Implant-Grade Neural Sensing with Artifact-Resilient Computational Intelligence Preserves Spatial Fidelity of HFO Biomarkers in Epilepsy.

High-frequency oscillations (HFOs) are clinically established biomarkers of epileptogenic tissue, yet their translation from hospital-based recording systems to implantable neural interfaces remains a major barrier to their utilization in adaptive neuromodulation in epilepsy. Here, we established a bedside translational framework using a benchtop implementation of a wireless implantable neural interface (Brain Interchange, BIC) to evaluate whether clinically meaningful HFO can be preserved under implant-grade sensing constraints. Ten patients with drug resistant epilepsy underwent simultaneous 24-hour intracranial EEG recordings using synchronized BICs and a clinical-grade amplifier. A sensitive detector identified candidate HFOs, while sparse signal processing and machine learning methods removed artifact related pseudo-HFOs. Despite its lower sampling rate, narrower analog bandwidth, and higher noise floor, the integrated sensing and computational intelligence framework preserved approximately 82% of clinical amplifier HFOs and maintained nearly identical spatial distribution of pathological activity. After pseudo-HFO elimination, BIC derived HFOs localized the seizure onset zone with 84% sensitivity and 90% specificity, comparable to the clinical amplifier despite lower fast ripple detection. These findings demonstrate that implant-grade neural sensing, when integrated with artifact resilient computational intelligence, can preserve clinically relevant HFO biomarkers and their spatial fidelity without reproducing every individual electrophysiological event. More broadly, this work establishes a translational framework for chronic tracking of evolving epileptogenic networks and future biomarker-guided adaptive neuromodulation.