LG Group’s AI think tank uses AWS to identify cancer risks earlier
Amazon SageMaker helps LG AI Research reduce genetic testing time from two weeks to less than one minute to accelerate patient diagnosis
At AWS re:Invent, Amazon Web Services (AWS), an Amazon.com, Inc. company (NASDAQ: AMZN), today announced that LG AI Research, the artificial intelligence (AI) research hub of South Korean conglomerate LG Group, has used the world’s leading cloud to develop its new pathology foundation model (FM) for earlier cancer diagnosis and treatment. The histopathology image-specific model, EXAONEPath, can securely analyze microscopic images of tissue samples from cancer patients to reduce genetic testing times from two weeks to less than one minute, helping medical professionals improve the speed and effectiveness of treatments.
EXAONEPath achieves an average accuracy of 86.1% across six benchmarks in correctly classifying cellular-level visual features, which is comparable to other leading pathology FMs trained on far larger data sets. With AWS, LG AI Research transfers terabytes of data to the cloud in less than an hour, shortening model training time from 60 days to one week. This improves EXAONEPath’s performance in diagnosing and detecting cancer, leading to improved clinical outcomes for patients. By running on AWS, LG AI Research can also reduce its data management and infrastructure costs by approximately 35% and cut data preparation time by 95%.
“AWS allows us to accelerate our AI research, bringing accessible and rapid cancer screening closer to reality,” said Hwayoung (Edward) Lee, vice president of LG AI Research. “By leveraging AWS, we can train our pathology model on a vast dataset faster—securely, and cost-effectively. This enhances EXAONEPath’s processing capabilities for delivering personalized, efficient cancer treatments to improve patient outcomes. EXAONEPath has the potential to transform cancer diagnosis and treatment globally.”
Leveraging Amazon SageMaker, LG AI Research trained and deployed its large-scale EXAONEPath model within eight months, using 285 million data points and more than 35,000 high-resolution tissue sample images. Processing and training AI models with extremely large datasets requires immense storage, high-speed data transfer, and significant compute power. With AWS and NVIDIA GPUs, LG AI Research is accelerating training and inference for its deep learning workloads.
LG AI Research uses Amazon S3 to store and retrieve massive volumes of data that are crucial for research. Amazon FSx for Lustre provides sub-millisecond latencies and delivers hundreds of gigabytes per second of throughput, essential for applications that require rapid access to large datasets. This high-performance file and storage system enables parallel data processing and analysis, significantly reducing the time needed to gain insights.
“The healthcare industry is making rapid progress in its use of AI on AWS to accelerate diagnoses and get patients into treatment faster,” said Dan Sheeran, general manager, Healthcare and Life Sciences at AWS. “Using AWS, LG AI Research can develop and use EXAONEPath at an unprecedented scale, reducing data processing and model training times and improving accuracy. This will allow healthcare providers to improve cancer diagnoses and treatments, reduce wait times, and personalize patient care.”
EXAONEPath is part of LG AI Research’s EXAONE, a 300-billion-parameter multimodal foundation model by LG AI Research, which was also built on Amazon SageMaker and Amazon FSx for Lustre. LG AI Research will continue to update and improve EXAONEPath by training it to detect more types of cancer using additional pathology images.
Explore AITechPark for the latest advancements in AI, IOT, Cybersecurity, AITech News, and insightful updates from industry experts!
