Deep Learning for Instrumented Ultrasonic Tracking: From Synthetic Training Data to In Vivo Application.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Ultrasound-guided percutaneous procedures rely on accurate needle localization.
• Current instrumented ultrasonic tracking faces challenges with reduced B-mode frame rates and low signal-to-noise ratios, impacting needle tip accuracy.

Purpose of the Study:

• To develop a convolutional neural network (CNN) framework to enhance ultrasonic tracking performance.
• To maintain spatial resolution and improve signal quality using fewer tracking transmissions.

Main Methods:

• Implementation of a CNN framework utilizing realistic synthetic training data.
• Interleaving tracking transmissions with B-mode image acquisition.
• Reconstruction of tracking images from detected ultrasound pulses using a fiber-optic hydrophone.

Main Results:

• The CNN framework improved axial and lateral spatial resolutions significantly, even with an eightfold reduction in tracking transmissions.
• Enhanced signal quality and maintained spatial resolution were achieved.
• The trained network demonstrated effective performance on both synthetic and in vivo data.

Conclusions:

• The proposed CNN framework offers a significant improvement in ultrasonic tracking performance.
• This leads to faster image acquisition rates and increased needle localization accuracy in minimally invasive procedures.