Covid_19_detection_from_Xray

Overview

COVID-19 is an infectious disease which has led to a dramatic loss of human life worldwide, its early detection is critical to control cases’ spreading and mortality. The actual leader diagnosis test is the Reverse transcription Polymerase chain reaction (RT-PCR)which is based on Nasopharyngeal swabs (NPSs).

This paper reports our experience with detection of COVID-19 using chest X-ray images(CXR). We here try to implement Un-Supervised, Supervised and Deep learning methods for binary image(COVID or NON-COVID) classification on CXR images. Implementation of PCA followed by application of K-means for the unsupervised techniques. In supervised learning we explored models such as Random Forest classifier, Decision Tree classifier, GaussianBayes. For the Deep Learning approach we developed a custom pipeline which involves usage of segmentation, SOTA(State of the art) CNN models and Multi-modular Approach.

We here propose a Multimodular Pipeline which uses feature maps of 3 Models with same scratch neural net architecture based on VGG and trained on Raw CXR Images, Segmented Masked CXR Images and Segmented Inverse Masked Images and finally does classification through dense classifier layers.

Best Accuracy Acheived: 96%

Dataset Links

X-ray images
Lung segmentation dataset

Built using:

• Scikit Learn: used ML Library
• PyTorch: used ML Libraries
• Tensorflow and Keras: used ML Libraries
• Streamlit: Javscript framework used
• Pandas: Python data manipulation libraries
• Seaborn: Data visualisation library
• Matplotlib: Plots and Data visualization
• OpenCV: Image Processing

Pipeline:

1. combined_model.ipynb

This is the main file with all the preprocessing, EDA, various Machine learning and Deep Learning Models.

• Installing libraries and dependency

• Importing the dataset

• Exploratory Data Analysis and Visualisation

• Data Preprocessing

  • Normalization
  • Resizing, Random Rotation, Random Flips
  • conversion to Tensor
  • Flattening of images

• Workflow

• 

• Unsupervised Learning Techniques

  • PCA
  • K-means clustering
  • Agglomerative clustering

• Supervised Learning Techniques

  • Decision Tree Classifier
  • Random Forest Classifier
  • Gaussian Bayes

• Deep Learning Techniques

  • ResNet-18
  • ResNet-50
  • MobileNet-v3
  • VGG-16
  • VGG-19

• Image Segmentation and U-Net

  • Workflow
  • 
  • The model returns the following 4 things :-
    • 1. Mask
    • 2. inverse (complimentary) Mask
    • 3. Masked image (Mask multiplied with original CXR image)
    • 4. Inverse Masked image (Inverse Mask multiplied with original CXR image)

• Proposed Combined Multi-Modular Approach

  • Workflow
  • 

• Comparitive Analysis

  • Various Evaluation metrices used
  • Plots for comparing performance
  • Performance Table

How to run:

• Run the cells according to above mentioned pipeline

Deployment

Team Members

Name	Roll No.
Mukul Shingwani	B20AI023
Saurabh Modi	B20EE035
Mitarth Arora	B20EE096