harmening/pyreite

🧠 Pythonic, Yet Rudimental, Electrical Impedance Tomography Expert

Pythonic, Yet Rudimental, Electrical Impedance Tomography Expert (pyreite ☠️)

Pythonic head modeling for Electrical Impedance Tomography (EIT):

Get up and running

Prerequisites

• Python3.6
• OpenMEEG 2.4 with python wrapping: compile with "-DENABLE_PYTHON=ON"

Install pyreite

git clone https://github.com/harmening/pyreite.git
cd pyreite
pip install -r requirements.txt
python setup.py install

docker 🐳

Build pyreite image

$ docker build -t pyreite .

or pull from docker hub

$ docker pull harmening/pyreite:v0.2

Example EIT simulation

import os.path.join as pth
from collections import OrderedDict
from pyreite.OpenMEEGHead import OpenMEEGHead
from pyreite.data_io import load_tri, load_elecs_dips_txt


# Load Colins surface meshes
geom = OrderedDict()
for tissue in ['cortex', 'csf', 'skull', 'scalp']:
    geom[tissue] = load_tri(pth('tests', 'test_data', tissue+'.tri'))

# Define conductivity values [S/m]
cond = {'cortex': 0.201, 'csf': 1.65, 'skull': 0.01, 'scalp': 0.465}

# Load electrode positions
sens = load_elecs_dips_txt(pth('tests', 'test_data', 'electrodes_aligned.txt'))


# Create EIT head model
model = OpenMEEGHead(cond, geom, sens)

# Calculate EIT voltage measurement array
V = model.V

pyreite's Issues

Slow jacobian + hessian computation for large matrices

Somehow numpy.dot seems to be ~10 times slower (170sec vs 14sec) for large head models (4x1922 vertices vs 4x800).
This slows down the whole optimization (especially for the hessian) by a factor >> 10 due to the many dot products in the hessian derivatives (jacobian is less severe).
Tested on HPC-Cluster and local MacBook Pro (factor was even larger on IOS).