Deep Learning on Ultrasound Images Visualizes the Femoral Nerve with Good Precision

Area of Science:

• Medical imaging
• Artificial intelligence in medicine
• Anesthesiology

Background:

• Hip fractures are a growing global health concern, projected to reach 6 million annually by 2050.
• Regional nerve blocks offer advantages for hip fracture pain management but are underutilized due to training challenges.
• Ultrasound-guided nerve blocks are effective, but success rates can be improved with advanced techniques.

Purpose of the Study:

• To develop and evaluate a deep learning semantic segmentation model for identifying the femoral nerve in ultrasound images.
• To enhance the accuracy and accessibility of ultrasound-guided regional anesthesia for hip fracture pain.
• To provide a potential solution for improving nerve block training and clinical application.

Main Methods:

• A U-net architecture deep learning model was employed for semantic segmentation.
• A dataset of 1410 ultrasound images from 48 patients was utilized.
• Images were manually annotated by a clinical professional, and the model was trained for 350 epochs.
• Model performance was assessed using 10-fold cross-validation.

Main Results:

• The deep learning model achieved a mean Intersection over Union (IoU) of 74% for femoral nerve identification.
• The interquartile range for IoU was 0.66-0.81, indicating robust performance.
• The model demonstrated significant accuracy in segmenting the target nerve in ultrasound images.

Conclusions:

• Deep learning semantic segmentation is a viable approach for accurate femoral nerve identification in ultrasound images.
• AI-powered ultrasound guidance can potentially increase nerve block success rates and improve training.
• This technology may help overcome barriers to the wider adoption of regional anesthesia for hip fracture pain management.