Genetic optimization of deep CNN based transfer learning for COVID-19 identification from radiography chest X-Rays

---

A. Environment

• Google Colaboratory: This project use Google Colaboratory for compute purpose.
• RAM: 12 GB, Disk Space: 80 GB
• Compute Engine: GPU
• Language: Python

B. Dataset

• Covid-19 Radiography Dataset
  • M. E. Chowdhury, T. Rahman, A. Khandakar, R. Mazhar, M. A. Kadir, Z. B. Mahbub, K. R. Islam, M. S. Khan, A. Iqbal, N. Al Emadi, et al., Can ai help in screening viral and covid-19 pneumonia?, IEEE Access 8 (2020) 132665–132676.
  • T. Rahman, A. Khandakar, Y. Qiblawey, A. Tahir, S. Kiranyaz, S. B. A.395 Kashem, M. T. Islam, S. Al Maadeed, S. M. Zughaier, M. S. Khan, et al., Exploring the effect of image enhancement techniques on covid-19 detection using chest x-ray images, Computers in biology and medicine 132 (2021) 104319.
• Sample Data: Following is the sample of dataset with class label.
  

C. Experimental Data

• Pickle Data: All experimental data of this project work are stored in Pickle Data.
• Data: All training histories are stored in this link.
• Models: All best models data found during optimization and exploitation are stored here.
• TL-Models: All best models data found during training of transfer learning models are stored here.

D. Codes

• **Model: **
  • VGG16: VGG16 model primirily used for this project. Following figure show the structure of this CNN model.
    
• Transfer_Learning_with_Optimization_of_Covid_19_Image_Classification.ipynb/py: Codes on optimization process on transfer learning for covid-19 identification.
  • collect_Data()
    • This function used to collect data based on the classes (folders, each folder represent a class in dataset folder) from given drive path.
  • Pickle
    • Pickle is used for store and retrive data from given drive path.
  • `train_test_split(…)
    • Used for data spliting for cross validation during training.
  • display_multiple_img(...)
    • Used for display some random data images with thier classes.
  • train(...), compile(...), fit(...), ... etc
    • Used for training the model and performance measures.
  • initial_individuals(model)
    • Used for initialization of unique individual for the initial population.
  • mutate(individual, model)
    • Used for mutation of individual with a mutation factor and provide unique gene.
  • crossover(individuals, model)
    • Used for crossover operation on a population (individuals) and provide unique individual in population using uniform crossover.
  • evolve(individuals, fitness, model)
    • Used for evolution of individuals based on fitness for next generation. In this work evolution is done upto 30 generations.
  • Performance Measures
    • After optimization of models performance measures for best 3 models are conduct based on accuracies, confusion matrix, etc.
• Exploitation_on_Transfer Learning_with_Optimization_Covid_19_Image_Classification.ipynb/py: Codes on further exploitation process on optimized transfer learning models found for covid-19 identification.
  • This python file includes codes for training of optimized transfer learning models found for further 100 epochs.
• Transfer Learning_without_Optimization_Covid_19_Image_Classification.ipynb/py: Codes on straight forward transfer learning training process for covid-19 identification.
  • This python file includes codes for training of straight forward transfer learning models without optimization for 300 epochs and saved best models found.
  • Model construction is done randomly based on fair coin toss.

E. Results and Performance

• Results: Performance of the Transfer Learning Models with and without Genetic Optimization.
  

F. Dependencies

• The dataset importation from Google Drive and saving model data into Google Drive required drive permission.