An infant with a congenital heart defect, an elderly man undergoing radiation treatment for head and neck cancer, and a young woman with infertility concerns: Each of these people could live a longer and healthier life thanks to entrepreneurial endeavors powered by big data and led by Caltech alumni. With innovative solutions that include everything from faster medical testing to artificial intelligence-assisted analysis, these start-ups are helping reshape the healthcare system for doctors and patients, one data point at a time.
An MRI heart exam can take up to 90 minutes. The diagnostic imaging is noninvasive, but the patient has to lie still, alone, inside a large, noisy machine throughout the process. It can be difficult for adults, says Shreyas Vasanawala (BS ’94), a pediatric radiologist at Stanford, and it is nearly impossible for children. Often, kids have to be anesthetized for the exam, which can be risky for anyone with a heart defect.
Faced with the difficulty of diagnosing his youngest patients, Vasanawala began experimenting with the use of compressed sensing, a method of data sampling that could capture high- quality information about a patient’s heart in significantly less time—requiring as little as 10 minutes in the MRI machine. But the process produced large, multidimensional data sets. “Radiologists were not used to looking at information like this,” Vasanawala says. “We needed new visualization techniques, too.”
That’s when fellow radiologist Albert Hsiao (BS ’00) joined the effort. Then a Stanford medical resident with a PhD in bioengineering, Hsiao had been working on software to analyze radiological data more accurately and efficiently. Building on that work, Hsiao created a software prototype for quickly analyzing 4-D flow MRI data sets (which image blood flow patterns in three spatial directions over time), so physicians can use the information to manage individual patients. This design became the foundation of Arterys, the company the duo launched in 2011. That initial prototype was the first AI-assisted, cloud-based medical imaging software to gain FDA clearance.
“Until recently, in cardiac imaging, we also used to measure the size of the heart and its function by drawing circles by hand,” Hsiao explains. “That’s really labor intensive. Our algorithms can now do much of this analysis automatically.” This technological advance saves several hours of Hsiao’s time for each cardiac MRI and allows doctors to diagnose more patients. Arterys has begun to apply this type of AI analysis to other types of diagnostic imaging. “The dream is for all of the imaging exams that we do—whether x-ray, CT, or MRI—to generate a preliminary report automatically,” Hsiao says. “Current AI applications show promise that we will get there someday soon.”
Every year, the top U.S. cancer centers collect medical information on hundreds of thousands of patients. Taken together, these often-underutilized data have the potential to offer insights that could lead to new courses of treatment and improved outcomes, says Christopher Berlind (BS ’11).
Berlind, who has a PhD in computer science, cofounded Oncora Medical in 2014 to use machine learning to harness the power of this information and help oncologists and patients make more informed decisions. The start-up is already contributing to the medical literature in the field of radiation oncology. One recent analysis examined data from more than 2,000 patients with head and neck cancer who underwent radiation treatment. Evaluating more than 700 variables, Oncora built a model that forecasts whether a new patient is likely to experience significant weight loss or require a feeding tube. Another study weighed the toxicity of radiation against the efficacy of the treatment for breast cancer under certain conditions.
Understanding who is at risk for serious side effects allows a physician to customize a treatment plan for each patient. “The doctor can come up with a proposed treatment, and we can get a pretty good prediction of whether these side effects are likely to occur,” Berlind explains. The predictions can also save money by reducing the need for emergency procedures to treat unanticipated side effects.
When the Philadelphia-based company launched, it focused on gathering, organizing, and analyzing data already collected by cancer centers. “That’s a lot of work,” Berlind admits. “That’s why no one had done it before.” Because not every institution has the infrastructure to capture high-quality data, Oncora also worked with MD Anderson Cancer Center at the University of Texas to create software that streamlines the process of recording information. Giving doctors instant access to data accelerates research insights that could lead to future breakthroughs. The company plans to expand its focus from radiation oncology to include medical and surgical oncology data.
“The more data we have, the more we can learn from it,” Berlind says. “The more we learn, the more we can improve health care.”
