Our client is a Chilean company with knowledge belonging to different scientific areas. Using this knowledge, they try to improve the tools used by the medical team to diagnose their patients.
They developed a model applying deep learning to develop markers directly from medical images that describe underlying biological processes within tissues.
The intention from the beginning was to publish the model. In this way, a user could upload magnetic resonances to the system and obtain processed information. But the difficulty was not only in adapting the model to a different framework, but also in having a 3D visualization of the result obtained, in addition to handling large files that could hinder the user experience.
A web platform was created that allows users to upload a list of cases to their accounts. Each case corresponds to a patient, so it has clinical and demographic information in addition to the MRI itself. After uploading the information to the site, it is processed in the background and the user is notified via email when the process ends. From that moment on, the user can download the calculated volumetric information, in addition to being able to manipulate the resonances on the web using Paraview Glance.
The platform was implemented with nest and react, and deployed on AWS, where services such as Amplify, Cognito, SageMaker, S3, EC2, and Lambda were used. The model and the processing pipeline were implemented with Python and adapted to be able to deploy it also on AWS.
We emphasize that in two months of work, a very good result was achieved, working closely and with a good relationship with the client. It was possible to adapt the 3D view as it was defined and the response times, upload, download, and manipulation of the information are much more than acceptable. Finally, the platform was fully functional in production and several workshop sessions were coordinated to instruct the client on the architecture and implementation of the solution.