The 105th Radiological Society of North America (RSNA) Scientific Assembly and Annual Meeting will take place between 1 and 6 December at McCormick Place in Chicago. In support of the event, AIMed has invited Dr. Eldad Elnekave, US and Israeli Board Certified Radiologist and Chief Medical Officer of Zebra Medical Vision and Dr. Rizwan Malik, Clinical Lead for Radiology at UK’s Royal Bolton Hospital, to highlight areas that delegates can look out for while at the meeting, as well as to discuss the general impact of artificial intelligence (AI) on radiology, over a near 20-minute long podcastthat debut on 26 November. 

Question 1: How do you integrate? 

The theme for RSNA 2019is “See Possibilities Together” and again, AI will be a dominating topic. With that, Dr. Malik asked Dr. Elnekave, as there are more and more companies and startups are coming to the annual meeting with new technologies and AI-driven solutions, it may be challenging for attendees to distinguish between them; he wondered if there are useful ways to interact with these participants in a critical manner? Dr. Elnekave suggested asking these tool providers, “how do you integrate?”

He explained many companies enter the realm of radiology with only one solution or service; be it identification of a lung nodule or quantification of ejection fraction on an echography. Most of them are unique and novel, they make people curious and excited about what they can achieve with AI. However, not many of them will put a thought on whether these fanciful tools will fit into the workflow of a radiologist or the PACS (picture archiving and communication system) that they are using at the moment. 

Dr. Malik agreed. He thought some of these innovations “sound great and look flashy” but when he tried to dig a little deeper and reflect on how they may help him in his clinical practice, he reasoned that they may actually increase his present workload. Potentially because he will have to “babysit” the AI algorithm on top of his work. 

Question 2: Does your solution add value to radiology/radiologist/healthcare? 

Dr. Elnekave said he would envision three purposes that AI has on medical imaging. First of all, AI could assist radiologists to look for things that they might have missed or found them earlier. Things like bleeding in the brain, the sooner they are being single out, the quicker radiologists can look into them. Second, AI could shoulder some of the mundane tasks. 

“I love it if I can click at something like volume, and an AI algorithm tells me how has it changed over the past five years… This is part of our job which can be streamlined. The other component is to characterize these images for me. So that I can derive at a diagnosis faster,” Dr. Elnekave explained. Third, AI could surface a disease pattern at the population level. For example, looking at every scan and quantify individual risks of getting diabetes or cardiovascular disease and so on. As such, Dr. Elnekave believes it will be worthwhile to ask if one’s AI-driven solution is going to add value in any of the above area. 

Question 3: What is your solution’s predictive value in a normal distribution? 

Dr. Malik pointed out whilst most companies focused on advertising their solutions, not many of them have received regulatory approval and sometimes, there are misconceptions with regards to the amount of work required to get a product approved. 

Dr. Elnekave added, presently, rules and regulations around AI are still not concrete as compared to other industries. At times, the way companies report their product statistics can also be confusing. “Some of them will say, our algorithm is 95% accurate and 90% specific. At a glance, you may think this looks great but when you look at the prevalence in radiographs. Those percentages may translate into 20-30 false-positive cases for every true positive case”. 

So, being able to ask a question like “what is your solution’s predictive value in a normal distribution” will provide delegates a better basis to evaluate an AI solution. On the other hand, companies also need to bear in mind that there is always a risk of over-failing when it comes getting a market clearance for a product.  

“It gets to a point that you will need to be able to develop a robust solution, which takes in studies of different parameters and still give reliable results. Aware that this is just the beginning because as you bring your product into the real world, you will meet individuals with different appetites for sensitivity and specificity and your product will need to cater to that,” Dr. Elnekave said. 

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

Hazel Tang

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