HEALTHCARE OUTCOMES

The use of edge AI in medical imaging offers the possibility of enormous benefits to stakeholders throughout the healthcare sector.

On the provider side, edge AI imaging can improve diagnostic accuracy, boost physician efficiency, speed case processing timelines, and reduce the burden on overstretched medical personnel. Patients benefit from shorter wait times for their diagnostic test results and a better overall quality of care.

But it can be challenging to develop AI-powered solutions needed to make this promise a reality. The computing requirements to implement edge AI in medicine are high, which has historically made it both difficult and expensive to obtain adequate computing resources. It can also be hard to customize the underlying hardware components well enough to suit medical imaging use cases.

It’s a frustrating situation for anyone wanting to offer innovative AI-enabled imaging solutions to the medical sector—because while the market demand certainly exists, it’s not easy to build products that are effective, efficient, and profitable all at the same time.

But now independent software vendors (ISVs), original equipment manufacturers (OEMs), and system integrators (SIs) are better positioned to innovate edge AI-enabled medical imaging solutions. The prevalence of rich edge-capable hardware options and the increasing availability of flexible AI solution reference designs make this possible.

AI Bone Density Detection: A Case Study

The AI Reasoning Solution from HY Medical, a developer of computer vision medical imaging systems is a case in point. The company wanted to offer clinicians an AI-enabled tool to proactively screen for possible bone density problems in patients so that timely preventive steps could be taken.

They needed an edge AI deployment that would put the computational work of AI inferencing closer to the imaging devices, thereby reducing network latency and bandwidth usage while ensuring better patient data privacy and system security. But there were challenges.

The edge computing power requirements for a medical imaging application are high due to the complexity of the AI models, need for fast processing times, and sheer amount of visual data to be processed.

In addition, special challenges involved developing an AI solution for use in medical settings: an unusually high demand for stability, the need for waterproof and antimicrobial design elements, and the requirement that medical professionals approve the solution before use.

The solution can automatically measure and analyze a patient’s bone density and tissue composition based on the #CT scan data, making it a valuable screening tool for #physicians. HY Medical (Huiyihuiying) via @insightdottech

HY Medical leveraged Intel’s medical imaging AI reference design and Intel^® Arc^™ graphics cards to develop a solution that takes image data from CT scans and then processes it using computer vision algorithms. The solution can automatically measure and analyze a patient’s bone density and tissue composition based on the CT scan data, making it a valuable screening tool for physicians.

The solution also meets the stringent performance requirements of the medical sector. In testing, HY Medical found that their system had an average AI inference calculation time of under 10 seconds.

Intel processors offer a powerful platform for medical edge computing, which allows the company to meet its performance goals with ease. Intel technology also provides tremendous flexibility and stability, enabling the wide-scale application of this technology in bone density screening scenarios.

Reference Designs Speed AI Solution Development

HY Medical’s experience with developing their bone density screening solution is a promising story—and one that will likely become more common thanks to the availability of AI reference designs. These reference architectures make it possible for ISVs, OEMs, and SIs to develop medical imaging solutions for a hungry market both quickly and efficiently.

Intel’s edge AI inferencing reference design for medical imaging applications supports this goal in several ways:

Tight integration with high-performance edge hardware: Ensures that solutions built with the reference design will be optimized for computer vision workloads at the edge. The result is improved real-world performance, better AI model optimization for the underlying hardware, and increased energy efficiency.

Flexible approach to AI algorithms: Because different software developers work with different tools, multiple AI model frameworks are supported. Models written in PyTorch, TensorFlow, ONNX, PaddlePaddle, and other frameworks can all be used without sacrificing compatibility or performance.

AI inferencing optimization: The Intel^® OpenVINO^™ toolkit makes it possible to optimize edge AI models for faster and more efficient inferencing performance.

Customized hardware support: The reference design also factors in the special needs of the medical sector that require customized hardware configurations—for example, heat-dissipating architectures, low-noise hardware, and rich I/O ports to enable connection with other devices in clinical settings.

The result of reference architectures such as this one is that they shorten time-to-market and reduce the inherent risks of the product development phase, giving innovators a clear path to rapid, performant, and profitable solution development. That’s a win for everyone involved—from solutions developers and hospital administrators to frontline medical professionals and their patients.

The Future of AI in Medical Imaging

The ability to develop innovative, tailored solutions quickly and cost-effectively makes it likely that far more AI-enabled medical imaging solutions will emerge in the coming years. The potential impact is huge, because medical imaging covers a lot of territory—from routine screenings, preventive care, and diagnosis to support for physicians treating diseases or involved in medical research.

Hospitals will be able to use this technology to improve their medical image reading capabilities significantly while reducing the burden on doctors and other medical staff. The application of edge AI to medical imaging represents a major step forward for the digital transformation of healthcare.
 

Edited by Georganne Benesch, Editorial Director for insight.tech.