BPSegSys: A Brachial Plexus Nerve Trunk Segmentation System Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Anesthesiology

Background:

• Ultrasound-guided nerve block anesthesia (UGNB) relies on precise nerve identification.
• Current methods require significant expertise, and real-time visualization is crucial.
• Deep learning offers potential for automating and improving nerve identification in UGNB.

Purpose of the Study:

• To develop and evaluate a deep learning system for automated brachial plexus (BP) trunk segmentation.
• To establish a public dataset for BP identification research.
• To compare the system's performance against experienced physicians.

Main Methods:

• Creation of a public dataset with 320 ultrasound images of the brachial plexus.
• Annotation of brachial plexus trunks by three experienced physicians.
• Development of a deep learning-based system (BPSegSys) for BP segmentation.

Main Results:

• BPSegSys achieved experienced-doctor-level nerve identification performance.
• Intersection-over-union (IoU) scores for BPSegSys exceeded those of experienced doctors across dataset groups (e.g., 0.5350 vs. 0.5205).
• The system demonstrated up to a 27% improvement in IoU for brachial plexus trunk identification.

Conclusions:

• A novel deep learning system (BPSegSys) was developed for brachial plexus trunk identification.
• BPSegSys demonstrates doctor-level performance, enhancing accuracy and efficiency in UGNB.
• The system holds significant potential for clinical application in nerve block anesthesia.