Objective Echocardiography with Superior Reproducibility Based on Artificial Intelligence Technology

Kenya Kusunose MD, PhD

Aplio i-series / Prism Edition provides powerful imaging with a whole spectrum of flexible, AI-assisted productivity functions powered by Altivity, Canon Medical's new AI brand. The smart, AI-driven algorithms allow clinicians to create simple and streamlined workflows to deliver fast, accurate results, and a more personalised treatment of the patients.

“The centre is set up to ensure smooth and efficient operation, with cardiologists on duty at all times to handle the diverse demands of cardiovascular examinations,” explained Dr. Kusunose. “We also focus on providing training for resident physicians and new operators under the close supervision of senior physicians.”

In cardiac ultrasound (echocardiography), M-mode, B-mode, Doppler, and other imaging techniques are used to examine the heart and assess both its structure (morphological characteristics, cardiac volumes, etc.) and function (wall motion, valve function, hemodynamics, etc.). In recent years, there have been many technological advances in diagnostic ultrasound systems, and many new scanning techniques and applications have been developed, such as 3D cardiac echo and strain analysis. The number of measurements that need to be performed in routine clinical practice is, therefore, rapidly increasing. Dr. Kusunose notes that a standard echocardiography report today includes twice as many items as a report from 15 years ago. The increase in the number of examination items is a testament to the excellent reliability of echocardiography, but in order to perform such specialized and complex examinations with high accuracy and efficiency, high-level automated diagnostic support technology is required. It is important to develop new functions to improve examination workflow, such as image quality optimization and examination support functions based on the latest advances in AI technologies, such as Deep Learning.

In addition, with the Auto Plane Detection, when strain analysis is performed in the software for myocardial wall motion analysis, 2D Wall Motion Tracking (2D WMT), the left ventricle (LV), left atrium (LA), right ventricle (RV), and right atrium (RA) can be detected by a single button click. With the Automatic initial contour trace, the contours are automatically traced together with the automatically detected planes, reducing the time required for analysis.
“Reliable 2D WMT ensures the stable strain measurement thanks to the system's high detection accuracy,” said Dr. Kusunose. “To maximise efficiency during analysis, the number of clicks to adjust the measurement points made by the operator should be minimised, but this means that the system must provide a sufficiently high level of accuracy.”
Strain analysis is attracting a great deal of interest today because it can be used to monitor changes in cardiac function in patients with cancer who are receiving treatment, such as chemotherapy. Although strain analysis results have been added as new measurement items in routine echocardiography reports, the Hospital has found that examinations can still be completed in the same amount of time as before thanks to these automatic measurement functions.

“AI can be expected to improve image quality and measurement reproducibility, thus improving workflow by reducing the time required for examination and creating new clinical value through automated diagnostics,” said Dr. Kusunose.

Although technological advances in diagnostic ultrasound systems have led to remarkable improvements in image quality, allowing examination and analysis to be performed based on more accurate images, Dr. Kusunose has always felt the need for new analysis technologies that are able to provide higher levels of objectivity and reproducibility. The application of AI technologies such as Deep Learning to medical image analysis continues to advance in the field of radiology.

“I thought that using AI in ultrasound imaging would improve image quality and enable objective and highly reproducible diagnosis employing new indices based on quantitative analysis and other approaches,” he said.

Dr. Kusunose identifies four phases in the process of implementing AI in automatic diagnosis. These are:
1) image quality evaluation, 2) plane/segment classification, 3) measurement, and 4) abnormality detection.

Tokushima University Hospital has developed a plane classification model (plane/segment classification)¹, a left ventricular ejection fraction (LVEF) estimation model (measurement)², and a regional wall motion abnormality detection model (abnormality detection)³ based on a convolutional neural network (CNN). Dr. Kusunose also notes that in order to apply AI to echocardiography, accurate classification of the planes and segments is essential for reliably recognising complex cardiovascular structures. He has, therefore, used such images to develop models for measuring LVEF and detecting wall motion abnormalities. These models have shown excellent agreement with the results of evaluation performed by experts, clearly demonstrating the great potential of AI in supporting cardiac examinations. In addition, Dr. Kusunose is currently working to develop a model that allows the AI system to learn the ‘visual EF,’ as determined by experts and reflect the results in automatic analysis.

“With regard to the future directions of AI development in the field of cardiac ultrasound, our immediate challenge is to further improve reproducibility,” Dr. Kusunose remarked. "It must be possible to obtain highly accurate data in the same easy way, no matter who is performing the examination. Although only a few guidelines, such as those in the field of cardio-oncology, include items related to 'reproducibility' in ultrasound examinations, high reproducibility is an essential factor in diagnosis and treatment planning based on ultrasound examination results. If AI can be used to obtain stable and reproducible images and measurement results, we can expect ultrasound systems to make even greater contributions to therapeutic decision making."

On the other hand, the main focus of Dr. Kusunose's AI research is the differential diagnosis of pathological conditions in the future.

"I believe that automatic measurement and diagnosis will be realised by manufacturers of medical equipment, so we, on the clinical side, will focus on the development of AI that can detect new pathologies and diseases," he added.

As the key institution in the imaging-related database project of the Japan Agency for Medical Research and Development (AMED), Tokushima University Hospital is taking the lead in the creation of a cardiovascular ultrasound imaging database in Japan. Future research and development to maximise reproducibility in echocardiography and further improve workflow by the introduction of automation is eagerly anticipated.//

(Interview conducted on June 27, 2022.)