Phase function estimation from a diffuse optical image via deep learning.

Area of Science:

• Biomedical Optics
• Computational Imaging
• Machine Learning

Background:

• Light propagation modeling in biological tissues relies on accurate phase functions for Monte Carlo (MC) simulations.
• Current methods often assume predefined phase function forms (e.g., Henyey-Greenstein) and estimate parameters.
• Determining the optimal form of the phase function directly from diffuse optical images remains an open challenge.

Purpose of the Study:

• To develop a data-driven approach using convolutional neural networks (CNNs) to estimate the form of the scattering phase function.
• To enable accurate phase function estimation from diffuse optical images without assuming a specific analytical form.
• To validate the proposed method on simulated biological tissue data.

Main Methods:

• A convolutional neural network (CNN) was designed to infer the phase function directly from diffuse optical images.
• A Gaussian mixture model (GMM) was employed to represent the phase function generally, facilitating accurate parameter learning.
• The CNN model was trained and validated using MC-simulated reflectance images of biological tissues.

Main Results:

• The proposed CNN-based method successfully estimated the scattering phase function form from diffuse optical images.
• Validation on simulated data showed a mean squared error of 0.01 for the phase function.
• The relative error for the anisotropy factor estimation was 3.28%, demonstrating high accuracy.

Conclusions:

• This work presents the first data-driven, CNN-based inverse MC model for estimating scattering phase function forms.
• The findings offer insights into the impact of field of view and spatial resolution on phase function estimation.
• The study provides guidelines for optimizing experimental protocols in diffuse optical imaging applications.