This is the first time AIMed is hosting Breakfast Briefing on the same theme for three consecutive days in three different locations. The sessions “Experience the future of AI in Radiology”, hosted in partnership with SOPHiA GENETICS, were held between 9 and 11 April in Chicago, Boston, and Toronto respectively. They were also live-streamed for the audience around the World. Their primary agendas were to explore various issues concerning artificial intelligence (AI) and how its impact on radiology as well as improving patient outcomes

AI is transforming radiological practice as machine learning and deep neural networks pledged to assist clinicians in detecting abnormalities and onset of diseases earlier and more accurately. However, as far as workflow and infrastructure are concern, integrating AI into the healthcare system can be challenging. Ensuring AI maintains clinicians’ expertise without adding additional demand and the developed AI solution will not contribute in silos as it only benefits a particular group of people, became some of the crucial topics to be addressed. 


Dr. Paul J. Chang, professor and vice-chairman of Radiology Informatics at University of Chicago gave a comprehensive yet contrarian perspective into the use of AI in Radiology. He pointed out a thought-provoking contradiction in present AI research. That is, we have rejected “one size fits all medicine” and yet we want to develop an AI algorithm that is generalizable across different patient population and even in different medical setting. 

Dr. Chang believes this whole idea of generalization is wrong. Fundamentally, it may not even be achievable. First of all, radiology terminology is very different from other medical terms. So, it may take a long time to train an AI to understand the semantics and be able to interpret the text without radiologists. 

Second, as Dr. Chang asserted radiologists may be early enthusiasts, but they are definitely not early adopters. This means that by the time an AI healthcare solution is deemed to be perfect to face the World, AI in other industries would have made their advancement. This is why, according to Dr. Chang, it is hard for healthcare professionals to find electronic health records (EHR) intuitive especially when they are armed with better and more humanized technology like social media. 


The Boston session was more open-ended and interactive, with issues such as AI liability and ethics were discussed. Sara Gerke, research fellow, medicine, AI and law at the Petrie-Flom Center for health law policy, biotechnology and bioethics, Harvard Law School, said, there are more important problems to tackle with the AI that we have in place currently. Primarily, there is an absent of a holistic ethical framework. She believes that even if autonomous robot becomes a reality one day, there should still be a human behind who is liable for any error or wrongdoing. Therefore, the focus of the ethical framework should not go beyond the legal status or market status. It should be who should be responsible for any medical malpractice. 

On the other hand, Neil Teneholtz, director of machine learning, MGH & BWH Center for Clinical Data Science said, it makes sense not for regulations not to be catching up with technology because the latter changes quickly. We probably have witness insufficient surveillance of drugs, let alone algorithms. Nevertheless, he kept his optimism as US regulation bodies are pushing AI initiatives forward and algorithms could be adequately controlled. 


Likewise, the Toronto session took place earlier today, has an interactive outlook as guest speakers took their turn to answer questions on data sharing. Generally, as Dr. Anna Golderberg, senior scientist of genetics and genome biology program at SickKids Research Institute pointed out, there are a lot of data in the ecosystem but AI researchers have never seen them. She suggested there should be investment into infrastructure for data sharing and believes that although the Canadian government is looking into it, things are moving relatively slowly. 

Dr. Errol Colak, radiologist at St. Michael’s Hospital agreed. He said radiology has many databases but radiologists do not have access to most of them. Even if there are tools available, only part of the data remained accessible. Dr. April Khademi, assistant professor of biomedical engineering at Ryerson University and principal investigator of the Image Analysis in Medicine Lab (IAMLAB) added that even downloading the data itself can be laborious. She said when she was completing her PhD, she has to manually extract case number, in order to retrieve the relevant images. 

Speakers urged for collaborations, to encourage the sharing of data, methods and tools, to prevent medical silos. 

More of the AIMed breakfast briefing will be shared at a later date. Meanwhile, please continue to keep an eye on our latest event at

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.