CASE STUDY

PneumoScan: PPE Safety Checkfor Medical Staff

sanitas team

Pneumoscan’s team implements PPE detection ML model as part of their AI tool for PPE monitoring and COVID-19 testing, to win AWS Marketplace Developer Challenge

Challenge

PneumoScan’s team established that PPE detection had to be part of their AI tool to protect medical staff and prevent the spread of COVID-19 in medical facilities. They were looking for a ready-to-use, highly accurate and secure ML model on AWS that they could integrate with their chest X-ray classification model and ID badge scanner in one web application.

Solution

Having researched various options of PPE detection models in AWS Marketplace, the team decided to opt for VITech Lab’s PPE Detector for Laboratory Safety. They implemented the model on Amazon SageMaker to make it work with live video streams captured from CCTV cameras, to detect individuals missing protective coats, glasses, gloves, and face masks.

Outcome

Since the machine learning model that powered PPE Detector for Laboratory Safety had been trained on an accurately curated dataset of 10K+ images, Pneumoscan was able to ensure high accuracy of PPE detection from the start. They completed the hackathon project in just two weeks and managed to become one of the winners.

EXTENDED USE CASE DESCRIPTION

Challenge

From the beginning of the SARS-CoV-2 outbreak, various organizations, including those in the technology sector pledged to invest their time and resources (and their best talent) to fight the pandemic. They sought to resolve such problems as:● Slow, inaccurate testing● Poor disease control & spread prevention● Lack of doctors and medical staff● Inefficient, cumbersome hospital operations● Patient care that exposes doctors to the virus
Amazon Web Services (AWS) was one among many companies that responded to the challenges of COVID-19. It provided its cloud infrastructure to bolster anti-COVID initiatives: mostly, vaccine research, but also the development of solutions helping governments and healthcare providers become more efficient in curbing the spread of the pandemic.
Tech communities around AWS were also contributing to the efforts. Among them, three engineers — Vi Ly, Sagar Bansal, and Mohit Gadkari — opted to participate in the AWS Marketplace Developer Challenge: ML Solutions hackathon, to build PneumoScan, an AI tool that would help doctors scan for COVID-19 faster and prevent disease spread through comprehensive PPE monitoring. Under the rules of the hackathon, they needed to utilize ML models from AWS Marketplace deployed on Amazon SageMaker to develop their project. All of the models had to be highly secure since medical data was to be processed.
The team came up with the solution design consisting of three elements:● ID Badge Scanner — To enable doctors to access the web application with patients’ confidential health information by scanning their ID.● PPE Detector — To check if doctors who are going to test/scan patients for the novel coronavirus follow PPE guidelines. All available in the web app, as well.● COVID-19 Chest X-Ray Scanner — To scale doctors’ expertise through AI and machine learning. X-ray images are uploaded to the app to be analyzed by ML.
Their major challenge was to find ML models that they could easily integrate with each other to deliver a comprehensive, ready-to-use solution accessible via a web application. One of those models was VITech Lab’s PPE Detector for Laboratory Safety. 

What VITech Lab Did


PPE Detector for Laboratory Safety was developed by the VITech Lab team as one of the range of Computer Vision solutions for PPE monitoring. The solutions are designed to detect and report any PPE compliance violations in real-time in a variety of industries; namely, in manufacturing, construction, and oil & gas.
PPE Detector for Laboratory Safety was built specifically to be implemented in research centers, laboratories, healthcare facilities, and other environments that require staff to wear face masks, protective coats, glasses, and gloves. The ML model that is used in the solution was trained on a curated dataset of 10K+ images of doctors, nurses, and laboratory workers wearing/not wearing these PPE items.
By design, the solution can be easily integrated with IP/CCTV cameras and monitoring systems. The camera footage is captured in real-time; then, images are resized and pushed to the ML model, where individuals and PPE items get classified and identified. If missing protective gear is detected, notifications are generated and sent to the monitoring system’s application.

ppe detection by vitech lab

PPE Detector for Laboratory Safety was chosen by the PneumoScan.ai team, because of:● High accuracy. On average, the solution demonstrates the accuracy of no less than 97% in detecting PPE. It classifies and identifies individuals, as well as four PPE items in real-time, with minimum latency.● Flexibility. The solution is available on AWS Marketplace, and it can be integrated with other AWS products. It can be hooked up with AWS services for storage, compute, analytics, and more.● Safety. The detector operates on a Shared Security Responsibility model, meaning that AWS guarantees infrastructure security while giving customers security controls for the data and applications that are deployed and stored in the cloud.
Having accessed the ML model from AWS Marketplace, the team implemented it using Amazon SageMaker, to capture, process, and analyze live video streams from CCTV cameras. Though they faced a few challenges working with Amazon SageMaker, creating endpoints of the pre-trained TensorFlow model, assessing accuracy of TensorFlow and SageMaker models, and calling Amazon SageMaker model endpoints using Amazon API Gateway and AWS Lambda, the team managed to build a working product in time.

Value Delivered

97%
accuracy of detecting PPE violations
Hackathon project completed in two weeks
Easily integrated with AWS ML models
anti covid by vitech lab

The pandemic of COVID-19 has presented unique challenges to governments, healthcare providers, and diverse communities all over the globe. However, every challenge is also an opportunity. Vi Ly, Sagar Bansal, and Mohit Gadkari proved that. In just two weeks, their team managed to design and build a comprehensive AI tool to help doctors and medical staff in their fight against the novel coronavirus.

Over the course of two weeks, they learned how to implement image recognition models from AWS marketplace, curate and manage data, train and finetune ML/DL models, and deploy and integrate all these models into a cloud-hosted web application.

With regard to VITech Lab’s PPE Detector solution, the team was able to take advantage of it right off the bat. When deployed, the solution demonstrated ~97% accuracy in detecting PPE non-compliance, and it did not consume lots of cloud resources. It proved to be extremely easy to integrate with other models and elements of the PneumoScan app.

The PneumoScan.ai team was satisfied with the final product they built for the hackathon and managed to land one of the winning spots at AWS Marketplace Developer Challenge: ML Solutions. Hopefully, their effort will not go unnoticed, and medical staff globally will benefit from their innovation to curb the spread of COVID-19.

Want to try PPE monitoring automation?

Feel free either to request a personal consultation concerning your case or try the demo now!