Peripheral Nerve Segmentation in Ultrasound Images Using Conditioned U-Net

Peripheral Nerve Blocking (PNB) is a regional anesthesia procedure that delivers an anesthetic in the proximity of a nerve to avoid nociceptive transmission. Anesthesiologists have widely used ultrasound images to guide the PNB due to their low cost, non-invasivity, and lack of radiation. Due to the difficulties in visually locating the target nerve, automatic nerve segmentation systems attempt to support the specialist to perform a successful nerve block. This work introduces a deep neural network for automatic nerve segmentation in ultrasound images. The proposed approach consists of a conditioned U-Net model that includes the kind of target nerve as a second input allowing the network to learn new features to improve the segmentation. The model is trained and tested on a dataset holding four different peripheral nerves, achieving an average Dice coefficient of 0.70. Results show that the proposed C-UNet outperforms the conventional U-Net, benefiting the ultrasound-guided regional anesthesia.