{
  "abstract": "Please confirm that an ethics committee approval has been applied for or granted: Yes: I’m uploading the Ethics Committee Approval as a PDF file with this abstract submissionApplication for ESRA Abstract PrizesBackground and Aims Accurate ultrasound-guided identification of anatomical structures is crucial for regional anaesthesia. Ultrasound guided regional anaesthesia using Artificial Intelligence (AI) has the potential for training and clinical assistance with regional anaesthesia. This study compared AI with human clinicians in identifying ‘Plan A’ nerve block structures.Methods This qualitative, cross-sectional study utilized 35 ultrasound images across 7 ‘Plan A’ nerve blocks. ‘Plan A’ nerve blocks include Erector Spinae Plane, Rectus Sheath, Interscalene, Axillary Brachial Plexus, Femoral Nerve, Popliteal Sciatic and Distal Femoral Triangle blocks. Twenty registrars, 5 consultants, and an AI system (ScanNav™ Anatomy PNB, Intelligent Ultrasound, Cardiff, UK) identified 125 anatomical structures each, totalling 3250 observations, against a regional anaesthesia expert.Results The AI system achieved superior overall accuracy (94%) compared to the consultants (79%) and registrars (59%) in anatomical structure identification. Ground truth annotations were established by the expert for the final evaluation which revealed that the initial expert annotation obtained an accuracy of 98%. While consultants showed high accuracy for many structures, both human groups struggled with specific nerves (e.g., Median Nerve: consultants 36%, registrars 16%). AI maintained high accuracy even for structures poorly identified by humans, such as the Fascia Iliaca (AI 80%, consultants 28%, registrars 24%) and Radial Nerve (AI 100%, consultants 56%, registrars 25%).Abstract P339 Table 1The accuracy of each group’s responses compared to the expert (gold standard) responses by structure and overall accuracy. The number represents the number of responses that matched the experts. The percentage is the number of responses that matched the experts divided by the total number of responses by participants (expert excluded). Table depicting accuracy of groupsAbstract P339 Figure 1A binary heatmap depicting responses by participants to demonstrate whether these were the same or different to the expert (gold standard) responses. Heatmap depicting individual responsesAbstract P339 Figure 2A continuous heatmap of the accuracy of the responses by participants compared to the expert (gold standard) responses. Continuous heatmap depicting group accuracyConclusions AI demonstrates superior accuracy in identifying anatomical structures for ‘Plan A’ nerve blocks. This underscores AI’s utility as a robust tool for assisting and training clinicians in regional anaesthesia especially where access to expert led training is limited.",
  "authors": [
    {
      "affiliations": [
        "Anaesthesiology, University of the Witwatersrand, Johannesburg, South Africa"
      ],
      "name": "Ruan Roscoe Rheeders"
    },
    {
      "affiliations": [
        "Anaesthesiology, University of the Witwatersrand, Johannesburg, South Africa"
      ],
      "name": "Zainub Jooma"
    },
    {
      "affiliations": [
        "Anaesthesiology, University of the Witwatersrand, Johannesburg, South Africa"
      ],
      "name": "Celeste Quan"
    },
    {
      "affiliations": [
        "Anaesthesiology, University of the Witwatersrand, Johannesburg, South Africa"
      ],
      "name": "Laura Indiveri"
    }
  ],
  "title": "P339 Variability between anaesthetists and artificial intelligence in identification of anatomical structures for ultrasound guided regional anaesthesia: a cross-sectional study in an academic department in Johannesburg",
  "uid": "7a335604-c5e5-5bf7-8568-3c71597206c7"
}