AI-based automated echo measurements at Juntendo University

Improving workflow of sonographers using Artificial Intelligence-based automated ECHOcardiographic measurements and the workflow of sonographers (AI-ECHO): Randomized Crossover Trial [Late Breaking Science, AHA 2024]

A research team from Juntendo University, led by Dr. Nobuyuki Kagiyama, recently presented their findings from the Randomized Crossover Trial: Artificial Intelligence-based Automated Echocardiographic Measurements and the Workflow of Sonographers (AI-ECHO) at the AHA Late-Breaking Science Session.

Dr. Nobuyuki Kagiyama, an Associate Professor at Juntendo University, is a renowned medical doctor and clinician-scientist, recognized for his commitment to advancing artificial intelligence in medicine. With a deep passion for AI research in healthcare, Dr. Kagiyama is dedicated to designing and implementing innovative AI applications that enhance patient care and transform clinical workflows.

We had the opportunity to speak with Dr. Kagiyama ahead of his presentation at AHA 2024, where he shared his experiences of implementing Us2.ai’s technology within his echocardiography lab.

What motivated you to incorporate artificial intelligence into the workflow of your echocardiography lab?

My motivation to incorporate artificial intelligence (AI) into the workflow of our echocardiography lab stems from my deep interest in AI applications within cardiovascular medicine. I recognized its potential to handle routine, repetitive tasks, such as the measurement of standard echocardiographic parameters, while addressing the growing demand for echocardiography that often leads to sonographer overwork and burnout.

Echocardiography is a complex field involving the interpretation of images and measurements to guide patient care. While these analytical tasks are intellectually stimulating, routine measurements can be monotonous and less rewarding for sonographers. By automating these repetitive tasks with AI, I hypothesized that AI could alleviate their workload and enable them to focus on the more complex, diagnostic aspects of their work that require expertise and clinical judgment.

Could you describe the process of implementing the Us2.ai software in your echo lab and which were the key stakeholders involved in the implementation process?

First, it was essential to ensure that hospital management fully understood the clinical significance and potential benefits of the Us2.ai AI system. Prior to initiating the research, we recognized that the AI could perform analyses at an expert level and significantly faster than humans, expediting the echocardiography process. By clearly communicating these advantages, I secured management’s support for integrating Us2.ai into clinical practice.

The next step was implementing the software, which required close collaboration with our hospital’s IT department. At our facility, echocardiographic images are first sent to an image server (PACS) for storage and reference in the reporting system. Since these images needed to be concurrently sent to the Us2.ai server, we established a new transmission pathway, while measurements from our echocardiography machine are sent to the existing reporting system via a separate pathway. As Us2.ai does not analyze all routinely measured parameters—such as the ascending aorta diameter—we developed a customized system to prevent overlaps between machine-measured values and those automated by Us2.ai. Although not every hospital will require this level of customization, many institutions may need similar adjustments to accommodate their unique workflows.

Successfully managing this system required close coordination with the IT department, equipment manufacturers with in-depth knowledge of the echocardiography machines, and extensive discussions with the sonographers who would use the system.

Even after the system launch, it did not operate flawlessly at first. Sonographers faced challenges with unfamiliar operations, and errors emerged as the system was used. To address these issues, we conducted a one-month trial before formally starting the study. This trial allowed sonographers to familiarize themselves with the new workflow and return to their standard performance levels. As a result, all our sonographers are now highly proficient with Us2.ai, and their performance has likely improved since the study began.

How was the AI technology (Us2.ai) introduced to the users (sonographers) in your lab, and what approach was taken for their training? Could you describe the structure of their training and how long was the training duration?

During the initial phase of implementation, we conducted a one-hour training session with M3, Inc (Japan) to teach the sonographers how to use the technology. Afterward, they tested the system on a few volunteers. While the initial feedback was positive, some stress was noted among the sonographers as they adjusted to the unfamiliar technology in a real clinical setting. With continued support and encouragement, most sonographers became comfortable with the workflow after about two weeks.

After three months, nearly all sonographers reported that the AI technology had improved the efficiency of examinations. They noted a reduction in the time spent on routine measurements, freeing up more time for complex analyses. This experience demonstrated that while AI systems offer significant potential, their transformative impact is not immediate. Achieving clinical benefits requires structured training and an adjustment period for users to adapt to the new workflow.

In summary, what would you identify as the benefits of integrating AI into your echocardiography lab workflow?

Integrating AI has not only reduced the workload for our sonographers, as demonstrated by our study, but it also appears to reduce variations that stem from individual preferences and habits among the sonographers. While accuracy checks are conducted to maintain quality, echocardiography is inherently a procedure where personal technique can influence measurements. With AI however, the software consistently generates the same measurements from identical images, thereby contributing to greater uniformity across our lab. 

What recommendations would you offer to other institutions considering the adoption of an AI tool in their practice?

One important consideration when implementing AI is recognizing that it is not a panacea. AI is not flawless; there are instances where it may be unable to perform certain analyses, and its accuracy is not absolute. The idea that AI enables complete novices to perform examinations independently is not realistic at this stage. Furthermore, integrating AI into existing workflows may not be seamless. Depending on your lab’s current processes, significant adjustments might be required, which could initially cause frustration among sonographers.

Like many clinical tools, AI comes with a learning curve. Despite early challenges, focusing on effective utilization can transform AI into a reliable partner. Our experience shows that, with time, AI can unquestionably enhance efficiency and become an invaluable asset in clinical practice.