Q&A: Stanford’s Curtis Langlotz on teaching AI to medical imaging students

The hype around artificial intelligence (AI) in medical imaging has led to plenty of discussions of its impact in clinical and academic spaces. To explore current and future implementations of AI in medical imaging at academic institutions, Health Imaging spoke with Curtis Langlotz, MD, PhD, Stanford University’s Medical Informatics Director for Radiology.

Langlotz discusses the implementation of AI and machine learning in clinical imaging as this year’s Dwyer Lecturer at the SIIM 2018 Annual Meeting from May 31 to June 2 in National Harbor, Maryland.

Health Imaging: Where do you think AI will first be deployed in medical imaging?

Curtis Langlotz, MD, PhD: Over the next decade, AI will be deployed throughout the image life cycle from image production to image interpretation. For example, machine learning algorithms will produce clearer images using less radiation and will alert technologists to suboptimal images at the scanner console. AI algorithms will also serve a triage function in both the developed and developing worlds, calling attention to images with urgent abnormalities and identifying images those that likely are normal. AI will also serve as a decision support function for radiologists, for example providing a differential diagnosis for unusual or difficult cases. We ultimately will see automatic drafting of radiology reports for some cases. Although there is a lot of hype around AI right now, some of the hype is real. I believe [AI] will have its earliest impact before the radiologist even sees images, such as image noise reduction, quality control and triage. Those areas are less regulated, don’t require high accuracy and provide rapid return on investment through improved efficiency.

What benefits can AI present to medical imaging trainees? Any disadvantages?

When MRI was first developed, many believed it would make radiologists superfluous. The theory was that the images were so clear, a radiologist wouldn’t be needed to interpret them. But we found that a deep understanding of the physics of MRI was required to recognize artifacts, to decide on the best imaging pulse sequences and to devise new ways to use MR imaging. That knowledge has been incorporated into radiology training, which is why radiologists still have primacy in MRI. AI’s future will be similar. Radiologists will learn how to apply AI in clinical practice and how to determine whether a system’s advice is sound. There is no question that AI will become an increasingly important part of what medical students and radiology trainees will need to learn.

Do you believe current medical imaging trainees are excited or discouraged by AI in medical imaging? Specifically, how has Stanford implemented AI technology into its medical imaging curriculum and research?

The trainees I encounter are excited. At Stanford, many of our trainees use their academic time to work as part of AI research teams. Some of our radiology trainees take advantage of popular graduate courses in imaging informatics, AI and computer vision.

For example, Andrew Ng, PhD, a computer science faculty member at Stanford, is one of the pioneers in deep learning and computer vision. We serve as faculty advisors in his AI for Healthcare Boot Camp for graduate students. We also collaborate with Fei Fei Li, PhD, a Stanford computer science professor, who developed the large ImageNet imaging database that has led to the recent progress in computer vision outside of medicine. We are also developing and deploying new AI models through our laboratory for AI in Medicine and Imaging.

Additionally, RSNA and SIIM recently co-sponsored the development of the online National Imaging Informatics Curriculum and Course for radiology residents, which includes AI training. The first session last fall attracted more than 200 residents from nearly 40 residency programs, including Stanford, and participation continues to grow with each session. The rest of imaging curriculum hasn’t changed much yet, but it will need to do so as AI products become widely available.

What should medical imaging trainees know about artificial intelligence before starting their studies and training?

I frequently hear from medical students who are attracted to radiology, but who have heard the hype and who are concerned that AI may replace the profession. On the contrary, new technologies have historically invigorated the field of radiology, but even unwarranted fears could become reality if they affect our ability to attract great people to radiology.