Fully Automated and Noise-Robust SEEG Electrode Localization under Practical Clinical Conditions

Abstract

Stereoelectroencephalography (SEEG) provides direct intracranial recordings of epileptic activity with high spatial and temporal resolution and is widely regarded as the gold standard for presurgical evaluation of drug-resistant epilepsy. Accurate localization of SEEG contacts is essential for seizure onset analysis and surgical planning. However, existing localization approaches are often semi-automated, sensitive to imaging quality, and require manual intervention, limiting scalability and routine clinical deployment. We present a fully automated and noise-robust SEEG contact localization pipeline that integrates adaptive artifact suppression, data-driven thresholding, and flexible geometry-based reconstruction. The method was evaluated on six patients from two centers, including diverse resolution CT acquisitions, comprising 57 electrodes and 652 contacts in total. The proposed framework successfully reconstructed all electrodes and achieved an overall recall of 98.3% and precision of 97.1% for contact localization compared with expert labeling, with a mean deviation of 0.248 {+/-} 0.066 mm. Processing time was approximately two minutes per subject on standard hardware. These results demonstrate reliable performance across heterogeneous imaging conditions. By reducing processing time from hours of manual effort to approximately two minutes per subject, the proposed fully automated workflow facilitates routine clinical deployment and scalable multi-center studies.