ChatGPT-4 reliability when applied to the regional anesthesia technique sonoanatomy.
Authors
Affiliations (4)
Affiliations (4)
- Department of Anesthesiology, Istanbul Health Science University Kanuni Sultan Süleyman Education and Training Hospital, Istanbul, Türkiye - [email protected].
- Department of Anesthesiology, Basaksehir Cam ve Sakura City Hospital, Istanbul, Türkiye.
- Department of Anesthesiology, Sisli Hamidiye Etfal Education and Training Hospital, Istanbul, Türkiye.
- Department of Anesthesiology, Istanbul Health Science University Kanuni Sultan Süleyman Education and Training Hospital, Istanbul, Türkiye.
Abstract
Ultrasound-guided regional anesthesia is commonly used to enhance perioperative analgesia. However, the technique is limited by inter-operator variability in identifying anatomical structures. Recent advances in artificial intelligence, particularly large language models, have raised the possibility of providing real-time interpretation support. This prospective observational study was conducted at a tertiary-level hospital between 20/10/2024, and 28/02/2025. A total of 111 adult patients (aged 18-70) undergoing ultrasound-guided regional anesthesia were included. Ultrasound images were obtained intraoperatively during routine care. Exclusion criteria included poor image quality, incomplete clinical metadata, and non-standardized block protocols. Each image was submitted to ChatGPT-4 with standardized prompts requesting: 1) identification of the block type; 2) labeling of key anatomical landmarks; and 3) assessment of block success based on local anesthetic spread. Model outputs were compared to expert anesthesiologist assessments. A total of 147 ultrasound images were analyzed. ChatGPT-4 correctly identified anatomical landmarks in 141 cases (95.9%), classified block types in 107 cases (72.8%), and predicted block success in 138 cases (93.9%). Model performance was highest in commonly performed blocks such as transversus abdominis plane (TAP), erector spinae plane (ESP), and femoral nerve blocks. Accuracy was relatively lower in more complex or less frequently performed blocks. ChatGPT-4 demonstrated high accuracy in identifying anatomical structures and predicting block success on ultrasound images. While performance in classifying certain block types was lower, these findings support the potential of large language models as decision-support tools in ultrasound-guided regional anesthesia.