COVID-2019, which first appeared in Wuhan city of China in December 2019, spread rapidly around the world and became a pandemic. Its very common for doctors to frequently use X-rays and CT scans to diagnose pneumonia, lung inflammation, abscesses, and/or enlarged lymph nodes. Since COVID-19 attacks the epithelial cells that line our respiratory tract, X-rays can be used to analyze the health of a patient’s lungs. It could be possible to use X-rays to test for COVID-19 without the dedicated test kits. In this repository, a new model for detecting COVID-19 using raw chest X-ray images is presented. The model developed presents accurate diagnostics for binary classification and morever achieved a good accuracy and very less loss rate.
The most common test technique currently used for COVID-19 diagnosis is a real-time reverse transcription-polymerase chain reaction (RT-PCR). Chest radiological imaging such as computed tomography (CT) and X-ray have vital roles in early diagnosis and treatment of this disease. Due to the low RT-PCR sensitivity of 60%–70%, even if negative results are obtained, symptoms can be detected by examining radiological images of patients. It is stated that CT is a sensitive method to detect COVID-19 pneumonia, and can be considered as a screening tool with RT-PRC. CT findings are observed over a long interval after the onset of symptoms, and patients usually have a normal CT in the first 0–2 days. In a study on lung CT of patients who survived COVID-19 pneumonia, the most significant lung disease is observed ten days after the onset of symptoms.
Reference to this information can be found at - https://www.sciencedirect.com/science/article/pii/S0010482520301621?via%3Dihub
The image dataset used for creating this classifier can be collected at - https://github.com/BIMCV-CSUSP/BIMCV-COVID-19
Number of positive covid images are - 100
Number of negative covid images are - 100
A sample image from the dataset for covid-19 positve is shown below :
A sample image from the dataset for covid-19 negative is shown below :
├── yolo_app
│ ├── admin.py
│ ├── apps.py
│ ├── Data
│ │ ├── init.py
│ │ ├── Model_Weights
│ │ │ └── data_classes.txt
│ │ └── Source_Images
│ │ ├── Test_Image_Detection_Results
│ │ │ ├── CCryct.2020200034.fig5-day7.jpeg
│ │ │ ├── Detection_Results.csv
│ │ │ ├── IM-0251-0001.jpeg
│ │ │ ├── IM-0253-0001.jpeg
│ │ │ ├── IM-0261-0001.jpeg
│ │ └── Test_Images
│ ├── Inference
│ │ ├── Detector.py
│ ├── init.py
│ ├── migrations
│ │ ├── 0001_initial.py
│ │ ├── init.py
│ ├── models.py
│ ├── requirements.txt
│ ├── tests.py
│ ├── urls.py
│ └── views.py
Step 1: Positive Samples has been taken from BIMCV-COVID-19 repository, Whereas negative samples has been taken from kaggle. Correspondingly, there are two folders, one for positive and another for negative. I have used Visual Object Tagging Tool (VoTT) of microsoft to perform tagging. Check this link for installation and usage of this software - How to install and use VoTT.
Once you are finished with tagging step, you can export csv file from the VoTT software which is needed for the next step. See the link below for viewing the status when annotation has been completed by me using VoTT.
Images Annotation using VoTT
Step 2: Convert the annotation to the yolo format. To achieve this step use convert to yolo format
Step 3: Download the pre-trained dark-net weights and convert them to YOLO format.TO achieve this step, please use the below code Download and convert to yolo
Step 4: It is the time for the training of the detector. We got the images as well as its annotation from the step 2. We got the weights from the step 3 in yolo format. Now you can execute the below code to get trained the model with the customized images. Train the yolo model with the customized images
Step 5: Once the training is finished you will get the new weights in form of h5 file. For this model I have made the file which one can find in the below link
Trained keras H5 file for the customized Dataset
Please keep these weights inside the Folder Data>>Model_Weights.
One can skip first four steps because the purpose of those steps were to create weights for the step 5. If one is training with different images those steps are needed. Please install all the requirements first from requirements file.
Step 6: Django model has been made for this. only the logged in user will upload the image and result will be displayed to their login page. Results are also saved in the database as well as history link is also created for login user to see the result on their login page. Application is linked with phpmyadmin mysql database. See the below screenshot
MySQL database in phpMyadmin
Step 7: History is created for every logged in user and for achieving this step data is retrieved from database using django queries.