An innovative learning-based framework for automated craniotomy planning in glioma resection

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Medical Imaging

Background:

• Glioma resection is complex due to tumor malignancy and prevalence, requiring precise craniotomy planning.
• Current automated planning methods lack explicit targets and geometric constraints for bone flap delineation.
• Minimizing neurological damage while maximizing tumor removal necessitates advanced craniotomy strategies.

Purpose of the Study:

• To develop an innovative learning-based framework for automated craniotomy planning in glioma resection.
• To address limitations in current automated methods by integrating imaging data and expert demonstrations.
• To determine optimal bone flap geometry for enhanced surgical intervention.

Main Methods:

• Utilized a reinforcement learning (RL) model integrating preoperative imaging and expert demonstrations.
• Incorporated self-supervised learning for implicit glioma quantification.
• Employed an encoding method for craniotomy patterns, a physics-based simulation engine, and imitation learning.

Main Results:

• Achieved a 92.31%±3.85% success rate with known craniotomy parameters.
• Demonstrated a 64.5% ±5.87% success rate in end-to-end planning from raw preoperative images.
• The method attained human-level performance in craniotomy planning.

Conclusions:

• The proposed learning-based framework shows significant potential for end-to-end automated craniotomy planning.
• This research offers a valuable reference for developing intelligent decision-support tools in neurosurgery.
• The AI-driven approach promises to enhance precision and outcomes in glioma resection procedures.