Welcome to AI’s Everything Era

“We are on the brink of a seismic shift, one that will see AI evolve from a unimodal ‘scan era’ into a multimodal tool capable of doing nearly everything—from making accurate diagnoses to providing personalized medical forecasting, preventing age-related diseases and monitoring patients remotely,” said Eric Topol, MD, a cardiologist and executive vice-president of the Scripps Research Translational Institute.

This by no means is meant to minimize the impact AI has already had on health care.

“Unimodal AI, which analyses just one data type, has been shown to significantly improve the accuracy with which physicians can interpret medical data, such as scans and pathology samples,” explained Dr. Topol, who made his remarks during a Tuesday plenary session. 

But Dr. Topol said that this is just scratching the surface of AI's capabilities.

“When we stop at the scan we leave so much data on the table,” he said.

That data includes a patient’s electronic health record, laboratory tests, genome, gut microbiome, proteome, epigenome, immunome, social determinants of health, environmental exposures—and more.

“You name it, we probably have the data on it,” Dr. Topol added.

With multimodal AI, one can put all this data together into a single model, creating a high-resolution human being and opening the door to a new era of medicine.

“We will have the ability to do things we've never done before, delivering individualized medicine that spans the patient’s entire life,” Dr. Topol noted. 

For instance, AI will reboot how we screen for cancer. “Instead of screening based on age, which isn’t a reliable indicator, AI will allow us to partition a patient’s risk and screen accordingly,” he said.

To illustrate, Dr. Topol pointed to a recent study where, by interpreting a mammogram, AI was able to predict breast cancer four to six years before it manifested. Another study used a deep learning algorithm to predict risk for pancreatic cancer based only on a patient’s electronic health record.

“AI essentially gives us the power to predict the future, to know a patient’s risk for a disease and to implement preventative measures well before they actually get the disease,” Dr. Topol said.

Although the potential of multimodal AI is clear, its actual use remains limited.

According to Dr. Topol, this is due in part to a lack of compelling evidence. For example, several recent studies looked at how AI can improve both diagnostic accuracy and patient communication. In these studies, not only did AI outperform physicians on both fronts, AI used alone also performed better than a physician using AI tools.

While outcomes like these make for good headlines, Dr. Topol notes that they are contrived and not based on the real world of medicine. “If AI is to make the shift from a tool primarily associated with radiology to being a genuine part of medicine, physicians and patients need to trust it—and that starts with transparency,” he said.

This is where radiology comes in. “Trust is built from experience, and medicine will rely on radiology’s expertise as we learn to leverage the power of AI,” Dr. Topol concluded. “Together, we can put this extraordinary opportunity into practice and use AI to launch a new era for medicine.”

Access the presentation, “AI's Transformation of Medicine,” (M4-PL02) on demand at RSNA.org/MeetingCentral.