Figure 1: SubtlePET’s AI-powered enhanced PET scan imagery (picture courtesy of Subtle Medical)
Subtle Medical’s artificial intelligence (AI)-powered technology allows doctors to enhance images from up to four times faster scans while maintaining clinically equivalent image quality when compared to standard PET scans. In addition to the increased revenue potential for healthcare institutions, faster scans provide patient benefits through shorter exposure times for children and those undergoing repeat scans. Subtle Medical uses the IntelⓇ Distribution of OpenVINO™ Toolkit to accelerate and seamlessly integrate their SubtlePET technology into the existing clinical computing infrastructure and workflow.
We have to work with the edge computing devices already in the hospital. In most cases, this is an IntelⓇ CPU. OpenVINO™ allows us to get an additional 2x inference speed to provide a product that integrates seamlessly into the radiologist's current workflow.
- Liren Zhu, Head of Engineering, Subtle Medical
The deployment of new technologies into a hospital often requires leveraging the existing infrastructure. Subtle Medical recognized that most radiologists work with existing infrastructure based on IntelⓇ architecture without additional hardware accelerators. The IntelⓇ Distribution of OpenVINO™ Toolkit takes advantage of the deep neural network optimizations found in the advanced vector extensions of IntelⓇ XeonⓇ processors and streamlines the inference performance of convolutional neural network (CNN) workloads through graph-level optimizations. In short, Subtle Medical can increase the inference speed on their product through Intel software without requiring the healthcare institution to upgrade its hardware infrastructure.
What the Results Show:
On integrating the IntelⓇ Distribution of OpenVINO™ Toolkit into SubtlePET, Subtle Medical observed that radiologists experienced a 2.1x improvement in inference speed (Figure 2). This performance improvement is critical to Subtle Medical’s focus on faster image acquisition and workflow, differentiating them from similar solution providers that focus on post-processing and computer-aided diagnosis products.
Figure 2. IntelⓇ Distribution of OpenVINO™ Toolkit boosts SubtlePET inference speed by 2.1x1. *Stock indicates TensorFlow was not optimized with IntelⓇ MKL-DNN instructions.
See this solution in action at the Artificial Intelligence Conference in New York City, April 15-18 2019. Join Enhao Gong, Founder and CEO of Subtle Medical for his 40 minute session “How to use AI to improve Efficiency, Safety and Patient Satisfaction in Radiology”
Clinical radiology is faced with several clinical issues: 1) improvement in imaging efficiency, 2) reduction of risks, 3) high imaging quality. Subtle Medical provides Deep Learning/AI solutions, powered and accelerated by industry solutions such as Intel® OpenVINO™ to address these problems by enabling faster MRI, faster PET and low dose imaging, providing real clinical and financial benefit to hospitals.
Benchmark testing was performed Jan 2019 by Subtle Medical on custom CNN topology (proprietary U-net variant) and dataset.
Compute AWS p3.2xlarge Intel® Xeon® E5-2686 v4@ 2.30 GHz with 8 virtual CPU cores, 61GB system RAM
Intel® distribution for Python v3.5, TensorFlow 1.12, Intel® OpenVINO™ 2018.2.319, gcc 5.3
Performance results are based on testing as of dates shown in configuration and may not reflect all publicly available security updates. No product can be absolutely secure. See configuration disclosure for details.
Optimization Notice: Intel's compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors.
Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more complete information visit: http://www.intel.com/performance