A naukri.com initiative
VentureBeat
2d
205
Learn how GE Healthcare used AWS to build a new AI model that interprets MRIs
- GE Healthcare has developed a full-body 3D MRI research foundation model (FM) that can use full 3D images of the entire body using AWS. The model is based on more than 173,000 images from over 19,000 studies and can support image-to-text searching, link images and words and segment and classify diseases in what's considered an evolutionary research phase. Already it has outperformed other publicly available research models in tasks including classification of prostate cancer and Alzheimer’s disease. GE Healthcare was able to train the model with five times less compute than previously required.
- This new model overcomes the massive hurdle of using 2D approximations of MRI images, allowing for intricate anatomical analysis, even for complex cases of brain tumours, cardiovascular diseases and other skeletal disorders. GE's FM is multi-modal and has exhibited up to 30% accuracy in matching MRI scans with text descriptions in image retrieval.
- The FM could enable real-time analysis of 3D MRI data, providing faster and more accurate diagnosis and treatment. GE HealthCare chief AI officer, Parry Bhatia, who built the model from the ground up using AWS, said the possibilities are huge.
- One of GE's flagship products is deep learning-based reconstruction algorithm AIR Recon DL, which removes noise from raw images, improving signal-to-noise ratio and cutting scan times up to 50%. Since 2020, 34 million patients have been scanned with AIR Recon DL.
- The MRI process requires a few different types of datasets to support various techniques that map the human body. To overcome challenges presented by diverse datasets, developers introduced a “resize and adapt” strategy so that the model could process and react to different variations.
- Semi-supervised student-teacher learning was also employed wherein two different neural networks were trained on both labeled and unlabeled data, with the teacher creating labels that help the student learn and predict future labels. This ensures the model performs well in hospitals with fewer resources, older machines and different kinds of datasets.
- According to the company, other challenges that needed to be addressed were the cost optimization as well as the limited computational power for 3D images that are gigabytes in size. GE Healthcare used Amazon SageMaker, Nvidia A100 and tensor core GPUs for AWS-based training to address these challenges.
- GE's FM's multimodality provides detailed information in one scan which can improve medical procedures like biopsies, radiation therapy and robotic surgery, Dan Sheeran, general manager for healthcare and life sciences at AWS, said.
- Sheeran also marvelled that with generative AI, GE's model could be expanded into areas such as radiation therapy or reduce scan time during X-rays and other procedures that currently require patients to sit still in a machine for extended periods.
- While GE's current model is specific to the MRI domain, researchers see great opportunities to expand into other areas of medicine by pre-training a single foundation model that can serve as a basis for other specialised fine-tuned models downstream.
Read Full Article
12 Likes
For uninterrupted reading, download the app