CME INDIA Presentation.

See this scientific paper with case studies and learning points.

Implementation of ‘COVID-Net’, an Artificial Intelligence (AI) platform in detection of COVID-19 through Chest X-Rays – Study at 2 centers in Bokaro

Anurag Agrawal1*, Sumit Kumar2, Sharad Kumar Singh3, Priyanka Jain4, Santosh Chaubey5, Gaurav Vishal6, Bibhuti Bhushan Karunamay7, Satish Kumar8, Arvind Kumar Singh9

  • 1 HOD, Department of Radiology and Imaging Sciences, Wellmark Hospital, Bokaro Steel City, Jharkhand, India.
  • 2 Solution Architect for Artificial Intelligence & High-Performance Computing, NVIDIA, MS (CS&AI), GeorgiaTech. University, Atlanta, GA, USA.
  • 3 DGM I/C (C & IT), Bokaro General Hospital (BGH), SAIL/BSL, Bokaro Steel City, Jharkhand, India
  • 4 HOD, Department of Radiodiagnosis and Imaging, Bokaro General Hospital,
  • 5 Consultant, Department of Medicine, Bokaro General Hospital
  • 6 I/C Molecular Biology Lab, Bokaro General Hospital
  • 7 I/C COVID-19 Unit, Bokaro General Hospital
  • 8 I/C Cardiology Unit, Bokaro General Hospital
  • 9 HOD & Director I/C, Medical & Health Services, Bokaro General Hospital, Bokaro Steel City, Jharkhand, India

*Corresponding Author: Dr. Anurag Agrawal, Wellmark Hospital, 15E/3, Western Avenue, B. S. City, Jharkhand, India, PIN – 827002. Email:


Coronavirus Disease 2019 (COVID-19) is an infectious disease caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). The disease was first identified in December 2019 in Wuhan, China and has since spread globally, resulting in the worst pandemic of recent times. In an attempt to contribute to the cause of relief, we attempt to use Machine Deep Learning and Convolutional Neural Network based Artificial Intelligence approaches to predict and understand the infection.

The Problem:

COVID-19 is known to attack the epithelial cells that line our respiratory tract.

A critical step in the fight against COVID-19 is effective screening of infected patients, with one of the key screening approaches being radiology examination using chest radiography. It was found in early studies that patients present abnormalities in chest radiography images that are characteristic of those infected with COVID-19. Hence, HRCT Thorax (the Gold Standard) and X-rays Chest are used to analyze the health of a patient’s lungs.

However, in the current scenario hospitals are overwhelmed with the increasing number of COVID-19 cases every single day, there is scarcity of CT scan machine availability with its associated high cost of imaging to the patient and increased radiation dose compared to X- Ray Chest, scarcity of radiology experts and the time taken for diagnosis for each patient sums up to an acute delay issue.

[Cost of a Chest X-Ray is somewhere around INR 250 while a Chest CT costs INR 4-6 thousand for each study. Similarly, a chest x-ray delivers on average 0.1 mSv, while a Chest CT delivers 7-10 mSv of radiation dose.]

Our Attempt:

Motivated by this and inspired by the open source efforts of the research community, we attempt to create an automatic analysis system to resolve the situation where we used various transfer learning techniques and introduce COVID-Net, a deep convolutional neural network design tailored for the detection of COVID-19 cases from chest X-Ray (CXR) images.1 It is open source, will be made available to the public, and will aid clinicians in improved screening.

Our team of doctors from Bokaro in the fields of Radiodiagnosis, Medicine and C & IT in association with AI expert Mr. Sumit Kumar, MS (CS & AI, GeorgiaTech, Atlanta, USA) have been exploring and training a system that could predict COVID-19 infection from digital Chest X-Rays transferred through our PACS (Picture archiving and communication system) of the hospital to the AI computer.

We are glad to announce that we have come with a novel algorithm presently in its Beta version (late testing and validation phase). The initial version of it has been already been studied extensively in US & Canada and numerous papers have been published or are into advance stages of publication.

We believe this AI-enabled detection and prediction system shall prove to be a boon in logistic starved areas like Jharkhand.

How it Works:

This AI system uses proven Darwin-AI corp.’s computer vision and image processing algorithm to segment the digital chest x-ray images into useful (the lungs) and non-useful (the mediastinum and peripheral regions) areas. The segmented lung images are then processed using Machine learning and Convolutional neural network algorithms which have been trained using available large image datasets of known COVID-19 positive patients and

COVID-19 negative patients (a total of more than 13975 chest x-rays at present1) to recognize the COVID-19 disease pattern on Chest X-Rays.

We are also planning clinical trials at the Wellmark Hospital as well as BGH for gaining our confidence over this platform and audit COVID-Net in a responsible and transparent manner to validate that it is making decisions based on relevant information from the CXR images.


  • Rapid triaging: CXR imaging enables rapid triaging of patients suspected of COVID-19 and can be done in parallel of viral testing (which takes time) to help relieve the high volumes of patients especially in areas most affected where they have ran out of capacity, or even as standalone when viral testing isn’t an option (low supplies). Furthermore, AI enabled CXR image analysis can be quite effective for triaging in geographic areas where patients are instructed to stay home until the onset of advanced symptoms, since abnormalities are often seen at time of presentation when patients suspected of COVID-19 arrive at the clinics/hospital and where the case load is high.
  • Availability and Accessibility: CXR imaging is readily available and accessible in many small and large hospitals, clinics and imaging centers as it is considered standard equipment in most healthcare systems. In particular, CXR imaging is much more readily available than CT imaging, especially in developing countries where CT scanners are cost prohibitive due to high equipment and maintenance costs.
  • Reduced cost: reduced cost of investigation to the patient especially in cases of repeat imaging during intensive care / follow-up.
  • Portability: The presence of portable digital CXR systems means that imaging can be performed within an isolation room, thus significantly reducing the risk of COVID-19 transmission during transport of patients to fixed systems such as CT scanners as well as within the rooms housing the fixed X-Ray imaging systems.

