A physical information neural network-based path planning method for flexible needle puncture of soft tissues

Flexible needle puncture is a critical minimally invasive technique where path planning accuracy directly impacts clinical outcomes. Traditional finite element methods are computationally complex and inadequate for real-time procedures.

We propose a physics-informed neural network (PINN) approach for soft tissue puncture path planning. PINN models generate puncturable regions where rapidly-exploring random tree star (RRT*) performs path optimization using neural network-derived soft tissue mechanics models, avoiding collision risks. Progressive learning control strategies provide real-time path optimization, ensuring accurate needle targeting.

Experimental results demonstrate dynamic puncture path correction with errors <1 mm, meeting clinical requirements. The progressive learning control strategy effectively optimizes path prediction models through data analysis.

The combined PINN-RRT* approach addresses modeling complexity, path planning difficulty, real-time adaptability, and expert dependence in traditional puncture techniques, significantly improving safety and efficiency.