Presented By
O’Reilly + Intel AI
Put AI to Work
April 15-18, 2019
New York, NY

How to use AI to improve the efficiency, safety and patient satisfaction in Radiology

Enhao Gong (Subtle Medical), Greg Zaharchuk (Stanford University)
2:40pm3:20pm Thursday, April 18, 2019
AI Business Summit, Case Studies
Location: Sutton North/Center
Secondary topics:  AI case studies, Computer Vision, Health and Medicine

Who is this presentation for?

Executives in healthcare industry (hospitals, pharma, imaging centers, etc.)

Level

Intermediate

Prerequisite knowledge

Background in healthcare and radiology industry

What you'll learn

Take-away: 1. Radiology is faced challenges to improve imaging quality and efficiency. 2. AI can not only be used for assisted diagnosis but also for improving imaging workflow and productivity, enabling direct financial returns to hospitals. 3. Solutions developed by Stanford University and Subtle Medical uses AI to significantly improve speed and quality of radiology. 4. The products are FDA cleared and pending and well received by clinicians. Evaluated at Stanford, UCSF, Hoag hospitals, RadNet etc. 5. Powered by industry solution such as Intel OpenVINO, the deployment can be highly accelerated and enables real-time solutions that benefit clinical applications.

Description

Take-away:
1. Radiology is faced challenges to improve imaging quality and efficiency.
2. AI can not only be used for assisted diagnosis but also for improving imaging workflow and productivity, enabling direct financial returns to hospitals.
3. Solutions developed by Stanford University and Subtle Medical uses AI to significantly improve speed and quality of radiology.
4. The products are FDA cleared and pending and well received by clinicians. Evaluated at Stanford, UCSF, Hoag hospitals, RadNet etc.
5. Powered by industry solution such as Intel OpenVINO, the deployment can be highly accelerated and enables real-time solutions that benefit clinical applications.

Currently the top 3 major issues in clinical MRI exams are: 1) MRI and PET exams are very slow, 2) usage of radiation and contrast dosage result in risks. 3) Insurance pressure and patients’ need urge radiology practice to improve both quality and productivity. Subtle Medical provide Deep Learning solution, powered and accelerated by industry solution such as Intel OpenVINO, to address these problems by enabling 4x-10x faster MRI scans, 4x faster PET scans and 10x dosage reduction.

【Method】

1) 4x-10x faster MRI (and PET)
For faster MRI, Deep Learning solution is proposed to replace conventional iterative optimization based algorithm. Using ResNet based recurrent structure and Generative Adversarial Network based adversarial loss, the model successfully reconstructs images from low-quality MRI into high-quality MRI with significantly better Signal-to-Noise Ratio and resolution information. This solution generates results much more efficient than conventional iterative methods with much better image quality and accuracy.
2) 10x contrast dose reduction
Gadolinium Deposition is one of the most urgent issues facing radiology community. In this work, we validated both technical performance and clinical applicability of the developed Deep Learning (DL) technology that algorithmically restores/boosts the contrast information in MRI. We further verified he generalization and robustness of the DL solution and show it improves workflow while maintaining diagnostic quality.

【Results】

The proposed DL solution demonstrate significant superiority over conventional algorithms and non-inferiority with standard slow and high-dose exams. Quantitative metrics demonstrated consistent (>5dB in PSNR and >10% in SSIM) and significant (p<0.001) quality improvement of the deep learning based solution, compared with low-dose CE-MRI. Qualitative ratings showed non-significant differences in image quality between the DL-enhanced and acquired full-dose CE-MRI images, which was also verified with reader study.

【Discussion and Conclusion】

Deployment with Industry solution such as Intel OpenVINO provides ~10x speed up of the inference by optimizing the Resnet and u-net based network model. This enables real-time feedbacks for clinical application. This application, with great accuracy, scalability and efficiency, can significantly improve the workflow for clinical radiology to provide faster, safer and smart imaging exams to patients.

Photo of Enhao Gong

Enhao Gong

Subtle Medical

Enhao Gong is founder and CEO at Subtle Medical. He is a serial entrepreneur and PhD in Electrical Engineering at Stanford, with research focus on applying AI and deep learning to improve reconstruction, analysis and quantification in medical imaging . His work that applies AI to accelerate and reduce dose for MRI and PET has been featured in numbers of academic journals and clinical conferences. Dr. Gong won several awards including 2018 Forbes China 30-under-30 for his work at Subtle Medical, an AI+radiology startup from Stanford and the winner of 2018 NVIDIA Inception Award in AI+Healthcare.

Photo of Greg Zaharchuk

Greg Zaharchuk

Stanford University

Greg Zaharchuk is a radiologist and professor in radiology at Stanford University and a neuroradiologist at Stanford Hospital. His research interests include deep learning applications in neuroimaging, imaging of cerebral hemodynamics with MRI and CT, noninvasive oxygenation measurement with MRI, clinical imaging of cerebrovascular disease, imaging of cervical artery dissection, MR/PET in neuroradiology, and resting-state fMRI for perfusion imaging and stroke.

Leave a Comment or Question

Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?

Join the conversation here (requires login)