M-VIC19: McMarvin Vision Imaging for COVID19

M-VIC19 is an AI Diagnosis platform is to help hospitals screen suspects and automatically locate the infected areas inside the lungs caused by the Novel Coronavirus using chest radiographs.

Comment 1

Inspiration

The total number of Coronavirus cases is 2,661,506 worldwide (Source: World o Meters). The cases are increasing day by day and the curve is not ready to flatten, that’s really sad!! Right now the virus is in the community-transmission stage and rapid testing is the only option to battle with the virus. McMarvin took this opportunity as a challenge and built AI Solution to provide a tool to our doctors. McMarvin is a DeepTech startup in medical artificial intelligence using AI technologies to develop tools for better patient care, quality control, health management, and scientific research.

There is a current epidemic in the world due to the Novel Coronavirus and here there are limited testing kits for RT-PCR and Lab testing. There have been reports that kits are showing variations in their results and false positives are heavily increasing. Early detection using Chest CT can be an alternative to detect the COVID-19 suspects. For this reason, our team worked day and night to develop an application which can help radiologist and doctors by automatically detect and locate the infected areas inside the lungs using medical scan i.e. chest CT scans.

The inspirations are as below:

1. Limited kit-based testings due to limited resources

2. RT-PCR is not as much as accurate in many countries (recently in India)

3. RT-PCR test can’t exactly locate the infections inside the lungs

AI-based medical imaging screening assessment is seen as one of the promising techniques that might lift some of the heavyweights of the doctors’ shoulders.

What it does

Our COVID-19 AI diagnosis platform is a fully secured cloud based application to detect COVID-19 patients using chest X-ray and CT Scans. Our solution has a centralized Database (like a mini-EHR) for Corona suspects and patients. Each and every record will be saved in the database (hospital wise).

Following are the features of our product:

  1. Artificial Intelligence to screen suspects using CT Scans and Chest X-Rays.

  2. AI-based detection and segmentation & localization of infected areas inside the lungs in chest CT.

  3. Smart Analytics Dashboard (Hospital Wise) to view all the updated screening details.

  4. Centralized database (only for COVID-19 suspects) to keep the record of suspects and track their progress after every time they get screened.

  5. PDF Reports, DICOM Supports, Guidelines, Documentation, Customer Support, etc.

  6. Fully secured platform (Both On-Premise and Cloud) with the privacy policy under healthcare data guidelines.

  7. Get Report within Seconds

Our main objective is to provide a research-oriented tool to alleviate the pressure from doctors and assist them using AI-enabled smart analytics platform so they can “SAVE TIME” and “SAVE LIVES” in the critical stages (Stage-3 or 4).

Followings are the benefits:

1. Real-world data on risks and benefits: The use of routinely collected data from suspect/patient allows assessment of the benefits and risks of different medical treatments, as well as the relative effectiveness of medicines in the real world.

2. Studies can be carried out quickly: Studies based on real-world data (RWD) are faster to conduct than randomized controlled trials (RCTs). The Novel Coronavirus infected patients’ data will help in the research and upcoming such outbreak in the future.

3. Speed and Time: One of the major advantages of the AI-system is speed. More conventional methods can take longer to process due to the increase in demand. However, with the AI application, radiologists can identify and prioritize the suspects.

How we built it

Our solution is built using the following major technologies:

1. Deep Learning and Computer Vision

2. Cloud Services (Azure in this case)

3. Microservices (Flask in this case)

4. DESKTOP GUIs like Tkinter

5. Docker and Kubernetes

6. JavaScript for the frontend features

7. DICOM APIs

I will be breaking the complete solution into the following steps:

1. Data Preparation: We collected more than 2000 medical scans i.e. chest CT and X-rays of 500+ COVID-19 suspects around the European countries and from open source radiology data platform. We then performed validation and labeling of CT findings with the help of advisors and domain experts who are doctors with 20+ experience. You can get more information in team section on our site. After carefully data-preprocessing and labeling, we moved to model preparation.

2. Model Development: We built several algorithms for testing our model. We started with CNN for classifier and checked the score in different metrics because creating a COVID-19 classifier is not an easy task because of variations that can cause bias while giving the results. We then used U-net for segmentation and got a very impressive accuracy and got a good IoU metrics score. For the detection of COVID-19 suspects, we have used a CNN architecture and for segmentation we have used U-net architecture. We have achieved 94% accuracy on training dataset and 89.4% on test data. For false positive and other metrics, please go through our files.

3. Deployment: After training the model and validating with our doctors, we prepared our solutions in two different formats i.e. cloud-based solution and on-premise solution. We are using EC-2 instance on AWS for our cloud-based solution.

Our platform will only help and not replace the healthcare professionals so they can make quick decisions in critical situations.

Challenges we ran into

There are always a few challenges when you innovate something new. The biggest challenge is “The Novel Coronavirus” itself. One of the challenge is “Validated data” from different demographics and CT machines. Due to the lockdown in the country, we are not able to meet and discuss it with several other radiologists. We are working virtually to build innovative solutions but as of now, we are having very limited resources.

Accomplishments that we're proud of

We are in regular touch with the State Government (Telangana, Hyderabad Government). Our team presented the project to the Health Minister Office and helping them in stage-3 and 4.

Following accomplishments we are proud of:

1. 1 Patent (IP) filled

2. 2 research paper

3. Partnership with several startups

4. In touch with several doctors who are working with COVID-19 patients. Also discussing with Research Institutes for R&D

What we learned

Learning is a continuous process. Our team learnt "the art of working in lockdown". We worked virtually to develop this application to help our government and people. The other learning part was to take our proof of concept to the local administration for trails. All these “Government Procedures” like writing Research Proposal, Meeting with the Officials, etc was for the first time and we learned several protocols to work with the government.

What's next for M-VIC19: McMarvin Vision Imaging for COVID19

Our research is still going on and our solution is now endorsed by the Health Ministry of Telangana . We have presented our project to the government of Telangana for a clinical trail . So the next thing is that we are looking for trail with hospitals and research Institutes. On the solution side, we are adding more labeled data under the supervision of Doctors who are working with COVID-19 patients in India. Features like Bio-metric verification, Trigger mechanism to send notification to patients and command room , etc are under consideration. There is always scope of improvement and AI is the technology which learns on top of data. Overall, we are dedicated to take this solution into real world production for our doctors or CT and X-rays manufacturers so they can use it to fight with the deadly virus.

+ 7 more
Share this project:

Updates