EIDORS: Electrical Impedance Tomography and Diffuse Optical Tomography Reconstruction Software

Hosted by

Samples for the EIDORS Workshop

>>run /path/to/eidors3d/startup.m

EIDORS Workshop

• Introduction

  CODE

  • Downloading and Installing EIDORS and Netgen
  • Test that EIDORS is working
  • EIDORS data structures
    • fwd_model
    • inv_model
    • rec_model
    • data
    • image
  • EIDORS functions overview
  • Documentation

    solvers	forward and inverse solverse
    meshing	Interface to FEM libraries: Netgen / GMSH / distmesh
    models	FEM modelling tools
    external	Read write external file formats
    graphics	Graphics / Display functions
    algorithms	Algorithms to process EIT images (time series)
    sample_data	Some sample data (deprecated, now use data_contrib)
    examples	Examples (deprecated, now use tutorials)
    tools	low level EIDORS tools
    tests	Tests for debugging (deprecated, now in UNIT_TEST
    arch	Mex files and architecture specific code
    compatibility	Compatibility with older matlab versions
    deprecated	Older, deprecated EIDORS files
    overloads	Override matlab functions to give extra features

  • Contributed Data
  • Defaults
  • Caching

    tic;
    fmdl= ng_mk_cyl_models(1,[16,.5],0.1)
    toc;
    
    fmdl.stimulation = mk_stim_patterns(16,1,[0,1],[0,1],{},1);
    img = mk_image( fmdl, 1);
    
    tic;
    vh = fwd_solve(img);
    toc
    

  • Challenge Questions:
    • What data are from a cute baby on EIDORS? load and display these data
    • What is the speedup for larger FEM models?
    • Look at inv_solve_diff_GN_one_step - what part is cached?

  • "EIDORS basics" tutorials

  • Forward Models

    • Finite Element Models (FEMs)
    • EIDORS FEM structure

      >> fmdl= ng_mk_cyl_models(1,[16,.5],0.1);
      >> fmdl.stimulation = mk_stim_patterns(16,1,[0,1],[0,1],{},1) 
      fmdl =
                           nodes: [1958x3 double]
                           elems: [7849x4 double]
                        boundary: [2326x3 double]
                boundary_numbers: [2326x1 double]
                        gnd_node: 350
                    np_fwd_solve: [1x1 struct]
                            name: 'ng'
                       electrode: [1x16 struct]
                           solve: 'eidors_default'
                        jacobian: 'eidors_default'
                      system_mat: 'eidors_default'
                            type: 'fwd_model'
                         mat_idx: {[7849x1 double]}
          normalize_measurements: 0
                     stimulation: [1x16 struct]
      >> fmdl.stimulation(1)
      
      ans = 
      
           stimulation: 'Amp'
          stim_pattern: [16x1 double]
          meas_pattern: [13x16 double]
      
      >> fmdl.electrode(1)
      
      ans = 
      
          nodes: [114 115 116 117 118 119 415]
          z_contact: 0.0100
      

    • Use shape_library to obtain a human thorax contour
      shape = shape_library('get', 'adult_male','boundary');
    • make a 2d model with 16 electrodes (width 0.1) with electrode 1 on the sternum (remember the shape must be counter-clockwise)
      fmdl = ng_mk_2d_model({flipud(shape),0.1},[16 -.17],[0.1 20]); show_fem(fmdl)
    • make a similar 3d model
      fmdl = ng_mk_extruded_model({1, shape, [4,49]},[16,1,0.5],[0.05 0 -1 0 60]); show_fem(fmdl)
    • add contrast using elem_select
    • solve for internal voltages
    • visualize using calc_slices
    • visualize using show_3d_slices
    • calculate current
    • calculate Jacobian
    • visualize Jacobian
    • change current pattern to N/2-1
    • compare Jacobian
    • (FEM generation 2D and 3D, Netgen, sensitivity, voltage and current distributions, stimulation patterns )
    • Challenge Questions:
      • Compare the sensitivity for a slice of a 3D vs 2D tank. Why are they different?
      • Create images of current flow in the thorax with various values for lung conductivity. How do they vary?

    • Further Tutorials on Forward Modelling

      • FE model generation in Netgen
        • Cylindrical models
        • Human (extruded) thorax model
        • Pre-packaged models
        • Effects of mesh refinement
      • FE approximations
        • Higher order (2nd, 3rd) finite elements in 2D
        • Higher order (2nd) finite elements in 3D
      • Electrode modelling
        • Estimating and managing electrode errors
        • Contact impedance
      • Physics

  • Reconstructions

    • finemdl =
    • coarsemdl
    • imdl = mk_common_mdl

      >> fmdl = ng_mk_cyl_models(0,16,0.05);
      >> imdl = select_imdl(fmdl,{'Basic GN dif'})
      imdl =
                    name: 'Basic imdl from select_imdl'
                    type: 'inv_model'
                   solve: @inv_solve_diff_GN_one_step
          hyperparameter: [1x1 struct]
               RtR_prior: @prior_laplace
          jacobian_bkgnd: [1x1 struct]
            reconst_type: 'difference'
               fwd_model: [1x1 struct]
      

    • select_imdl(imdl, 'Basic GN')
    • inv_solve with inverse crime
    • inv_solve w/o inverse crime
    • show_slices
    • use mk_pixel_grid to setup a dual model
    • manually change hyperparam
    • select_imdl('Choose NF')

      >> imdl = mk_common_model('c2c2',16);
      >> imdl = select_imdl(imdl,{'Choose NF=1.0'})
      

    • ( 2D and 3D FEMS, inverse crime, visualization functions, hyperparameter choice, functions: select_imdl , functions: mk_pixel_grid )
    • Challenge Questions: Using simulated and real data, test the effect of
      • different model shapes
      • different priors
      • hyperparameter.value
      • noise figure

    • Further Tutorials on Reconstruction

      • Dual modelling
        • Introduction to dual models
        • Reconstruct a 3D volume onto a triangulated mesh
        • Reconstruct a 3D volume onto a rectangular mesh
      • Absolute reconstructions
        • Iterative Gauss-Newton solvers in 3D
        • Iterative vs one-step Gauss-Newton solvers using RPI data
        • Total Variation solvers in 2D

  • GREIT

    • String interface
    • Custom model (for a 2 ring cyl)
    • Use ng_write_opt to increase minimum elem size and build a coarse pig model
    • imdl1 = mk_GREIT_model (fmdl)
    • Set the lungs to have 0.3 conductivity
    • vh =
    • vi =
    • compare reconstructions
    • ( GREIT algorithm, adapting shape, background conductivity, working with sample data )

    • Further Tutorials on GREIT

      • 2D human thorax
      • Multiple electrode layers
      • 3D reconstruction

  • Live data

    • Working with live data from tank
    • GREIT competition
    • How to contribute data?

  • Assorted Tutorials on specific applications of EIT

    • Respiratory Monitoring

      • Lung recruitment in ARDS
      • Lung injury after PEEP trial images
      • Ventilation changes after lung injury
      • Neonate lung images
      • Human shallow breathing
      • Stomach emptying images

    • Geophysical

      • Pont Péan region
      • Borehole EIT
      • EIT in a tunnel
      • 3D surface electrode system
      • Stimulation patterns and pseudo-section representation

    Other interesting Tutorials

    Here are useful tutorials to start with:

    • EIDORS Basics

    • Absolute and difference solvers

    • Lung injury images

    • Netgen FEM models

    • GREIT