Even before Ding Xiaowei entered the medical industry, he knew hospitals well. He spent his childhood in them, tagging along as his grandparents treated patients. As a young boy, he saw how healthcare touched the lives of normal people, especially those who couldn’t afford the best treatment.
“A lot of my pets were from villagers who were too poor to pay for the care they received as patients,” the 28-year-old recalls. “They would give us a lot of dogs and cats as a way to thank my grandmother.”
Growing up in that environment – normally a depressing place for healthy individuals, he says – planted the seed that eventually guided his computer science career to medicine. In 2016, Ding co-founded Voxelcloud, an artificial intelligence company specializing in medical imaging, with the aim of improving access to accurate diagnosis.
“The fact that medical resources are not evenly distributed is a clear problem,” he tells Tech in Asia.
“Doctors in China sometimes see more than 100 patients a day.”
In China, for instance, the disparity between hospitals in Shanghai and Beijing and those in smaller cities can make the difference between life and death. Ding’s own grandmother experienced that when she entered a hospital in Qingdao – a city of more than nine million inhabitants – for a blood infection. It was only because Ding was able to consult two doctors in Shanghai that she survived, he says. “The doctors at the local hospital didn’t know what to do.”
Doctors in China are also extremely overworked, sometimes seeing more than 100 patients a day – with some earning less than US$500 a month. That means boosting doctors’ efficiency is crucial to managing caseloads, especially in places where hospitals are shorthanded. Some tasks, such as preventive health screenings, can also be moved from the clinic to the household, says Ding.
It’s not easy for those without professional medical knowledge to carry out these screenings – say, for nearsightedness in teens and children – yet there’s no need to visit a doctor either.
“AI is fundamentally about solving simple and repetitive problems at scale,” he adds.
Out of the lab and into the fire
In a sense, Voxelcloud is a continuation of Ding’s work as a PhD student at the University of California, Los Angeles (UCLA). There, he worked with doctors at Cedars-Sinai Medical Center to automate certain analyses of cardiac images, such as those related to coronary artery disease.
After graduating, he considered staying on and becoming a professor. “Professors and startup founders have one thing in common – no one manages you,” says Ding. But he chose the entrepreneurship route in the end because it was faster.
“Professors and startup founders have one thing in common – no one manages you.”
“Pulling together a research group as a professor is a long and difficult process,” he explained. “You have to find a faculty position, apply for national grants, and hire students. It’s definitely slower than a startup that’s raised capital.”
Indeed, less than two years after its founding, the Shanghai-based company has three products, one of which has already sealed approval from the US Food and Drug Administration (FDA) and the Conformite Europeenne – Europe’s certification for health, safety, and environmental protection standards. Called Autoplaque, the product helps doctors analyze buildups of plaque inside a patient’s heart.
Voxelcloud’s other two products, which screen CT scans and images for lung cancer and retina diseases, are currently undergoing clinical trials to obtain approval from China’s regulators. According to Ding, the company is partnering with over 100 medical centers and hospitals in China to conduct these experiments. The outcome will demonstrate how accurate Voxelcloud’s analytics system is compared to standard diagnostics by human doctors.
But the point of Voxelcloud’s diagnostic tools isn’t only to free up time spent poring over medical images, emphasizes Ding. The diabetic retinopathy exam, which checks for retina damage incurred by diabetes, is a simple diagnostic test that optometrists can conduct in five to 10 seconds. It won’t be useful for hospitals in Shanghai and Beijing, but it could fill the gap for rural villages and smaller cities.
“It can be used in places that don’t have skilled eye doctors, and given to a broad group of users for checkup,” he explains. “It becomes a kind of resource, going to places that Shanghai and Beijing doctors can’t easily access.”
In the future, the startup also plans to develop supplementary tests that parents can carry out at home, such as checking for vision disorders in children under three years old. Near and farsightedness can also be assessed through photos of the eye’s exterior, says Ding.
“These kinds of applications don’t have to be done by a doctor,” he says. “But for those who need [consultation by a doctor], they can first learn about the problems they’re having before going to the hospital.”
Artificial intelligence startups often face the challenge of obtaining enough data. In the medical industry, it’s even worse. If you’re training a natural language processing system, you can obtain data by paying freelancers to read you sentence samples. Developing AI systems for medical image analysis, however, requires real data from actual patients and doctors.
“You need professional [doctors] to help you,” says Ding. “And professionals are very busy people. That’s why this kind of data is very rare. The cost and pressure of collecting it is much higher than in other industries.”
Protecting patient privacy is also paramount. Some hospitals will not let companies take data outside of its premises, so Voxelcloud’s team has to work on-site in order to access hospital data. For its retina disease screening tool, the company trained its system on more than 15,000 images.
Despite the challenges, however, the AI industry for medical image analysis has exploded in recent years. This year alone, several AI startups have raised significant funding rounds, including SigTuple which uses artificial intelligence to analyze blood samples, and AIdoc Medical, an Israeli company that helps radiologists spot abnormalities in CT and MRI scans.
Tech giants too are getting in on the action. Earlier this year, Alibaba Cloud showed off a suite of AI healthcare applications, including tumor identification for both lung and thyroid cancer.
On the topic of competition, Ding admits that the industry is quickly becoming crowded. However, with more than US$28 million in funding – the most recent being a US$15 million “A+” round last month – from big-name investors like Sequoia Capital and Tencent, Voxelcloud is well-equipped to continue its research for the long term. At the moment, the company is not focusing on generating profit, though a software-as-a-service model could work in the future, he says.
In addition to funding, the company also has a strong R&D team in Los Angeles, where Demetri Terzopoulos, Ding’s former PhD advisor and Voxelcloud’s co-founder and chief scientist, resides. The LA office has been a crucial hiring hub for Voxelcloud as it’s been difficult to find talent in China whose experience in technology also overlaps with the medical industry, says Ding.
The third co-founder, Jianming Liang, also comes from an academic background. Liang is an associate professor of biomedical informatics and computer science at Arizona State University, as well as the company’s vice president of R&D.
Currency converted from Chinese yuan. Rate: US$1 = RMB 6.63.