Neural Network-Based Depth Retrieval of Fluorescent Targets From Wide-Field Fluorescence Intensity Images in...

Area of Science:

• Biomedical optics
• Medical imaging
• Computational biology

Background:

• Fluorescence imaging (FI) is crucial for in vivo and cellular studies.
• Accurate depth determination of fluorescent targets in vivo is challenging due to photon scattering and absorption in biological tissues.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based framework for accurate depth estimation of near-infrared (NIR) fluorescent targets in turbid media.
• To overcome the limitations of traditional fluorescence imaging in determining target depth within biological tissues.

Main Methods:

• Developed an optimized Monte Carlo (MC) model to simulate NIR fluorescence photon propagation.
• Generated datasets of fluorescence images corresponding to target depths from 0.1 to 1 cm.
• Experimentally validated the ML framework using tissue-mimicking phantoms and an in vivo model with fluorescent gold nanostructures, employing computed tomography (CT) and wide-field FI.

Main Results:

• The ML framework achieved accurate depth estimation for NIR fluorescent targets in turbid media.
• Quantitative validation using rounded accuracy and root mean squared error (RMSE) demonstrated the framework's effectiveness.
• Successful experimental validation was performed over a depth range of 0.1-0.7 cm in phantoms and in vivo models.

Conclusions:

• The developed ML framework, supported by MC simulations, provides a robust solution for accurate in vivo depth estimation of NIR fluorescent targets.
• This approach has the potential to significantly advance biomedical optics and fluorescence-guided interventions.
• The study highlights the synergy between computational modeling and machine learning for solving complex challenges in biological imaging.