Executive Summary:

While the university client had experience in building neural networks for radiology solutions, they needed help with scaling their training across a cluster of GPU’s in the cloud, and in adapting open source models to their datasets. 

• The university approached Theta Tech AI because they knew about our experience in building AI models utilizing healthcare data, and needed a team that could ramp up quickly. 
• We were able to rapidly adapt their data and algorithms to a streamlined cloud infrastructure, help them integrate some open source AI models, and build a whole cloud training infrastructure with live dashboards across multiple GPU’s.
• We delivered a multi-sequence 3D radiology segmentation model based on their data, sent a torch checkpoint file, along with Weights & Biases dashboards reporting the Dice scores. We gave them cloud infrastructure that allows the model to be easily retrained as the clients gather more data in the future.
• Other expert AI consultants or software vendors would have required at least 3 times as long, and would have needed education on proper handling of 4-sequence MRI data, and would not have necessarily known the best radiology models to use.

‍