Flexible needle insertion has emerged as a promising technique for minimally invasive procedures, yet their curved trajectories make feasible path planning challenging. In this study, we propose a novel deep learning enhanced particle swarm optimization (EPSO) algorithm for puncture path planning. A dense network-based model is first trained to generate a spatial probability distribution of optimal paths. This guidance is then integrated into the key components of PSO—initialization, and velocity update—to enhance sampling efficiency and trajectory quality. The proposed method generates smoother, shorter, and safer needle trajectories while maintaining rapid convergence performance. Experiments demonstrate the effectiveness and generalizability of the approach across diverse scenarios, suggesting its potential for clinical application in image guided interventions.