Autonomous Planning and Intraoperative Augmented Reality Navigation for Neurosurgery

Abstract

Neurosurgical interventions in the brain are challenging due to delicate anatomical structures. During surgery, precise navigation of surgical instruments supports surgeons and allows prevention of adverse events. Here, an augmented reality-based navigation aid with automated segmentation of risk structures and path planning is presented. Superimposed patient models are visualized during neurosurgical interventions on the example of the ventricular puncture. The proposed system is experimentally validated in a realistic operating room scenario with expert neurosurgeons to determine its quality of support as well as its potential for clinical translation. The automated segmentation reaches a F1-Score of 95-99%. Paths are planned correctly in 93.4%. The entire process enables navigation aid in under five minutes. Validation shows that the system allows for a puncture success rate of 81.7% with mean accuracy of 4.8 ± 2.5 mm. A control group who performed the standard-of-care procedure reached a rate of 71.7% with 6.5 ± 2.4 mm accuracy. Acceptability analysis shows that 85.7% of the participating surgeons approve of the system's convenience and 92.9% expect accuracy improvement. The presented navigation aid for ventricular puncture enables automated surgical planning and may improve accuracy and success rates of neurosurgical interventions.