Multierror learning enhanced fluorescence diffusion tomography in reflection geometry

Area of Science:

• Biomedical Optics
• Medical Imaging
• Computational Biology

Background:

• Accurate 3D localization of intravital fluorescent probes is crucial for understanding neural circuits and tumor progression.
• Fluorescence diffusion tomography in reflection geometry (rFDT) enables deep tissue imaging but faces challenges with detection sensitivity and tissue heterogeneity.
• Existing methods struggle with precise localization due to perturbations in photon diffusion paths.

Purpose of the Study:

• To develop a novel method for precise 3D localization and sensing of intravital fluorescent probes at subcentimeter depths.
• To enhance the accuracy and fidelity of fluorescence diffusion tomography in reflection geometry.
• To overcome limitations of current rFDT techniques in complex biological tissues.

Main Methods:

• Introduced multierror-learning-enhanced rFDT (MEL-rFDT), a deep learning approach.
• Embedded physical priors from photon transport models and spatial attention into the neural network.
• Trained the network using limited <i>in silico</i> samples for adaptive error compensation.

Main Results:

• MEL-rFDT achieved high-fidelity reconstruction of intravital fluorescent probes.
• Demonstrated unprecedented 3D localization and sensing accuracy in <i>ex vivo</i> brain and <i>in vivo</i> subcutaneous tumor imaging in mice.
• Showcased volumetric and functional generalization capabilities across different samples.

Conclusions:

• MEL-rFDT significantly improves the precision of deep tissue fluorescence imaging.
• The method effectively compensates for depth-dependent sensitivity and model imperfections.
• Enables advanced applications in intraoperative pathology, dynamic imaging, and healthcare decision-making.