The oncologists at Columbia Medical Center were in a quandary: Should they advise their patients with lung cancer to get the flu vaccine? The flu can be extremely dangerous for a patient weakened by cancer, but the vaccine also can wreak havoc on the immune system. The answer lay buried in reams of patient data—until Virtualitics, a Pasadena-based data-analytics start-up, plotted it in a 3-D virtual reality landscape.
It’s the interplay between AI and 3-D visualization in virtual reality that allows us to really see the insights in the data.
“This isn’t a question you can address with a 2-D graph of the number of flu deaths among lung cancer patients,” explains Virtualitics cofounder Michael Amori (MS ’07). “You need to consider what stage the cancer is, how old the person is, how long they’ve had cancer, and more.”
When Columbia Medical Center doctors put on virtual reality goggles to view Virtualitics’ 3-D landscape, they gained a whole new perspective. They could stand amid the data, with variations depicted by color, shape, size, and texture. They could even watch the data around them change over time.
Most important, Virtualitics’s AI could direct the doctors to patterns that they otherwise might have overlooked in the vast amounts of data. “It’s the interplay between AI and 3-D visualization in virtual reality that allows us to really see the insights in the data,” Amori says.
Amori first encountered the technology he would adopt for Virtualitics at a Caltech alumni event where he met George Djorgovski, director of the Institute’s Center for Data-Driven Discovery. Djorgovski had been using virtual reality and machine learning to analyze astronomy data. Amori, who had spent most of his career in quantitative finance, realized that the technology could be a boon in that sector as well. Together with Djorgovski, then-Caltech computational scientist Ciro Donalek, and Scott Davidoff of the Jet Propulsion Laboratory, Amori launched Virtualitics in 2014 with the goal of building “a platform that allows you to understand your data even if you aren’t a data scientist.” He adds, “That’s going to be more and more important as more and more data are generated in every sector.”
Today, the company is working to solve complex data problems in fields from marketing to defense to medicine.
At Columbia Medical Center, the doctors used Virtualitics to navigate through their data and zero in on the most interesting trends. Their findings could save lives: Some lung cancer patients, they discovered, should get the flu vaccine, and others should not.
When Katie Brenner (PhD ’09) found herself struggling with fertility issues and frustrated with the frequent doctor-administered blood tests necessary to measure her hormone levels, she did what her Caltech education taught her. “I went to the literature,” says Brenner, who holds a master’s in electrical engineering and a PhD in bioengineering. There, she discovered a simpler way to monitor reproductive health: saliva. As it turns out, scientists have known for decades that the hormones found in blood are also present in saliva. Brenner decided that women should have the option of using saliva to measure their hormone levels by themselves. And bluDiagnostics was born.
“I don’t know that there’s a millennial woman left who isn’t concerned about her fertility,” Brenner says. “But there aren’t a lot of ways for a woman to get clear answers about her health. We needed a new solution.”
bluDiagnostics, based in Madison, Wisconsin, and cofounded by Brenner in 2015, created a device that allows women to chart their hormone levels with an at-home saliva test. Each day, the user spits into a lipstick-sized tube and inserts the container into a cylindrical base. The machine measures the hormones in the saliva and automatically transmits the information to an app, which allows the woman and her doctor to monitor long-term trends and short-term changes in hormone levels that affect fertility, perimenopausal symptoms, and migraine headaches, among other things.
The at-home test, which the company expects to bring to market in the coming year, is part of a broader trend toward consumer engagement with health care. “There’s a change under way in how people interact with their health care,” Brenner says. “They’re much more proactive. They’re fed up with not having answers, and they want to take control. They want to have the data.”
The data collected by bluDiagnostics will allow a woman’s physician to make diagnoses and monitor the impacts of treatments in real time. It also could contribute to more significant advancements in the medical understanding of women’s health, a traditionally understudied topic. With users’ permission, anonymized bluDiagnostics data will be used “for research to better women’s health and to illuminate new pathways for diagnosis and treatment of women’s health conditions,” Brenner says. “We want women to have better options in the future.”