Since the virus in question attacks our respiratory system and hinders breathing ability, doctors have concluded that an efficient way to predict the prognosis is to examine the X-ray of patients.
We have employed Deep Learning in our application which uses the VGG16 which is a deep convolutional network for object recognition.
We have used the VGG16 as the base model and on top a custom made model to suit the requirements of the application.
DATASET USED - Dataset of Covid patients
Accuracy obtained - 90%
Specificity - 80% i.e. People who dont have covid and are diagnosed as negative
Sensitivity - 100% i.e. People who have covid and are diagnosed as positive which shows that there is no scope of error in the prognosis
What it does
- Use of deep learning in the app to predict whether the user is prone to coronavirus.
- Real time location updates of coronavirus infections all around the globe with current news on covid-19 of that location.
How I built it
The deep learning model uses VGG-16 architecture which is a deep convolutional network for object recognition developed and trained by Oxford's renowned Visual Geometry Group (VGG), which achieved very good performance on the ImageNet dataset. It is used as a base model while a top model is used to refine the model based on the requirements of this application. The weights used are pre-trained of the imagenet dataset and the architecture of the entire model is shown as below: -
Challenges I ran into
The dataset is scarce due to the novelty of the virus.
Accomplishments that I'm proud of
- Avoided queues needed to see an x-ray specialist in a hospital.
- Avoided getting infected in the hospital which is a hot zone for getting the virus
- Avoided putting load on already exhausted doctors by going to any local radiologist and uploading a scan image on our app by sitting at home
- Spread awareness by showing a real time location update on the coronavirus such that you can secure your location
What I learned
- Deep knowledge about covid19
- Practice with deep learning networks
- API interfacing (REST API)
What's next for COVID19 PREDICTOR
With time as the number of COVID19 cases increase, dataset shall increase and the accuracy of the deep learning network will improve naturally.
- Python 3.7.6
- Tensorflow 2.1.0
- Keras 2.3.1
- Flask 1.1.1
Accuracy and loss plot
The application is developed in flutter. It can be run simply by downloading app-release.apk and installing in your mobile phone. It accesses the API(link above) to use the trained model to predict COVID-19. A live tracker is also included in the application to keep track of status of COVID-19 cases around the world.
- cupertino_icons: ^0.1.3
- image_picker: ^0.6.3+4
- webview_flutter: ^0.3.14+1
- dio: ^2.1.16
- http: ^0.12.0+4
- flutter_spinkit: ^4.1.2