Physics in Medicine and Biology 2022

Phase Function Estimation

Phase Function Estimation from a Diffuse Optical Image via Deep Learning

Yuxuan Liang, Chuang Niu, Chen Wei, Shenghan Ren, Wenxiang Cong, Ge Wang

Current or former lab member. Select a name to see all of their projects.

Method overview of Phase Function Estimation
Example results: diffuse optical images with the ground-truth and estimated phase functions (Fig. 2 of the paper).

Abstract

The phase function describes how light is deflected when it scatters in tissue. This work estimates it with a convolutional neural network directly from a diffuse optical image, without assuming a fixed form. A Gaussian mixture model represents the phase function flexibly and still gives an analytic expression for sampling deflection angles in Monte Carlo simulation.

The method was validated on Monte Carlo-simulated reflectance images of typical biological tissues with different anisotropy factors. It reached a mean squared error of 0.01 for the phase function and a relative error of 3.28% for the anisotropy factor. The study also analyzes how field of view and spatial resolution affect the estimate.

What’s in the repository

  • A modified CUDAMCML program and scripts that generate the simulated diffuse reflectance data
  • Training, cross-validation and testing code
  • Scripts that produce the figures, and saved test results

Citation

@article{liang2022phase,
  title   = {Phase Function Estimation from a Diffuse Optical Image via Deep Learning},
  author  = {Liang, Yuxuan and Niu, Chuang and Wei, Chen and Ren, Shenghan and Cong, Wenxiang and Wang, Ge},
  journal = {Physics in Medicine and Biology},
  volume  = {67},
  number  = {7},
  pages   = {074001},
  year    = {2022},
  doi     = {10.1088/1361-6560/ac5b21}
}
Back to projects