COVID-QU-Ex Segmentation

Introduction

• We propose an end-to-end realtime system to detect, localize, and quantify COVID-19 infection from X-ray images.

Proposed Architecture

Presentation Slide

• AIO 2023 End Course Project Slide

Experimental Results

Task	Backbone	Accuracy	IoU	DSC
Lung Segmentation	MobileNet v3	98.09	92.05	95.77
Infection Segmentation	MobileNet v3	97.77	80.17	85.65

CPU running inference: Intel(R) Xeon(R) CPU @ 2.20GHz
Inference time on average per image: 0.02 s
Achieve realtime segmentation with 50 FPS
Fully code for training and reimplementing experimental results: Kaggle Notebook

Installation

pip install -r requirements.txt

Streamlit App

App Link

Data Preparation

COVID-QU-Ex Dataset: Kaggle

Organize the dataset as follows:

|- datasets
   |- Infection Segmentation Data
   |  |- Test
   |  |   |- COVID-19
   |  |   |- Non-COVID
   |  |   |- Normal
   |  |- Train
   |  |   |- COVID-19
   |  |   |- Non-COVID
   |  |   |- Normal
   |  |- Val
   |  |   |- COVID-19
   |  |   |- Non-COVID
   |  |   |- Normal