieved an average correctness score of 2 (1 = strongly agree, 4 = strongly disagree), indicating general clinical plausibility, with most associations falling within the “likely agree” range. Minor inaccuracies involved partial identification when seizures affected multiple regions and slight misclassifications of relationship intensity. This represents the first knowledge graph-based systematic analysis of semiology-EZ relationships. The framework provides a data-driven approach for objective semiological analysis with reasonable clinical accuracy. The methodology offers potential utility as a supplementary diagnostic tool for surgical planning, though further clinical validation is warranted.