The U.S. Food and Drug Administration (FDA) has cleared a new AI-powered pregnancy ultrasound tool.
The FDA gave clearance to Butterfly Network‘s AI-powered ultrasound tool designed to estimate gestational age without the need of specialist interpretation. It uses artificial intelligence to interpret videos captured using a “blind sweep” movement.
In an official statement, the company shared plans to incorporate the AI tool into its portable ultrasound device and app. The tool is only commercially available to licensed healthcare professionals. Originally deployed in clinical settings in emerging African markets, like Uganda and Malawi, the FDA clearance allows the company to expand into a larger market. Butterfly Network plans to move into the U.S. while simultaneously deploying it across other Sub-Saharan African countries.
“Butterfly is proud to be the first to bring this type of technology to mothers globally,” said Dr. Sachita Shah, the vice president of global health.
The Technology Behind The Tool
The underlying technology was invented by University of North Carolina professors Jeffrey Stringer, Ben Pokaprakarn, and Juan Prieto.
The university licensed the technology to Butterfly Network, where they further developed the deep learning model. The Gates Foundation helped to support the company’s development of the tool with a $5 million grant.
How The AI Ultrasound Device Works
According the Butterfly Network, the gestational age tool provides a reliable estimate in less than two minutes through a three-step process.
A user enters the patient’s fundal height, applies gel to the abdomen, and performs guided sweeping motions. The AI tool automatically identifies the fetus within the ultrasound video and calculates its age.
In this method, a technician moves the probe across the abdomen in a pre-set pattern. They don’t need to identify landmarks or have a spatial expertise of fetal organs on the screen.
Potential Ethical Gray Area With The Ultrasound Tool
Butterfly Network shared they trained their algorithm on more than 21 million ultrasound images. They developed the dataset through research partnerships with the University of North Carolina at Chapel Hill. The model was also trained on clinical data from patients in Zambia and other Sub-Saharan African clinics.
Some critics pointed to that being an ethical gray area since the algorithm relies on medical data collected from volunteers in Zambia, Malawi, and Uganda to power a commercial product for a U.S.-based company.
Gestational Aging May No Longer Be A Specialized Practice
According to experts, ultrasound screening automation could potentially shift diagnostics from specialized clinics to general practitioners and non-specialists.
In the clinical study, the AI tool provided statistically equivalent gestational dating compared to a high-end cart-based systems for the 16-to-37-week window. One doctor in the study described device as an “accurate, fast, and simple way for lower skilled healthcare workers to get expecting women the right care plans for the stage of their pregnancy journey.”
Gestational aging has primarily been a referral only practice. Pregnant women are routinely referred to specialist clinics, obstetrician-gynecologists and American Registry for Diagnostic Medical Sonography (ARDMS). Now, newly cleared automated tools, such as the Butterfly iQ3, disrupt this high-cost and training barrier.
These handheld devices retail between $4,000 and $8,000, allowing non-specialist clinicians to perform the screening. To contrast, ultrasound machines can cost upwards of $100,000 and require raining to master.
In addition to Butterfly Network’s new tool, several other automated ultrasound tools have received FDA clearance this year. Another tool, called Delivery Date AI, analyzes ultrasound image to predict delivery date.
While AI can now automate measurement and biometry, there are still some things it cannot do, placing an emphasis on the need for medical professionals with proper training.
The Society of Diagnostic Medial Sonography voiced support for responsible AI integrations, while cautioning against replacing trained specialists. The American Registry for Diagnostic Medical Sonography even released an article last year recommending the use of AI as a clinical-support tool not replacement for clinical expertise.
“AI won’t replace sonographers, but it will certainly be disruptive. Fortunately, imaging professionals are in a unique position to help drive the technology’s advancements as it requires skilled sonographers to facilitate the right decision and assessment paths,” the article read. “Because of this, AI will be valuable as a teaching tool, as well as for testing and validation. The bottom line: it won’t replace clinical expertise but will be a clinical support tool.”
