Utilizing Deep Learning to Identify an Ultrasound-guided Nerve Block Target Zone.

Ultrasound guided nerve blocks are increasingly being used in perioperative care as a means of safely delivering analgesia. Unfortunately, identifying nerves in ultrasound images presents a challenging task for novice anesthesiologists. Drawing from online resources, here we attempted to address this issue by developing a deep learning algorithm capable of automatically identifying the transversus abdominis plane region in ultrasound images. Training of our dataset was done using the U-Net architecture and artificial augmentation was done to optimize our training dataset. The Dice score coefficient was used to evaluate our model, with further evaluation against a test set composed of manually drawn labels from a pool of (n=10) expert anesthesiologists.Across all labelers the model achieved a global Dice score of 73.31% over the entire test set. These preliminary results highlight the potential effectiveness of this model as a future ultrasound decision support system in the field of anesthesia.