Case Study: Duke University accelerates clinical-grade ultrasound imaging research with Google for Education

Duke researchers leverage deep learning on Google Cloud to improve medical imaging quality

Duke University researchers needed a way to make ultrasound imaging more consistent and clinically useful across different scanner manufacturers. Using Google for Education’s Google Cloud tools, including Colab and Google Compute Engine, the team set out to build an open-source framework that could translate raw ultrasound data into clinical-grade images.

Google for Education helped Duke develop MimickNet, a deep-learning model trained on Google Compute Engine and built in Colab to mimic proprietary post-processing from top clinical scanners. The approach enabled researchers to run up to 100 experiments in parallel and get results in one day instead of three weeks, while achieving a mean SSIM score of 0.940, making the outputs nearly indistinguishable from commercial clinical images.

Duke University

Ouwen Huang

Head of Biomedical Engineering