Intel and GE Healthcare's X-ray machine uses embedded AI to prioritize scans

Technology experts worked with radiologists to find the best ER use cases for artificial intelligence and computer vision.

AI in medicine: Tech that can improve the patient-doctor experience Learn how using artificial intelligence in healthcare can benefit the patient—regardless of gender, age, or race. AI can especially make a difference in the medical care of women and premature babies.

GE Healthcare and Intel have a new software + hardware combination that can get the most critical X-rays to the top an ER doctor's review stack.

The new solution puts GE's mobile, digital X-ray system, its artificial intelligence-powered Critical Care Suite, and Intel's computer vision platform OpenVINO in one product. The Critical Care algorithms are embedded in the imaging devices to speed up image processing time. OpenVINO improves computing power to allow hospitals to deploy this new service on existing hardware.

"So, the thing that's actually capturing the images is also doing the processing," said Todd Minnigh, CMO X-Ray, GE Healthcare, in a press release. "It's not in the cloud and not on a server downstream somewhere."

To build this service, GE and Intel asked radiologists and technologists how technology could improve the X-ray review process. The healthcare providers had two requests: Help us spot rare, life-threatening, and easy-to-overlook conditions, and automatically move those cases to the top of the priority list for review.

Pneumothorax is that kind of condition. This happens when air leaks into the space between the lungs and chest wall, causing the lung to collapse. The problem is diagnosed with an X-ray. There are only about 200,000 cases each year. 

The new product also can flag critical cases like pneumothorax and send to radiologists for immediate review. 

Dr. Rachael Callcut, Associate Professor of Surgery at UCSF, a surgeon at UCSF Health and Director of Data Science for the Center for Digital Health Innovation, worked with GE Healthcare to develop Critical Care Suite. "When a patient's X-ray is taken, the minutes and hours it takes to process and interpret the image can impact the outcome in either direction. AI gives us an opportunity to speed up diagnosis, and change the way we care for patients, which could ultimately save lives and improve outcomes," Callcut said in a press release.

SEE: Telemedicine, AI, and deep learning are revolutionizing healthcare (free PDF)

GE Healthcare received FDA approval in September for Critical Care Suite, which includes the X-ray processing enhancements. In addition to flagging X-rays that suggest life-threatening problems, Critical Care Suite includes an intelligent auto-rotate algorithm. This is typically a manual process. One estimate suggests that automating this task could save technicians about 70,000 clicks a year.

Katelyn Nye, X-ray global product manager, artificial intelligence and analytics at GE Healthcare, said in a press release that the goal was to make the new AI service available on its large install base of Intel systems through software upgrades or via a clear hardware upgrade path for older systems.

"Our approach to building intelligent machines is avoiding any additional steps, workflow, or infrastructure if the task can be performed with what the customer already has today," Nye said.

Improving computer vision 

Intel's OpenVINO toolkit allows hospitals to use this new analysis capability on existing hardware. OpenVINO is short for Open Visual Inference and Neural Network Optimization. This software combines computer vision and deep learning inference to allow an algorithm to spot X-rays that should be reviewed first. The process was tested on more than 100 public and custom models to understand what warning signs to look for in chest X-rays. 

Intel worked with several healthcare companies to implement OpenVinoPhilips Medical used the platform in their CT scanners and imaging machines to improve bone-age-prediction models. In a white paper about Intel's and Philip's work around AI inferencing of healthcare workloads, researchers explain how the deep learning inference models were developed. The product team found that a 37.7x speed improvement for the lung-segmentation model over the baseline measurements.

The toolkit is based on convolutional neural networks and extends workloads across Intel hardware to improve performance. OpenVINO also:

  • Enables deep learning inference at the edge
  • Supports heterogeneous execution across computer vision accelerators—CPU, GPU, Intel® Movidius™ Neural Compute Stick, and FPGA—using a common API
  • Speeds up time to market via a library of functions and pre-optimized kernels
  • Includes optimized calls for OpenCV and OpenVX*

Intel built the toolkit for software developers and data scientists who work on computer vision, neural network inference, and deep learning deployment capabilities and who want to accelerate their solutions across multiple platforms.

Also see

screen-shot-2019-10-29-at-12-18-05-pm.png

When comparing the OpenVINO toolkit to standard methods of image analysis, OpenVINO performed better, particularly with parallel execution of the software.

Image: Intel