“[DeepMetrics’] AI demonstrates the capability to control ventilators at the level of an intensivist. We are excited to continue championing their journey as they advance global patient care.”
Mayo Clinic, considered one of the top hospitals in the United States, showcased DeepMetrics during a graduation ceremony for a startup accelerator program held in Rochester, Minnesota, on March 28. DeepMetrics was the only Korean company among the innovative medical technology startups selected for acceleration by the Mayo Clinic in October 2023.
Song Hyun-oh, a professor in the Department of Computer Science at Seoul National University and an AI expert, founded DeepMetrics in February 2021. After three years of research and development, the company developed an “autonomous AI for ventilators.” This AI controls variables related to artificial respiration, such as the patient’s oxygen saturation and blood test results, enabling the ventilator to operate independently.
In intensive care units (ICUs), medical staff must manually adjust ventilator settings, such as the number of breaths per minute, the inspiratory-expiratory ratio, and respiratory pressure, while monitoring patients’ conditions. This not only increases the workload for the medical staff but also risks missing critical deteriorations in a patient’s condition. If not adjusted in time, the patient can suffer severe hypoxia, leading to brain death.
Currently active in New York, Song said, “We saw that problems in the critical care field could be solved with autonomous AI using reinforcement learning,” in a ZOOM interview with ChosunBiz. “It’s a technology that improves patient conditions through continuous interaction and real-time data analysis.”
Reinforcement learning rewards a pet dog with praise or treats for spontaneously performing a specific action, rather than continuously explaining it. Similarly, if AI randomly achieves a desired outcome and receives a reward, it learns to find solutions independently. Google DeepMind used this method to develop AlphaGo, the AI that defeated Go master Lee Se-dol. “AI interprets patterns in large datasets through machine learning and mathematical-statistical analysis,” Song explained.
DeepMetrics validated their autonomous AI for ventilators in the ICU of Seoul National University Hospital with 2,286 patients over 4,243 days, making it the largest clinical data set in the world for such research. To date, DeepMetrics is the only company that has developed an AI capable of automatically supplying oxygen and removing carbon dioxide from patients. There are no existing approvals by the U.S. Food and Drug Administration (FDA) for such technology. If DeepMetrics successfully navigates FDA approval and commercialization, it will be a world first.
“It took about three years to develop, and we are soon submitting a paper with our research findings to a top journal,” Song said. “We are also proceeding with the U.S. patent application process.” He expects the FDA approval process to take about a year and a half from the second half of this year.
The market outlook is promising. The DeepMetrics CEO said, “With the ventilator autonomous AI developed by DeepMetrics, patient safety can be enhanced, and hospitals can reduce medical costs.” By reducing complications associated with artificial ventilation, patients can recover faster, shortening hospital stays and improving bed turnover rates.
According to U.S. health authority statistics, about 620,000 patients receive respiratory therapy annually, with 211,200 cases of ventilator-associated complications. “By applying AI technology, it’s possible to save about $6,700 (around 9.15 million won) per patient per day, resulting in an annual cost savings of $4 billion for hospitals across the U.S.,” said Song.
