Artificial Intelligence for Ultrasound Image Acquisition and Interpretation in Regional Anesthesia: A Narrative Review of Current Evidence

UGRA (ultrasound guided regional anesthesia) principle is the process of ultrasound acquisition and interpretation, which should be used to delineate sono-anatomy and guide needle insertion.In many cases, however, the interpretation of the images is still difficult, especially for the less experienced, and some experts disagree in their interpretation.Artificial intelligence (AI) in medical image interpretation and ultrasound, particularly deep learning (DL), is now more and more utilized.AI based devices have developed which use color overlays or segmentation maps to provide real-time B-mode ultrasound information with the anatomical structures highlighted during UGRA scanning.This narrative review aims to consolidate the existing evidence regarding the use of AI in ultrasound imaging acquisition and interpretation (ultrasound-AI) in the field of regional anesthesia, assessing its accuracy, clinical application, training impact and perceived safety.They carried out the narrative review of the literature restricting it to research of efficacy of AI-enabled devices and systems in the UGRA setting.Constraints imposed on the studies included were: the studies included the assessment of AI-based anatomical identification, scanning performance with and without the inclusion of AI support, training programmes involving AI, and the validation of the accuracy of AI models with external assessments.The quality of the AI highlighting, accurate block view acquisition, anatomical structure identification accuracy, user feedback, modified risk assessment of the patient, and clinical complication rates were key outcomes.The overall accuracy rate of the models using AI for identifying eight anatomical landmarks across nine peripheral nerve block regions was 93.5%" with a 3.5% false-positive rate and a 3.0% false-negative rate.Non-experts who used an AI device had significantly higher results with regards to the correct block view acquisition (90.3% vs. 75.1%,p = 0.031) and correct anatomical structure identification (88.8% vs. 77.4%,p = 0.002).Non-experts said that AI was more helpful for training, whereas experts said it was more helpful for teaching.AI training was significantly correlated with a reduction in paresthesia from puncture (4.12% vs. 14.06%,p < 0.001).AI highlighted in the expert panels was shown to decrease the perceived risk of block failure by 81.3% and the risk of needle trauma to safety-critical structures by 62.9 -86.2% in cases.AI-powered devices can help augment ultrasound image acquisition and interpretation in UGRA, complement training of non-experts and reduce perceived complications risks.But what evidence is there now is incomplete, and mostly from a handful of research groups.Further large-scale and multicenter randomized controlled trials with clinical patient outcomes are warranted to confirm these benefits.