Ultrasounds Where You Need Them: How AI Is Improving Diagnoses at Point of Care

by Emily Bryce

In medical emergencies, a quick diagnosis based on the information at hand can be a matter of life or death.

The same can be true for non-emergencies. When limited medical equipment is available, delays in diagnosis can turn these situations into emergencies.

Ultrasound enables the accurate, efficient and non-invasive diagnosis of a host of ailments, including appendicitis, heart abnormalities and many urological and gynecological conditions.

But most emergency responders and medical professionals either aren’t trained or aren’t equipped to use the technology.

DESKi, a member of the NVIDIA Inception program, based in Bordeaux, France, is using AI to make ultrasound technologies more effective at the point of care for these personnel and their patients.

“The fact that two-thirds of the world population have no access to medical imaging technologies is a public health issue,” said Bertrand Moal, CEO at DESKi. “Ultrasound is non-invasive, affordable and can be used to diagnose ailments of multiple organs, which makes it the perfect tool to support diagnosis at the point of care, by non-specialists.”

Benefits of Ultrasound

To help those on the medical front lines make more accurate diagnoses and better informed decisions about patient care, DESKi has created DeepEcho.

This system combines deep learning algorithms, trained on NVIDIA DGX Station, and cutting-edge handheld ultrasound devices, which can be linked up to mobile phones and tablets, to deliver the expertise of cardiac health specialists in emergency situations.

Using a wealth of training data from leading cardiology units, DESKi has developed a series of neural networks that can determine whether or not the DeepEcho’s ultrasound probe is in the correct position for acquiring accurate and insightful views of the heart.

The company is also training its algorithms to automatically measure the left ventricle ejection fraction, which can help diagnose heart failure.

“By deploying ultrasound in the field with AI software, we’re helping to bring medical imaging expertise to those who need it most,” said Moal.

Protecting Patient Privacy

To train their deep learning algorithms, DESKi needs to collect high-quality data that has been reviewed and interpreted by cardiology experts.

Recently, the startup entered into a framework agreement with Bordeaux University Hospital for the development of AI projects, including DeepEcho.

Over 20,000 cardiac ultrasound examinations are performed by experienced cardiologists every year at the hospital. DESKi uses anonymized data from these examinations to train its deep learning algorithms.

To accelerate the training, DESKi turned to the power of NVIDIA DGX Station. The portability of the deskside supercomputer enabled them to build the initial framework in-house; when it was time to deploy, they transported the system to the hospital itself.

“By deploying NVIDIA’s DGX Station onsite in the hospital, we’re able to combine cutting-edge AI technology with cardiology expertise, all while ensuring that patient data is secure and never comes off premises,” noted Victor Ferrand, co-founder and CTO at DESKi.

In the future, DESKi plans to expand its tools to other specialties such as gynecology, gastroenterology and urology.

Learn more about their work with Bordeaux University Hospital in our webinar “Deep Learning for Automatic Cardiac Ultrasound Analysis.”