From Consensus to Standardization: Evaluating Deep Learning for Nerve Block Segmentation in Ultrasound Imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Anesthesiology

Background:

• Deep learning (DL) offers automated nerve identification in ultrasound images.
• This technology learns from expert-labeled data to detect and highlight nerves.
• Accurate nerve identification is crucial for ultrasound-guided nerve blocks.

Purpose of the Study:

• To evaluate the performance of deep learning models in identifying nerves for ultrasound-guided nerve blocks.
• To compare the effectiveness of various convolutional neural network architectures for nerve segmentation.

Main Methods:

• 3594 ultrasound images from public sources across 9 nerve block regions were collected.
• 10 images per region were used for testing, labeled by 10 anesthesiologists.
• 3504 images were augmented to create a 25,000-image training set per region, with 908 negative images included.

Main Results:

• The R2U-Net model achieved the highest average Dice score (0.7619), outperforming other models.
• Statistically significant performance differences were noted for the Transversus Abdominis Plane (TAP) nerve region.
• Expert evaluations showed high accuracy, with agreement for needle insertion in 100% of Popliteal nerve predictions.

Conclusions:

• Deep learning models, particularly R2U-Net, show potential for consistent nerve segmentation in ultrasound-guided procedures.
• These findings support the integration of DL tools to enhance nerve block accuracy and safety.
• Further validation and refinement of DL models can improve clinical application in regional anesthesia.