Handheld interventional ultrasound/photoacoustic puncture needle navigation based on deep learning segmentation

Abstract

Interventional ultrasound (US) has challenges in accurate localization of the puncture needle due to intrinsic acoustic interferences, which lead to blurred, indistinct, and even invisible needles in handheld linear array transducer-based US navigation, especially the incorrect needle tip positioning. Photoacoustic (PA) imaging can provide complementary image contrast, without additional data acquisition. Herein, we proposed an internal illumination to solely light up the needle tip in PA imaging. Then deep-learning-based feature segmentation alleviates acoustic interferences, enhancing the needle shaft-tip visibility. Further, needle shaft-tip compensation aligned the needle shaft in US image and the needle tip in the PA image. The experiments on phantom, ex vivo chicken breast, preclinical radiofrequency ablation and in vivo biopsy of sentinel lymph nodes were piloted. The target registration error can reach the submillimeter level, achieving precise puncture needle tracking ability with in-plane US/PA navigation.