Computer Vision in Healthcare: What It Can Offer Providers

AI-backed computer vision is breaking out as a way to improve certain care models.

Solving a challenge: This was the first task set out by the Mount Sinai AI Consortium, a group of scientists, physicians and researchers at New York City–based Mount Sinai Health System dedicated to developing artificial intelligence in medicine.

“We wanted to [apply AI] in the healthcare context and tackle a problem that is clinically impactful and relevant to our practices,” says Eric Karl Oermann, instructor in the department of neurosurgery at the Icahn School of Medicine and director of the AI program, dubbed AISINAI.

The challenge the group landed on was to identify markers of acute neurological illnesses, such as hemorrhages and strokes. Time matters because a patient’s “clinical condition is something that worsens, in some cases, by the minute,” says Oermann. “They’re extremely time-sensitive.”

With this in mind, the group set out to see if they could find a way to use AI and deep learning to save some of those precious minutes. The attempt was a success: By leveraging the application of computer vision in the medical field, Mount Sinai’s system can now identify a problem from a CT scan in 1.2 seconds — 150 times faster than it would takes a physician to read the image.

GET STARTED: Register for the HealthTech Insider program today.

What Is Computer Vision?

AI is beginning to have real world implementations in healthcare, especially in the burgeoning field of computer vision, which is tasked with the incredibly difficult job of training computers to replicate human sight and understanding the objects in front if it.

To do this, computer vision uses algorithms to process images with the aim of making faster and more accurate diagnoses than a physician could make.  

How Computer Vision Optimizes Medical Diagnosis

Organizations have begun tapping deep learning, like that used in computer vision, for everything from predicting heart rhythm disorders to estimating blood loss during childbirth.

Oermann, who completed a post-doctorate fellowship at Verily Life Sciences (formerly Google Life Sciences), also hopes to tap computer vision to allow doctors spend more time with their patients.

“By bringing more machines into medicine, it will let physicians focus more on patients,” says Oermann. He believes computer vision in healthcare can also help cut costs in care delivery by transferring time-consuming and tedious tasks to machines, allowing clinicians to provide better patient care, boosting patient outcomes as a result.

Digital%20Transformation_IR_1.jpg

Computer Vision in AI: Modeling a More Accurate Meter

An example of computer vision’s promise in healthcare is Orlando Health Winnie Palmer Hospital for Women & Babies, which taps computer vision via an artificial intelligence tool developed by Gauss Surgical that measures blood loss during childbirth.

“Post-partum hemorrhaging is one of the biggest causes of mortality in childbirth,” says Lorraine Parker, patient care administrator at the Orlando Health Winnie Palmer Hospital for Women & Babies. The AI technology uses pictures taken with an iPad device and analyzes images of surgical sponges and suction canisters. 

“It was really a guessing game. This technology takes that out,” she says.

Since implementing AI at the hospital, where 14,000 babies a year are delivered, doctors have learned that they often overestimate how much blood women lose during delivery. With computer vision they can understand the amount more accurately, allowing them to treat the women appropriately.

The largest challenge is implementation, Parker says, because it’s another step in the workflow, especially for C-section surgeries. Otherwise, the computer vision rollout “has gone extremely well.”

MORE FROM HEALTHTECH: Check out how to keep your data secure in the cloud.

Computer Vision in Healthcare: Seeing the Problem Faster

At Mount Sinai, to develop the AI necessary to detect acute neurological illnesses — what Oermann calls a “weakly supervised learning approach” — the organization used 37,236 head CT scans from across Mount Sinai Health System to train a deep neural network to determine if an image showed an acute neurological illness. But the provider didn’t just let the AI run wild; it also tested it in a blind, randomized controlled trial in a simulated clinical environment.

To achieve the necessary infrastructure, Oermann notes that Mount Sinai has invested in Nvidia’s graphics processing units for AISINAI. Nvidia has also partnered with the Scripps Research Translational Institute to establish a center of excellence for AI in genomics and digital sensors to develop best practices, tools and biomedical research infrastructure.

The Future of Computer Vision in Artificial Intelligence

So, what’s next for the technology that’s already showing signs of promise in healthcare? In a study published in the journal PLOS last year, Oermann and his team found that deep-learning models could not be just picked up from one healthcare system and plopped into another. The study found that when AI from one healthcare system was tested on a model to detect pneumonia from chest X-rays at another institution, it was less effective.

This plug-and-play AI is the next step in research for computer vision. 

Is it possible to train a model that works at one place to also work somewhere else?” Oermann asks. “How do we make sure that medical AIs do what they’re supposed to do all the time?”

cosmin4000/Getty Images
Jan 30 2019

Sponsors