Each year, more than 6.5 million brain scans are performed in the United States alone. With this amount of data, it is difficult for clinicians to carefully analyze healthy / pathological cases and compare them with historical databases.
To meet this challenge, our team is working with a technology startup to develop tools to automatically analyze brain scans, check them for multiple pathologies (such as tumors or aneurysms), and obtain insights by comparing the analyzed images with several historical pathology datasets (including comparing changes in patients' results over time) - all, in order to better support physicians' day-to-day decision-making.
Our focus is on processing medical 3D scans (including MRI and MRA imaging), using imaging techniques and neural networks in conjunction with large databases of historical medical data to segment normal and pathological structures, as well as to detect anomalies in patients' brains. For regulatory (FDA) reasons and requirements, we have taken additional steps to prepare a versioning system to track machine learning models and data.
Client benefits: With this application, the diagnosis of brain tissue pathology can be carried out faster and more accurately, minimizing the risk of late- and mis-diagnosis, thus protecting patients' lives and reducing treatment costs.
Industry: Healthcare Project Category: AI/ML, image processing, software development, data analytics, MLOps Lead Member: Datarabbit Technology used: Python, Pytorch, Terraform, AWS, DVC, VTK