Related Experiment Video
Updated: Sep 25, 2025
09:31
High-Speed Ultraviolet Photoacoustic Microscopy for Histological Imaging with Virtual-Staining assisted by Deep Learning
Published on: April 28, 2022
3.2K
Improving needle visibility in LED-based photoacoustic imaging using deep learning with semi-synthetic datasets
Mengjie Shi1, Tianrui Zhao1, Simeon J West2
1School of Biomedical Engineering and Imaging Sciences, King's College London, London SE1 7EH, United Kingdom.
Photoacoustics
|May 2, 2022
Summary
This study introduces a deep learning framework to enhance metallic needle visibility in LED-based photoacoustic imaging. The method improves needle detection during minimally invasive procedures, reducing potential complications.
Area of Science:
- Medical Imaging
- Biomedical Engineering
- Artificial Intelligence
Background:
- Photoacoustic imaging (PAI) shows promise for guiding minimally invasive procedures by identifying tissue targets and medical devices.
- Light-emitting diodes (LEDs) offer an affordable and portable excitation source for PAI, accelerating clinical translation.
- Needle visibility in LED-based PAI is often limited by low optical fluence, hindering accurate guidance.
Purpose of the Study:
- To develop and evaluate a deep learning framework using U-Net to enhance the visibility of metallic needles in LED-based photoacoustic and ultrasound imaging.
- To address challenges in training data generation by creating semi-synthetic datasets combining simulated needle features and in vivo tissue data.
- To improve the accuracy and safety of percutaneous needle insertions through enhanced needle visualization.
Main Methods:
- A U-Net based deep learning framework was designed for image enhancement.
- Semi-synthetic training datasets were generated using simulated data for needles and in vivo measurements for tissue background.
- The framework was evaluated using phantom studies, ex vivo tissue samples, and in vivo human volunteer measurements.
Main Results:
- The deep learning framework significantly improved metallic needle visibility in photoacoustic imaging.
- Improvements included a 5.8-fold increase in signal-to-noise ratio and a 4.5-fold improvement in modified Hausdorff distance compared to conventional reconstruction.
- The method effectively suppressed background noise and image artifacts in vivo.
Conclusions:
- The proposed deep learning framework substantially enhances metallic needle visibility in LED-based photoacoustic imaging.
- This advancement can aid in reducing complications associated with percutaneous needle insertions by improving real-time needle identification.
- The framework offers a viable solution for clinical translation of photoacoustic imaging in guided procedures.
