Leveraging Artificial Intelligence to address pitfalls of modern Echocardiography

By Tony Gallagher, Director of Noninvasive Services, Baptist Health Lexington

As healthcare steamrolls down the path of artificial Intelligence, we all need to remain acutely aware of the effect it will have on our various customers: patients, providers, and healthcare workers. Any progress in Artificial Intelligence (AI) should be viewed through the lens of how these groups might be affected. Questions like the following should be considered:  “Will this enhance or improve the life of a patient?” “Will this advancement aid or hinder a provider arriving at a diagnosis?” and” Will the innovative technology simplify or complicate the job of the person performing the work?”

In the last decade, the field of ultrasonography has seen notable AI initiatives. This article will specifically refer to cardiac ultrasound. AI growth in the industry includes automated measurement of Doppler signals for vascular ultrasound that eventually translated to cardiac ultrasound and the creation of 3D images derived from sonographer-traced 2-D images. Today, AI applications can perform standard image recognition, automatic measurements, and even disease state recognition, such as predicting heart failure or calculating valvular stenosis. AI-based ultrasound machines are currently being endowed with the same advanced features and capabilities as some of the more robust cardiac PACS systems. 

Examples of new systems that have the potential to reduce the work of the healthcare worker and, at the same time, improve or maintain accuracy are: 

1. Siemens is releasing a new cardiac ultrasound machine (the AcusonOrigin), that has image recognition capabilities and can automatically configure the Color Flow Doppler box appropriately for each cardiac valve. Then, it can also automatically place the spectral Doppler sample and measure Doppler signals in real time, saving the sonographer the time it takes to freeze, activate the measurement system, and perform multiple measurements.
2. Us2.AI is an AI application that reportedly has the ability to automatically measure over 90% of the measurements performed in an echocardiogram. If this application proves its validity and reliability, it could reduce the time-to-perform an echocardiogram by as much as 10 to 15 minutes.

Leveraging an AI application has the potential to reduce that study time by 15+/- minutes. If a sonographer performs eight studies/shift, there is potential to conservatively recoup eighty minutes/shift.

Ongoing research in the field of echocardiography, advancing technologies and increasing AI features have continued to increase the amount of data derived from an echocardiogram. Vendors are entering the field of advanced AI from different angles, each with unique features that promise to help facilities derive and compile an increasing amount of data. The question becomes, which AI features seem more beneficial and how can you leverage them and the increased amount of data to improve the work performed by the sonographer, increase the accuracy of the physician’s diagnosis, and ultimately improve the life of the patient?

As an operations director, my perspective on how we should leverage AI is from the vantage point of quality, workflow, and capacity. To address these, we must discuss pitfalls that directly affect a cardiac ultrasound.

• Pitfall one, the environment: The performance of a Cardiac ultrasound is optimal when the sonographer can control the lighting, patient positioning as well as their own positioning to maintain proper ergonomics. However, most hospital-based echocardiograms are performed at a patient’s bedside, where there is minimal environmental control. Practically speaking, you do not need AI to address this.
• Pitfall two, Intra-reader variability: Much of reading an echocardiogram has historically been qualitative, not quantitative. However, qualitative interpretation can lead to results variations among readers and increase the margin of error. As an example, echocardiography conferences usually have a meeting called ‘Read-With the Experts.’  At these meetings, you will see that the qualitative interpretation of an echocardiogram can vary enough among physicians to change a diagnosis.
• Pitfall three, intra-sonographer variability:  The overall quality of a cardiac ultrasound is very sonographer dependent. The sonographer performing the exam must make fine adjustments to optimize the image quality of each still image and cineloop that they capture and then perform on-screen measurements on those optimized images. There are multiple problems with this method of measuring, training, skill level and each individual’s vision. Each of the adjustments can cause variances in the measurements performed. Every sonographer strives to do the best of their ability to ensure the highest image quality and most accurate measurements that they can for their patients. However, the variants in vision and training leave a scenario where there is no perfect or true standard.

Leveraging AI

There are currently around 350 measurements that are either performed or derived during an echocardiogram. Theoretically, a beneficial AI would be one that could accurately and consistently reproduce these measurements with at least the same accuracy as an experienced cardiac sonographer. Thus, mitigating Pitfall three.

Over the past year, Baptist Health in Kentucky has been investigating the opportunity to leverage an AI application to mitigate intra-sonographer variability and standardize our cardiac measurement and quantification processes. The proper use of AI will potentially minimize variation in the cardiac measurements performed, and theoretically reduce the time it takes to perform a study. The measurement portion of an echocardiogram consumes roughly 30% of the duration of the study.  If we can leverage an Artificial Intelligence application with the ability to perform all, or most, of the routine measurements performed in an echocardiogram. This could create a positive cascade that would help us maintain quality (due to the AI standardization), reduce the time-to-perform a study (as the measurement portion will be significantly reduced), and increase capacity.

From a healthcare operations viewpoint, an average echocardiogram takes 45-60 minutes to complete, including measurements. Leveraging an AI application has the potential to reduce that study time by 15+/- minutes. If a sonographer performs eight studies/shift, there is potential to conservatively recoup eighty minutes/shift. That time can be utilized to perform an extra echocardiogram per shift without a reduction in quality. Increased capacity without increasing staff equals increased revenue. The larger the number of cardiac sonographers in a facility, the greater the potential for operational efficiencies, which will likely result in increased revenue.

Leveraging the right AI application has 1) The potential to increase the diagnostic quality of echocardiograms, which is a win for patients and providers, 2) Increase capacity without additional staff, and 3) Increase revenue. As Micheal Scott would say, that’s a win, win, win.