Enhancing Photoacoustic Visualisation of Clinical Needles with Deep Learning

Photoacoustic imaging has shown great potential in guiding various minimally invasive procedures by providing complementary information to ultrasound imaging, visualising critical tissue targets as well as surgical tools such as metallic needles with rich optical contrast. The use of light emitting diodes (LEDs) as excitation sources further accelerates the clinical translation of this modality due to its favorable affordability, portability, and cost-efficiency but results in the degradation of image quality associated with low pulse energy. In this work, we propose a framework based on a modified U-Net to enhance the visualisation of clinical metallic needles with a commercial LED-based photoacoustic and ultrasound imaging system. This framework included the generation of semi-synthetic datasets combining both simulated data and in vivo measurements. Evaluation of the trained neural network was performed with needle insertions into a blood-vessel-mimicking phantom, and pork joint ex vivo tissue. This deep learning-based framework significantly enhanced the visualisation of the needle with photoacoustic imaging, achieving 4.3- and 3.2-times higher signal-to-noise ratios (SNRs) compared to conventional reconstructions, which could be helpful for guiding minimally invasive procedures.