The Road Ahead:

Implementation of Risk stratification of the COVID-19 patients based on AI evaluation of Chest X-Rays by geographic and opacity extent scoring of SARS-CoV-2 lung disease severity.


We would like to emphasize that we are not proposing the use of the Al model as an alternative to the conventional/ prevailing guideline (national or institutional) supported diagnostic tests for COVID-19 and the decision of the treating clinician should be given the precedence over the Al generated reports.

These are the results for the COVID-Net models1.

COVID-Net (100 COVID-19 test x-rays)

Sensitivity (%)


Positive Predictive Value (%)


Sample X-rays that have been used to train the AI system for analysis and detection of COVID- 19 on Chest X-ray.

Covid X-Ray Scan

These samples have been collected from RSNA Pneumonia Detection challenge dataset2, ActualMed COVID-19 Chest X-Ray Dataset Initiative3, COVID-19 radiography database4, Figure 1 COVID-19 Chest X-Ray Dataset Initiative5 and COVID-19 Image Data Collection6.

**We acknowledge the teamwork all the staff specially the radiology technicians lead by Mr G. Sharma of the department of Radiodiagnosis and Imaging of Bokaro General Hospital, Bokaro in imaging of the COVID-19 suspect and proven positive patients.


  1. Wang L., Lin Z. Q., Wong A., COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images, , Electrical Engineering and Systems Science, 2020 MAY 11, arXiv:2003.09871v4 [eess.IV].
  2. of North America, R. S., RSNA pneumonia detection challenge. detection-challenge/data (2019).
  3. Chung, A. Actualmed COVID-19 chest x-ray data initiative. chestxray-dataset (2020)
  4. of North America, R. S., COVID-19 radiography database. radiography-database (2019).
  5. Chung, A. Figure 1 COVID-19 chest x-ray data initiative. dataset (2020).
  6. Cohen, J. P., Morrison, P. & Dao, L. COVID-19 image data collection. arXiv 2003.11597 (2020).

CME INDIA Case Studies

CASE 1: By Dr Satish Kumar.

Admitted, No h/o fever, no cough or SOB, SPO2 98% at admission, only complaint loose motions for one day and general weakness.

Routine chest x-Ray

Covid X-Ray Scan
COVIDNET prediction 99.8% done

Covid Net X-Ray Scan

Dr Satish Kumar, Bokaro: I thought it is some error, sent to a separate cabin till confirmation by TruNat

Covid Net X-Ray Scan

Alert/ No fever no Cough, only Ac GE, but Chest XRay by this system predicted 99.8% COVID positivity, Found later POSITIVE. Only thing, CXR showed HRCT Initially not done as there were no respiratory symptoms at all.

Covid Net X-Ray Scan

Later HRCT showed

Covid Xray Scan

CASE 2: By Dr Satish Kumar.

Relative of an anaesthesiologist colleague. She is in Nepal at a small place. All symptoms of COVID. They had sent a x-ray snap on WhatsApp not properly taken and skewed. Again it was repeated with a better obtained image. The results is the same… 💯% COVID

Covid Net X-Ray Scan

Case 3: By Dr N K Singh

I sent this CXR for analyses on this system. This is a case of diabetes, came to me with history of one day fever 6 days back, Patient SpO2 93%, did CXR.

Covid X-Ray Scan

Sent this CXR to COVID NET system.

Covid Net X-Ray Scan

Later HRCT showed:

Covid X-Ray Scan
Covid X-Ray Scan

rtPCR : Positive, Admitted.

CME INDIA Learning Points

  • With limited testing kits, despite all efforts by governments in different countries, it is just impossible for every patient with respiratory illness to be tested using conventional techniques like Chest CT SCANS & RT-PCR to detect the virus. These tests also have long turn-around time and are not available everywhere.
  • Data scientists and doctors have been brainstorming to make simple and readily available tests like chest x-rays to transform into dependable tool for detecting with surety the possible COVID-19 infections. This may also help quarantine high risk patients while test results for RT-PCR are awaited.
  • To achieve this, researchers from various universities in US, Canada and other countries have been involved in developing Artificial Intelligence methods using deep convolutional neural network to generate highly accurate diagnosis and predictions. One of such an artificial intelligence (AI) algorithm is called COVID-Net for detecting COVID-19 by analysing just plain chest x-rays. This highly complex software set is also open-sourced, which means they are available free of cost for uses, by doctors and scientists.
  • This COVID-Net algorithm is trained with the help of an another Open-Source AI platform called DarwinAI.
  • Till now, COVID-Net has extensively been studied using more than 7,500 X-rays. The reported accuracy of prediction of COVID infection is more than 95%.

CME INDIA Tail Piece:

  • Recently, Dr. Anurag Agrawal, HOD of Radiodiagnosis & Imaging at the coming up health facility at Bokaro, Wellmark Hospital has installed this system with the help of Mr. Sumit. It is undergoing clinical trials in association with clinicians at Bokaro General Hospital (BGH) which is the pivotal COVID-19 hospital in the city. They are finding these results very exciting and dependable.
  • The system is in constant improvement with more data sets and uses and the latest version of this may even express the disease severity and possible outcomes. They continue to open source this model to the community in hopes of developing a robust tool to assist researchers and health care professionals in combating the pandemic.

Discover CME INDIA

Discover CME INDIA