neuracovid-py has a AI model which will give inferences by using tensorflow.

This file covers all requirments and dependencies from scratch for application Neuracovid-py

follow the below steps to run model

Step 1

Create pubsub topic and subscription

Go to Google Cloud Platform
Click on Topics, create a new topic
create a subscription for the topic
Click on Subscriptions, select topic and create subscription
In Google Cloud Platform select storage click on browser and create bucket to store artifacts eg. model , credential file etc
create another bucket to store logs.

Create a vm instance and run commands on SSH terminal window from root directory

recommendation is to have 2vcpu configuration to process max 3 messages at a time and 8vcpu configuration to process more than 25 messages

Step 2

sudo apt-get update

apt-get update downloads the package lists and "updates" them to get information on the newest versions of packages and their dependencies

sudo apt-get install python3-venv -y

python3 -m venv venv_name

source venv_name/bin/activate

The python3-venv module allows us to create “virtual environments” and source will activate "virtual environment"

pip install --upgrade pip

This will upgrade pip to latest available version

Step 3

sudo apt install git-all -y

Install Git to clone repository

Step 4

cd /opt ; sudo mkdir inference_module ; cd inference_module

sudo mkdir src ; sudo mkdir bin ; cd src

create path to clone repository

Step 5

git clone https://github.com/neurahealth/neuracovid-py.git

It will clone the Git Hub repository dedicated for neuracovid Application

cd neuracovid-py

pip install -r requirements.txt

below dependences will get install

Tensorflow==1.14.0
numpy
opencv-python
scikit-Learn
matplotlib
pydicom
pandas
firebase_admin
google-cloud-pubsub
python-dotenv

Application flow diagram

NeuraCovid Youtube video link

https://www.youtube.com/watch?v=9erwBwCPPzU&feature=emb_title

Step 6

git clone https://github.com/lindawangg/COVID-Net.git

COVID-Net.git is a open source repository to fight againest the pandemic COVID-19, above command will clone it in neuracovid- py directory

git clone https://github.com/haydengunraj/COVIDNet-CT.git

Step 7

save models in google drive

[Model-CXR-Large] go to https://bit.ly/CovidNet-CXR-Large

[Model-SEV-GEO] go to https://bit.ly/COVIDNet-SEV-GEO

[Model-SEV-OPC] go to https://bit.ly/COVIDNet-SEV-OPC

[COVIDNet-CT-A] go to https://bit.ly/2BAPyvM

google drive to google cloud storage.ipynb

This file will help to save model from google drive to storage bucket,run commands in colab to save model in storage bucket

Step 8

cd COVID-Net ; sudo mkdir models ; cd models

create path to store model

gsutil cp gs://bucket_name/COVIDNet-CXR-Large .

gsutil cp gs://bucket_name/COVIDNet-SEV-GEO .

gsutil cp gs://bucket_name/COVIDNet-SEV-OPC .

This will copy model files from storage bucket (these files should be inside models/COVIDNet-CXR-Large/).

cd ..

gsutil cp gs://bucket_name/COVIDNet-CT-A COVIDNet-CT-A/models

cd ..

come back to root directory to run script inference.py

Copy crerdential file and environment file to working directory(COVID-Net)

save credential .json file and .env in storage bucket

gsutil cp gs://bucket_name/credential.json .
gsutil cp gs://bucket_name/.env .

give grants to /opt path” to successfully run python inference.py command

run

for inference only > python inference.py

for inference and heatmap > inference_heatmap.py

for CT innference > COVIDNet-CT-A/covidnet_ct.py

open another shell

python inference_severity.py

How inference.py | inference_heatmap.py and inference_severity.py work ?

inference.py | inference_heatmap.py script has a updated AI model
This script will get trigger when a new message comes in pubsub topic, message will sent to topic upon the image upload by the user, findings will be store in firebase-database and heatmaps to storage for every user uniquely

Inferences will be:

prediction, sensitivity and severity
Normal, Pneumonia, COVID-19 and sensitivity will stored in pecentage

Deployment

application can be deployed to GCP kubernetes engine

Done Cheers!