Helping Leading Pharma Client Automate Disease Discovery with Amazon Sagemaker

About the Client

One of the leaders in pharmaceutical industry, the client is a multi-national corporation employing over 20,000 people.

Business Challenge

The client wanted to leverage state-of-the-art technologies to effectively use deep learning for disease detection and improve the accuracies by making it comparable to or better that what human radiologists can achieve. They wanted the entire tech stack engineered using Cloud computing making the disease detection uniform, scalable, efficient and universally accessible from any corner of the globe.

Solution

Brillio solution revolves around radiological diagnostics wherein the challenge was to ascertain the incidence of diseases by leveraging National Institute of Health (NIH) X-ray dataset. Getting X-rays diagnosed through Radiologists can be an expensive process both in terms of time and money. Also, the quality of diagnostics can vary a lot within different regions as the same quality of experts aren’t available uniformly. Our solution potentially can help bridge and establish uniformity in terms of diagnostic standards while maintaining highly accurate and easily interpretable results.

Our automated solution helps various data scientists to implement inbuild or custom machine learning algorithms to train and deploy the model.

AWS services used

AWS Cloud Formation Templates, AWS SageMaker Notebook, AWS Lambda, AWS IAM, AWS SNS, AWS S3, AWS ECR, AWS Cloud Watch, AWS EFS

Benefits and Business Impact

With Brillio’s solution – an improvement of over 10 percent relative to the accuracy of client organization was achieved.  It ensured uniformity with respect to disease detection vis-à-vis human radiologists’ prediction. There was significant savings in cost and time for lung disease prediction.  It is also slated to helped reduce healthcare services cost for those who can’t afford the expertise of radiologists.

This project also paved the way to incorporate newer diseases into the algorithm after training it afresh. This enabled doing away with the need to hire expensive experts for newer disease detection as well.