AWS and machine learning enabled the client to analyze medical images effectively

A leading pharmaceutical and biotechnology company that conducts research, production, and distribution of drugs and medical products on a global scale.

The project aimed to deliver a solution to the client, enabling effective analysis of the quality of sensors for medical devices manufactured on the production line. This tool was intended to operate in high availability mode with an automatic scaling mechanism while ensuring that the training model would be based on powerful machines with a GPU processor.

One of our clients’ areas of activity is to provide doctors with the right conditions for effective disease diagnosis, which is made possible through the practical analysis of the quality of sensors for medical devices. The main challenge in the mentioned quality analysis was the unreliable human factor, which increased the percentage of defective devices released for use.

After conducting an in-depth requirements analysis, the Billennium team designed a full application infrastructure in the AWS environment, adjusting it to the client’s precisely defined needs.  

Our team has built an application logic that enables effective processing of medical sensor images through machine learning. After training, reports from them are automatically saved in object data stores (AWS S3 buckets), where they can be read and analyzed by data science specialists who work on improving learning algorithms. 

Thanks to a thorough analysis of the client’s needs and the use of best practices in the field of cloud native architecture, our team proposed and implemented changes in the infrastructure to best adapt it to the use of AWS services. The automatic scaling of microservices on the AWS EKS platform allowed for more effective work of the application, which processes more data in a shorter time. The combination of the S3 data warehouse with the application enabled direct access to reports and training models for users (data science specialists) while maintaining the separation of access layers to critical parts of the system.

The work of the Billennium project team resulted in a ready-to-implement, built-from-scratch application that allows the classification of medical sensor images using machine learning. 

Thanks to this high-performance tool for describing and training medical images, the client has gained a practical application that will streamline the production process of sensors for medical devices, eliminating the unreliable human factor in quality testing, which will significantly increase the reliability of these devices in everyday use.