Google’s work uncovered a way to estimate refractive error with high accuracy from retinal fundus photos using deep learning, previously unknown.

This will improve the screening process for diabetic eye disease. They also have projects researching assisting pathologists in detecting cancer [1].

We spoke to Lily Peng, MD, PHD, Product Manager at Google Brain AI Research Group to learn more about this landmark work in medical imaging:

AIMed: What trends are you currently seeing emerging in the space of medical imaging and biomedical diagnostics and AI that you find exciting?

Lily Peng: It’s pretty exciting to see many examples of how machine learning (ML) may be able to bring high-quality care to everyone, especially to rural and underserved communities where there is a shortage of experts.

There is a good amount of existing work on deep convolutional neural networks (a particular type of deep neural networks that has been optimized for imaging) for imaging in general. And the technology is fairly mature and has been applied widely in the consumer space.

AIMed: How does Google collaborate with healthcare providers to drive innovation and results in clinical settings?

Lily Peng: Our group collaborates closely with providers every step of the way — from development/validation to deployment of machine learning models.

For example, for the diabetic retinopathy (DR) project [2], we’ve been working with the same partners (Aravind, Sankara, EyePACS) for all of these phases for a few years now and it’s been amazing to see these ML models actually in use in a pilot setting in some of these clinics at Aravind.

AIMed: What is most important to know when brining these tools into clinical settings?

Lily Peng: Machine learning models are only as good as the training data you give it. So it’s quite important to make sure that you have really good ground truth for model training.

5) How can clinicians learn more about the potential for medical imaging and biomedical diagnostics coupled with AI?

Lily Peng: There now *many* resources freely available to learn about ML in general.

For a bit of a hands-on demo that helps you understand neural networks, I’d recommend playing around with this demo/tutorial: https://playground.tensorflow.org/

If you are technically savvy, TensorFlow is open sourced and there are also a lot of tutorials available to get you started.

8) Where do you see the space moving in the short term (2-3 years) and in the long term (9-10 years) and what will be the main driver?

Lily Peng: There are now several products approved by the FDA that will utilize Deep Learning. I think we’ll see more products being approved soon.

Long term, I think we’ll probably see a suite of AI powered solutions that will help doctors organize and make better use of the huge amount of medical information generated every day.

This interview originally appeared in AIMed Magazine issue 03, you can download the complete magazine here.

References

[1] https://ai.google/research/teams/brain/healthcare-biosciences

[2] https://ai.googleblog.com/2016/11/deep-learning-for-detection-of-diabetic.html

 

Interviewee: Lily Peng
Lily Peng, MD, PHD, is a non-practicing physician and product manager for a team that works on applying deep learning and other Google technologies and expertise to medical imaging.

Before Google, she was a product manager at Doximity, the “linkedin” for physicians, and a co-founder of Nano Precision Medical (NPM), a medical device start-up developing a small implantable drug delivery device. She completed her M.D. and Ph.D. in Bioengineering at the University of California, San Francisco and Berkeley. She received my B.S. with honors and distinction in Chemical Engineering from Stanford University.