Thailand’s Siriraj Hospital has developed a chest X-ray imaging solution for tuberculosis that greatly reduces the time needed to analyse scans and expedites decision-making to save more lives.
The solution is developed on and powered by two NVIDIA DGX A100 systems and NVIDIA Clara software — making the hospital the first in Southeast Asia to deploy these NVIDIA technologies for medical research and clinical applications.
Siriraj Hospital is Thailand’s largest government hospital with more than 2,500 beds and over 8,000 outpatients daily. It is the university hospital of Mahidol University, whose Faculty of Medicine focuses on research, medicine and healthcare.
Massive amounts of data to analyse
The hospital’s radiology department is the first in Thailand to explore research and clinical application of AI on radiology images, including plain radiography, computed tomography (CT), magnetic resonance imaging (MRI), and ultrasonography.
Each day, the hospital collects 1,000 to 2,000 CT scan images and 500 to 1,000 MRI images for each patient. Multiply this by the 200 CT studies and 50 to 60 MRI studies each day and the amount of data is massive, making it difficult for radiologists to analyse and interpret in a timely manner required for clinical efficiency.
“We have to focus on image classification to pick up the possible lesions among the vast number of images to help radiologists interpret the data. The volumetric measurement of organ and tumour on initial evaluation and follow-up are also a time-consuming process that requires manual drawing,” said Associate Professor Trongtum Thongdee of Radiology Department, Faculty of Medicine Siriraj Hospital at Mahidol University.
Siriraj Hospital leveraged AI in a solution that includes two NVIDIA DGX A100 systems, two workstations with NVIDIA V100 GPUs, six workstations with NVIDIA P6000 GPUs, as well as NVIDIA Clara Deploy and Train, CUDA, cuDNN, and TensorRT.
Designed as a universal platform for AI workloads, the five-petaflop NVIDIA DGX A100 features NVIDIA A100 Tensor Core GPUs to enable enterprises to consolidate training, inference and analytics into a unified, easy-to-deploy AI infrastructure that includes direct access to NVIDIA AI experts.
NVIDIA Clara Train features techniques such as AutoML, privacy-preserving federated learning and transfer learning to help researchers build robust AI models. NVIDIA Clara Deploy provides a reference framework to take an AI model and write an application workflow around it to enable interfacing in a hospital environment.
Faster and more accurate
AI has proven to have high accuracy for segmentation of medical images.
“Training a 3D dataset with NVIDIA V100 takes just two days compared to 80 days with a CPU. The GPU is 40x faster and accuracy is increased. The CPU is not practical in research and clinical applications,” said Thongdee.
“Furthermore, because of the high throughput of NVIDIA GPUs, clinical workflows can proceed smoothly without any bottlenecks. In emergency situations, faster interpretation of images can help doctors make better treatment decisions to save patients’ lives,” he added.
“NVIDIA Clara is enabling hospitals to innovate and accelerate the journey to precision medicine. The decision by Siriraj Hospital to invest in an NVIDIA-based AI solution empowers its radiologists to make faster and more accurate decisions that in turn will save more lives, as well as create a blueprint that healthcare facilities in the region can use to adopt leading-edge technologies that will dramatically improve patient care and outcomes,” said said Dennis Ang, Director of Enterprise Business for the SEA and ANZ Region at NVIDIA.