Flexible needle puncture path planning for liver tumors based on deep reinforcement learning.

Area of Science:

• Medical Imaging
• Surgical Planning
• Artificial Intelligence

Background:

• Minimally invasive surgery is standard for liver tumors.
• Image-guided ablation needles are crucial for precise tumor treatment.
• Accurate path planning is vital to avoid damage to surrounding organs.

Purpose of the Study:

• To propose a novel path planning method for CT-guided ablation needles in liver tumor surgery.
• To enhance the safety and efficiency of liver tumor puncture procedures.
• To provide surgeons with a reliable tool for preoperative planning.

Main Methods:

• Voxel classification and 3D model reconstruction of liver and hepatic vessels from CT data.
• Multi-agent reinforcement learning for optimal needle entry point selection.
• Double deep Q-learning network (DDQN) for incremental path optimization.

Main Results:

• The method successfully generates optimal puncture paths for liver tumors.
• Training on multi-scale models improved network convergence and performance.
• Clinical relevance was validated by human surgeons.

Conclusions:

• The developed method robustly identifies optimal puncture paths for liver tumors.
• This approach can serve as a valuable reference for surgeons during preoperative planning.
• The technique holds promise for improving outcomes in minimally invasive liver tumor surgery.