Electrical Impedance Tomography (EIT for friends) is a non-invasive medical imaging method that generates images of the interior electrical properties of a subject from voltage measurements acquired from electrodes placed outside the surface of that object. EIT is a low-cost, radiation-free imaging method and, thus, allows for monitoring in real-time and long-term without any harmful effects. Because of this EIT has been gaining particular importance in cases of pulmonary ventilation and early detection of breast cancer.

The EIT dataset contains \(1000\) voltage measurements obtained via simulation of current propagation for different electrical properties. The dataset was generated with the detection of breast cancer in mind. Hence, all simulations were obtained over a circular domain and with close relation to the electrical properties of human tissues.

The electrical properties of a domain are defined by a conductivity profile. In Fig. 1, we provide an example present in the EIT dataset.

Fig. 1: Demonstration of some dataset simulations. The darker region symbolizes cancer tissue inside an otherwise healthy breast.

For each simulation, we set a conductivity profile by defining the conductivity of the medium and adding a single circular anomaly parameterized by its position, radius, and internal electrical conductivity. The voltage measurements are obtained after injecting electrical currents into the domain through electrodes attached to the boundary of the domain. The simulation uses the Finite Element Method to compute the voltage measurements for each conductivity profile.

To download the dataset, head to this link.

Structure

In the dataset folder, you find a setup.json file containing the metadata for the dataset. This metadata contains the description of the dataset and the voltage measurement setup, such as the number of samples, the number of electrodes, the description of the injected current, and some numerical model parameters. For this dataset, it contains the following information:

{'nmb_samples': 1000,
 'nmb_electrodes': 16,
 'current_pattern': 'trigonometric',
 'current_amplitude': 3.0,
 'domain_radius': 10.0,
 'mesh_elements': 17887,
 'mesh_nodes': 9109}

The inputs and outputs for each simulation are stored inside a folder named with the respective simulation ID. In each folder, you find the input data stored in an anomaly.json file and the output data stored in a voltages.npy file.

Each anomaly input file contains the information that describes the conductivity profile parameterization, i.e., the centre position of the circular anomaly, its radius and the conductivity inside and outside the anomaly. One example of such input is as follows:

{'anomaly_radius': 1.3314534425735474,
 'x-center': 3.4796831607818604,
 'y-center': 3.9940834045410156,
 'conductivity_inside': 0.061698172241449356,
 'conductivity_outside': 0.0011272493284195662}

The voltage output file contains the respective voltage measurements acquired at the surface electrodes.

A scheme of the file structure is shown below:

├── sim_0000
│   ├── anomaly.json
│   ├── voltages.npy
├── ...
├── sim_0999
│   ├── anomaly.json
│   ├── voltages.npy
└── setup.json

Along with the dataset, you can also find a helper function to load all data and a Jupyter notebook with instructions on how to load the simulation data.

Ideas on how to use the dataset

This dataset can be used for different purposes, such as:

  • Empirically test algorithms to obtain images of the subject’s inside from the voltage measurements;
  • train machine learning models to learn the voltage measurements from the anomaly, or the anomaly from the voltages measurements.

For more information on how this is useful, take a look at our paper on Automatic differentiation as an effective tool in Electrical Impedance Tomography.

⚡️ Have fun! ⚡️

If you have any doubts regarding the dataset, don’t hesitate to reach out at datasets@inductiva.ai!

Also, in case you’re using this data on your projects, we would love to know how!

Besides this, don’t forget to check out the other cool datasets!

Recent posts from our blog

Sofia Guerreiro

Sofia Guerreiro

Cristiana Carpinteiro

Cristiana Carpinteiro

In this series of blog posts we will explore a specific case of the use of AI in the pharmaceutical industry - using Graph Neural Networks for predicting binding affinity. But for now, let’s start by understanding the problem of drug discovery and some fundamental concepts like binding affinity.

Ivan Pombo

Ivan Pombo

Luís Sarmento

Luís Sarmento

Coastal engineering projects protect the coast from erosion, flooding and other events. Due to their cost, the design phase of these projects is heavily based on computational simulations. In this blog post, we enlighten how Inductiva API allowed researchers to scale their coastal engineering simulations.