According to Dr. Lawrence Tanenbaum, vice-president and director of advanced imaging at Radnet, in one of the panel sessions of AIMed North America 2018, deep learning research is on the rise and most of the work are going into diagnostic imaging. Its application mainly varies between the screenings of breast, lungs and chest. In spite of being one of the most matured fields for artificial intelligence (AI) in medicine, diagnostic imaging adoption is still at its dawn. Roughly half of the practicing radiologists are into it or thinking of using it. 

Macro vs Micro adoption

The myth: radiologists are afraid to lose their jobs to AI in 10-15 years, is not the reason which deters the adoption. Often, radiologists are unsure of the scale of adoption. Whether it should be a macro one which involves PACS (picture archiving and communication system) or one that relies on individual device and application. 

Besides, the payment model for diagnostic imaging is not in place, so if radiologists will like to adopt it, should patients continue to pay them for their expertise or should they pay for the new technology? Dr. Tanenbaum continued, somewhat 10% of the insurance company may reject reimbursement as a result, while 70% may question the expenditure at the moment. 

Coupled with other factors such as rules and regulations, interoperability between infrastructure of different medical establishments and patients’ acceptance. The environment does not favour taking up a new technology although it is beneficial in the long run. 

Another case of workflow integration 

Jack Hidary, chairman of the Hidary foundation said, diagnostic imaging comes with many advantages even though it is extremely complex. One good thing about radiology is most data are digitized and readily available to work on. Unlike pathology, which is much slower because 99% of the past slides have never been digitized. That’s why not only Google but even Facebook is trying to tap on radiology. 

However, there remains a gap in what physicians are looking for and what diagnostic imaging can do. Jack Po, product manager of Google added, most physicians are interested in demographics like age and gender from ancient images. But diagnostic imaging highlights the risks of how likely a patient is down with certain medical condition. 

The robust data in radiology is a strength but the inability to turn them into something which radiologists find it worthwhile undermines the true potential of AI-powered medical imaging. Kevin Lyman, chief executive officer of Enlitic, commented, “radiology data is highly nuanced and unintended bias is hard to avoid without proper education. Yet, most of the time is spent building tools that enable us to build clinical AI rather than actually building clinical AI.”

Daniel Kraft, chair for medicine & neuroscience of Singularity University and founder & chair of Exponential Medicine added, “A lot of tech are moving very quickly… A lot of data are being mentioned but the challenge is how to make them talk to one another so that we not only have the data but medical utility, which can be translated to clinicians, patients and workflow.”

Author Bio
synthetic gene empathy chinese artificial intelligence data medicine healthcare ai

Hazel Tang

A science writer with data background and an interest in current affair, culture and arts; a no-med from an (almost) all-med family. Follow on Twitter.