United Imaging Healthcare Enhanced the Speed and Quality of MR Imaging

The product and engineering teams at United Imaging Healthcare found that reconstructing high-quality MR images requires extremely compute-intensive workloads. They also needed to leverage more streaming sensor data to achieve higher throughput in less time. United Imaging Healthcare began using NVIDIA’s high-performance computing (HPC) and AI platforms to improve their proprietary algorithms, which aim to improve the speed and accuracy of the imaging process. However, even with these algorithms, processing streaming sensor data still took too much time and expense, which didn’t meet the needs of the clinicians they served. Their team recognized that a full-stack, GPU-accelerated solution was necessary to bring intelligent MR machines with competitive, innovative capabilities to market.

The United Imaging Healthcare team used NVIDIA’s rich suite of GPUs and SDKs, which provided them with a wide range of acceleration, from AI inference to matrix calculations. Using NVIDIA data center GPUs, their products and engineering teams increased the performance and speed of their HPC algorithms, necessary for streaming sensor data to image pipelines. With the NVIDIA CUDA® Toolkit, the team optimized performance of their proprietary algorithm for developing and deploying AI applications across multiple environments.

United Imaging Health also used cuSolver, NVIDIA’s high-performance matrix computation library, to accelerate the reconstruction of MR images. To further improve performance, NVIDIA engineers tailored an algorithm that increased computational speed by more than 10X and significantly reduced the time required for MR reconstruction.

The NVIDIA AI Enterprise software suite came with enterprise support—critical to the success of developing and launching United Imaging Healthcare’s AI-enabled MR machine. NVIDIA experts, including DevTech engineers and solution architects, helped their teams train models, analyze code, and efficiently migrate to GPUs, reducing development costs. Finally, the United Imaging Healthcare team participated in NVIDIA Deep Learning Institute courses to deepen their understanding of how to develop on CUDA.

Using NVIDIA’s platform has had a significant impact on MR image reconstruction. After easily integrating AI into their medical instruments with NVIDIA’s GPUs and AI computing stack, United Imaging Healthcare’s team reduced image construction time by nearly 95 percent. To optimize this reduction, NVIDIA’s support teams helped United Imaging Healthcare further accelerate algorithms by 3X. Clinical teams can now limit the amount of time that patients spend in constrained MR machines, and hospitals can increase access of MR procedures to more patients.