Bangkit YOG-1 C Final Project

Classifying COVID-19 and Non-COVID-19 Lungs From CT Scan Using CNN

0. Dataset

What Dataset Did We Choose?

We choose COVID-19 Lung CT Scans dataset for our final project. This dataset is a collection of COVID-19 related papers from medRxiv, NEJM, JAMA, Lancet, etc. Total images in this collection is 746 CT scan images. It consists of 349 COVID positive images and 397 COVID negative images.

1. Reason

Why Did We Choose the Dataset?

1. We choose COVID-19 CT scans dataset due to the current situation, COVID-19 pandemic outbreak.
2. COVID-19 Lung CT Scans is easy to understand and has some public kernels. Public kernels are useful to gain insight and compare our work with the others.

2. Result

Baseline CNN Implementation

3. Reason

Why Did We Choose the Specific Improvement?

4. Result

Improvement

5. Documentation

ARCHIVE CONTENTS

CT_NonCOVID.zip: Contains non covid lungs ct scan images. CT_COVID.zip: Contains covid lungs ct scan images.

HARDWARE: (The following specs were used to create the original solution)

We are using google colab’s facility with GPU Hardware accelerator.

SOFTWARE:

Python 3.6.9 EfficientNet 1.1.0 Tensorflow 2.2.0 Numpy 1.18.5 Pandas 1.0.4 Sklearn 0.22.2.post1 OpenCV 4.1.2 Matplotlib 3.2.1

DATA SETUP

!git clone https://github.com/UCSD-AI4H/COVID-CT.git !unzip "COVID-CT/Images-processed/CT_NonCOVID.zip" -d "dataset/" !unzip "COVID-CT/Images-processed/CT_COVID.zip" -d "dataset/"

DATA PROCESSING

Normalization Make the pixel range between 0 and 1. Reshape 224x224

MODEL BUILD:

Call Efficientnetb0BaseModel(config_json) class.

6. Propose/Ideate

Local (Indonesia) implementation of the project you've done