Cerebriu trains life sciences AI models 3X faster on OCI AI infrastructure

The Danish startup chose Oracle Cloud Infrastructure to develop, train, and deploy its patented AI models to improve radiology.


As we create generative AI solutions in radiology to improve clinical care and improve patient outcomes, OCI provides performance and economic benefits for training our data-intensive models. For the same price, we get more GPU compute and memory for results that are 3X faster than AWS and Azure.

Akshay PaiCo-Founder and CTO, Cerebriu

Business challenges

The healthcare company’s mission is to boost innovation in radiology with generative AI by streamlining brain MRI exam workflows and helping to ensure accurate first-time image capture. Cerebriu’s software detects possible critical findings during MRI scanning while the automated MRI workflow helps ensure that the right image is obtained at an early stage so that findings are promptly flagged for radiologists. This shortens intervention time, which improves acute patient treatment and outcomes.

To meet the resource-intensive process of AI model training while keeping costs down, Cerebriu required computing infrastructure that would scale to meet its global expansion plans and growing product development pipeline.

Why Cerebriu chose Oracle

Cerebriu chose Oracle Cloud Infrastructure (OCI) Compute bare metal instances with NVIDIA GPUs to accommodate the startup’s operational and budgetary constraints over a 12-month term. Oracle offered the company a greater number of available GPU instances than AWS and Azure for the same price, ease, and efficiency when launching new GPU instances, in addition to dedicated technical support.

The startup expects to use OCI’s distributed cloud, including numerous public cloud regions worldwide, to serve hospitals across Europe, the Middle East, and Asia with reliability, performance, and security.  


After moving all of its AI model training to OCI AI infrastructure with NVIDIA A100 40GB GPUs, Cerebriu accelerated parallel processing by 3X. The high GPU memory allows for training multiple images in parallel, helping with gradient descent during the training phase and increasing accuracy and speed. Also, training time dropped from eight weeks to three for greater efficiency. The company plans to add model inferencing on OCI to assist radiologists within MRI diagnostic workflows with insights to improve treatment and patient outcomes. Using OCI, Cerebriu can scale to any size workload with its partners, such as Siemens Healthineers and other manufacturers of MRI machines, to aid hospital providers across the globe.

Published:June 28, 2024