Happy American Heart Association month, everyone! While celebrating this month at EMMA International, we bring you this blog on a medical software tool that utilizes Artificial Intelligence (AI) for providing echocardiography. This is the story of Caption Guidance, the world’s first and only FDA-granted AI-guided cardiac ultrasound software...

1. Caption Guidance – FDA’s First Authorized AI-based Cardiac
Ultrasound Software
By: Govind Yatnalkar
Happy American Heart Association month, everyone! While celebrating this month at
EMMA International, we bring you this blog on a medical software tool that utilizes Artificial
Intelligence (AI) for providing echocardiography. This is the story of Caption Guidance, the
world’s first and only FDA-granted AI-guided cardiac ultrasound software.
Built by the California-based Caption Health, the Caption Guidance tool employs AI
algorithms to provide its intended use. The major functionality of this tool is to guide healthcare
professionals in getting the best quality images or videos while performing ultrasound or
echocardiography. From a software perspective, the principal emphasis of the algorithm is to
provide a distinction between images or videos which are acceptable, and which do not have the
expected clarity.1
For users, the application provides directions while they are moving the probe on the
patient’s chest for cardiac imaging. The application instructs to shift the probe towards an
accurately computed direction if the currently generated image or video has noise or is not clear.
Also, the images and video frames with the right quality are auto-captured. But in the end, I
wondered, how is this image or video quality measured? That is where I found about DCNNs
and the ‘learning’. 1
For Caption Guidance, the AI Algorithm used for image or video frame selection is the
Deep Convolutional Neural Network (DCNN). DCNNs are the most powerful and preferred AI
algorithms when it comes to imaging applications. The learning begins by training this DCNN
with a massive data set of existing images and videos. When the user shifts the probe while
performing an ultrasound on the patients’ body, new images or videos are generated and these
are then compared with the learned data. Based on the minimal difference from image quality
comparison, current images and video frames are selected.2
This imaging software tool followed the FDA De Novo pathway, indicating the device fell
in the low-to-moderate risk category. Truly, as the selected images and videos assist healthcare
professionals to make significant decisions relating to patient care, the software’s reliability,
performance, and most important, safety must be thoroughly tested.
1 FDA (February 2020). FDA Authorizes Marketing ofFirst Cardiac Ultrasound Software That Uses Artificial Intelligence to Guide
User. Retrieved on February 9th, 2021 from https://www.fda.gov/news-events/press-announcements/fda-authorizes-marketing-
first-cardiac-ultrasound-software-uses-artificial-intelligence-guide-user.
2
Schneider,M., Bartko, P., Geller, W. et al. A machine learningalgorithmsupportsultrasound-naïve novicesin the acquisition of
diagnostic echocardiographyloopsand provides accurate estimation of LVEF. Int J Cardiovasc Imaging (2020).

2. Software validation entails testing the software design and application thoroughly for
ensuring the software reaches the safety, quality, and efficiency levels expected by the FDA.
These include activities such as documenting the software requirements, architecture,
performing risk-assessment, and in-depth code testing with inspections. Specifically, with AI-
based medical devices, the requirements, training datasets, and AI models may change
frequently. To regulate such a software that has a dynamicdesign pattern, the FDA recently
released a detailed action plan that specifies what steps they will be implementing for
specifically regulating AI-based medical devices. The action plan includes activities such as
gathering data for real-time AI performance monitoring and building the best consensus
standards for improving AI models, but the major takeaway is that this year, they will be
providing a specific guidance document that will focus on regulating AI/ML-based software as a
medical device.3
As discussed, one of the most advanced echocardiography software tools in the
healthcare industry is the Caption Guidance. It uses AI to guide novice healthcare professionals
to get quality images and videos while performing cardiacimaging or ultrasound. But
manufacturers or AI-engineers should ensure that their tool is completely validated and tested
before it is hosted in the market. This is where we can help. Do you have anAI/ML-based
software tool that needs to be validated or FDA-compliant? Our quality and software experts can
get your software tool completely validated and guide you through the FDA regulatory process to
ensure your AI/ML-based medical device is FDA compliant. Contact us at 248-987-4497 or
email us at info@emmainternational.com for more information.
3 FDA (January 2021). FDA Releases Artificial Intelligence/Machine Learning Action Plan. Retrieved on February 9th, 2021 from
https://www.fda.gov/news-events/press-announcements/fda-releases-artificial-intelligencemachine-learning-action-plan.