Last May, U.S. researchers reported in the journal Nature Genetics that they had demonstrated that mutations in apparently useless “junk DNA” are strongly related with autism. The study was not the first to associate the condition with DNA that does not code for genes, but it harnessed the power of machine learning to analyze the genomes of 1,790 people with autism, their parents and siblings. It’s one of countless efforts through which artificial intelligence is transforming clinical and research medicine.

Takuya Kitagawa, director of Rakuten’s Global Data Supervisory Department, introduced a session on AI and medicine at the recent New Economy Summit (NEST) in Tokyo by drawing attention to the Nature Genetics paper and other studies using AI. He then kicked off a discussion with three practitioners who use AI in their work.

Takuya Kitagawa, director of Rakuten’s Global Data Supervisory Department, moderated a session on AI and medicine at the recent New Economy Summit (NEST) in Tokyo.
Takuya Kitagawa, director of Rakuten’s Global Data Supervisory Department, moderated a session on AI and medicine at the recent New Economy Summit (NEST) in Tokyo.

Reading genomes to predict health

“How can we use AI in medical treatments?” asked Kitagawa. “One example is reading brain signals directly and translating them into audio signals, giving those who are mute the power of speech.”

Panelist Puneet Batra is associate director of machine learning at the Broad Institute of MIT and Harvard, which was founded in 2004 to continue the legacy of the Human Genome Project and examine the genetic basis of disease. He noted how the cost of genetic sequencing has dropped dramatically over the last 15 years and outlined what genetic profiling can mean for predictive medicine.

For example, Broad researchers are developing techniques such as polygenic risk scores, which are genetic calculations that can evaluate the chances of developing medical conditions.   

“Now, looking across the entire genome, looking at millions of different variants in a single person, we can predict your risk of myocardial infarction, stroke, atrial fibrillation, or even obesity or breast cancer,” said Batra.

“Now, looking across the entire genome, looking at millions of different variants in a single person, we can predict your risk of myocardial infarction, stroke, atrial fibrillation, or even obesity or breast cancer,” said Puneet Batra, associate director of machine learning at the Broad Institute of MIT and Harvard, during a presentation on what genetic profiling can mean for predictive medicine. 
Now, looking across the entire genome, looking at millions of different variants in a single person, we can predict your risk of myocardial infarction, stroke, atrial fibrillation, or even obesity or breast cancer,” said Puneet Batra, associate director of machine learning at the Broad Institute of MIT and Harvard, during his presentation on what genetic profiling can mean for predictive medicine.

Better diagnostic tools

Fellow panelist Yuki Shimahara, CEO of Tokyo-based AI venture LPIXEL, discussed how AI can improve both the accuracy and efficiency of medical diagnoses, as well as generate new criteria for diagnosing illness.

Backed by major companies such as Canon, Fujifilm and Olympus, LPIXEL provides image analysis and diagnosis tools for researchers working in medicine, pharmacology and agriculture. Its EIRL platform can be used to analyze magnetic resonance angiography images for signs of vascular deformations and potential brain aneurisms.

“Aneurisms can be overlooked,” said Shimahara. “Compared with a single doctor’s assessment, combined with AI, the accuracy improves by several percentage points. In the case of less experienced doctors, it may improve by about 10%.”

Yuki Shimahara, CEO of Tokyo-based AI venture LPIXEL, discussed how AI can improve both the accuracy and efficiency of medical diagnoses, as well as generate new criteria for diagnosing illness. 
Yuki Shimahara, CEO of Tokyo-based AI venture LPIXEL, discussed how AI can improve both the accuracy and efficiency of medical diagnoses, as well as generate new criteria for diagnosing illness. 

Panelist Seigo Hara, a former physician who is currently CEO of healthcare startup MICIN, told attendees that AI techniques such as machine learning can also be harnessed to help doctors identify conditions that sometimes don’t present with clear-cut symptoms, such as postpartum depression. Established in Tokyo in 2015, MICIN operates a telemedicine business for remote diagnosis, including the delivery of home testing kits for influenza, and an AI-based medical analysis service that tries to find links between lifestyles and disease.

Medical AI technologies are still being developed, and their most compelling use cases have yet to be imagined. Asked about potential moonshot achievements they could bring years down the road, provided regulatory, clinical and technological hurdles can be overcome, the panelists described several intriguing scenarios.

Smart devices such as cloud-connected bathroom scales, fitness bands and other internet of things (IoT) tools will play an increasingly important role in health maintenance, the panel agreed. In the next five years, Shimahara foresees the advent of an app store-like platform for medical imaging software tools, which radiologists could select according to patient needs. Looking further out, Hara predicted that advanced analytical tools will help rewrite medical textbooks.

“We could redefine diseases in 10 years with AI,” said Hara. “For example, there are two types of diabetes, but using cluster analysis, there are actually five types… That type of classification or redefining diseases can happen using AI.”

Seigo Hara, a former physician who is currently CEO of healthcare startup MICIN, told attendees that AI techniques such as machine learning can also be harnessed to help doctors identify conditions that sometimes don’t present with clear-cut symptoms, such as postpartum depression. 
Seigo Hara, a former physician who is currently CEO of healthcare startup MICIN, told attendees that AI techniques such as machine learning can also be harnessed to help doctors identify conditions that sometimes don’t present with clear-cut symptoms, such as postpartum depression.

Looking to the future

Theoretically, some diseases could be entirely eliminated. If societies are able to discuss and eventually accept the concept of genome editing, genetic conditions could be successfully treated.

“I would like to see us address all single-point mutations in the genome—50% of all mutations that lead to disease are related to single base changes in the genome,” said Batra. “If the world focused on it, I think we could address those mutations, and that would cover 40% of populations in sub-Saharan Africa that have sickle-cell anemia… A decade of focus could really get us there.”

A major challenge is making good use of the mountains of health data being generated by medical AI technologies. But the panelists expressed optimism that it can be done.

“There’s something about this data, this problem, this field today that feels like the golden ages of physics,” said Batra, a former physicist. “The challenge of our time is to figure out how to put this data together to make a difference in our understanding of humans. We’ve reached the point where there’s an intersection of data, technology, and business need. I think it’ll now happen.”