Performance of various Machine Learning Algorithms on Electrical Impedance Tomography Images

Overview

This report provides an overview of current practice of Electrical Impedance Tomography (EIT), its imaging and use-cases. Electrical Impedance Tomography is a non-invasive type of medical imaging. These advances are improving our capacity to treat and even prevent cancers. The full implications of the subject remain to be explored. Examples of research techniques used in this project are detailed.

Dependencies

1. Python 3.x
2. Numpy
3. Scipy
4. Pandas
5. Matplotlib
6. Sci-kit learn
7. OpenCV Python

Important Files and Folders

eit
│   README.md   
│
└───assets
│      datasets - contains datasets in csv
│      eit_images - generated images
│   
└───classification
│      *.ipynb - Classification ML algorithms
│      results.ipynb - Final results and graphs
│   
└───docs
│      documentation and reports
│
└───main
       eit_analysis.py
       eit_classify.py
       eit_dataset.py 

Usage

1. generate_image.py

   • Generates 1000 images
   • Linspace and Meshgrid are numpy methods

2. read_img.py

   • Reads an image into code
   • matrix contains three dimensional array of image
   • img contains three dimensional array of image - image import
   • grayscale contains two dimensional array of image
   • x contains x dimension of image
   • y contains y dimension of image

3. eit.py

   • Plots a contour graphs
   • Adds list of colors and be saved as an image

4. eit_dataset.py

   • Generates dataset without labels - creates file eit.csv
   • intensity_range_strings contains ranges of intensities
   • classify_dict contains dataset in the form of dictionary
   • df contains final file to be converted to csv

5. eit_analysis.py

   • Assigns targets 1 or 0 and created another dataset - creates file eit_data.csv
   • target contains target array 0s and 1s

6. eit_classify.py

   • Generate classification plots - generates eit_contour_plot.csv
   • autolabel - function labels bar graphs

7. <*>.ipynb - All classification ML algorithms - '<*>' means all files

Results

No	Algorithms	Accuracy (%)
1	K Nearest Neighbours	93.6%
2	Decision Tree Classification	98.8%
3	Kernel Support Vector Machines	94%
4	Logistic Regression	88.4%
5	Naive Bayes	92.4%
6	Random Forest Classification	99.2%
7	Support Vector Machines	88%

License

Copyright(c) 2018, Faststream Technologies

Authors:

• Sudhanva Narayana
• Shreyas S

Credits

Vinod Agrawal

CTO, Faststream Technologies