Lungs X-ray Chest Classification(COVID-19 or Other pneumonia)

This project uses a convolutional neural network to build a classifier to determine whether a patient's pneumonia is due to neocoronavirus or other types of pneumonia. The data set is described below.

Structure

.
|-- main.py                     # train model
|-- utils.py                    # define modules
|-- data
|   |-- public
|   |   |-- train
|   |   |   |-- positive
|   |   |   |-- negative
|   |   |-- val
|   |   |   |-- positive
|   |   |   |-- negative
|-- model
|   |-- resnet50_public-40.pth  # saved model

Dataset

• Kaggle Dataset

  The dataset is organized into 3 folders (train, test, val) and contains subfolders for each image category (Pneumonia/Normal). There are 5,863 X-Ray images (JPEG) and 2 categories (Pneumonia/Normal). Chest X-ray images (anterior-posterior) were selected from retrospective cohorts of pediatric patients of one to five years old from Guangzhou Women and Children’s Medical Center, Guangzhou. All chest X-ray imaging was performed as part of patients’ routine clinical care.

• Github Dataset

  This project extract images from publications.

Details

Datasets description

	COVID-19	Others
train	48	48
val	21	21

Lab environment

• Hygon C86 7185 CPU
• Pre-Wukong DCU
• 128GB RAM

Result

	ACC	AUC	SEN	F1
train	0.9792	0.9970	0.9792	0.9792
val	0.9762	1.0	0.9524	0.9756

ROC