Stanford Researchers Release Large Pediatric Echocardiography Video Dataset for Computer Vision Research

The most important first-line diagnostic tool for heart disease in the United States is echocardiography, as it is portable, effective, and noninvasive while providing high-quality images. The most commonly used and readily available imaging modality for evaluating and treating congenital and acquired heart disease in children is pediatric echocardiography, also known as cardiac ultrasonography. A wide range of pediatric disorders, including those treated with arrhythmia, heart failure, post-surgical ventricular function, genetic abnormalities, and acquired heart disease, require evaluation of left ventricular function to monitor disease development and focus treatment.

The researchers also presented a huge new echocardiogram video dataset for machine vision research in addition to the deep learning model. The database includes patients with a wide range of sizes and ages (0-18 years) (43% are women). The 7,643 echocardiography films annotated in the EchoNet-Peds database, along with measurements, tracings, and calculations by human experts, serve as the starting point for research on heart motion and chamber sizes.

As demonstrated by EchoNet-Dynamic, machine learning has shown tremendous promise in adults, particularly in its ability to improve assessment of left ventricular function. However, the difficulty of collecting large data sets and the rarity of congenital and acquired heart disease sometimes restrict research in children. Children’s environments differ from those of adults in several ways, including a broader range of patient size, age, heart rate, and involvement, all of which are known to affect image quality. In addition to the absence of open databases that allow for cooperation, ensure reproducibility, and build infrastructure to support unbiased generalization, they have somewhat restricted machine learning in pediatric cardiology. Furthermore, it is debatable whether models that have only been trained on adult data will be suitable for young patients. By training and verifying a model with only pediatric data and adding echocardiographic images relevant to the pediatric community for assessment of left ventricular function, the researchers set out to expand the work of EchoNet-Dynamic with EchoNet-Peds. The researchers provide the EchoNet-Peds data set of 7,643 echocardiography videos, covering the range of typical echocardiography laboratory imaging conditions, with corresponding labeled measures such as ejection fraction, left ventricular volume at end of systolic and end-diastolic, and human expert tracings of the left ventricle as an aid to the investigation of machine learning methods to assess cardiac function. The researchers also demonstrate the model’s ability to classify videos using a three-dimensional convolutional neural network design. This model serves as a standard for future collaborations, comparisons, and the development of architectural designs for specific tasks. It is used to semantically segment the left ventricle, assess ejection fraction to judge human performance, and segment the left ventricle. This is the largest collection of labeled pediatric medical videos made openly available to researchers and medical specialists.

data set

• Echocardiogram movies – An average complete resting echocardiogram exam consists of several videos and images that show the heart from various perspectives using various imaging methods. The data set includes 4,424 parasternal short-axis echocardiography recordings and 3,176 apical 4-chamber echocardiography videos from patients who received imaging procedures between 2014 and 2021 as part of routine clinical care at Lucile Packard Children’s Hospital in Stanford. Each video was cropped and masked to exclude text and material outside of the scan sector. The generated images were then cubically interpolated and subsampled into uniform 112 × 112 pixel movies.
• Measurements: Each investigation is linked to clinical measurements and calculations performed by a registered sonographer and confirmed by an expert echocardiographic physician using the usual clinical process and the video itself. Left ventricular ejection fraction, used to diagnose cardiomyopathy, assess eligibility for specific chemotherapy treatments, and establish the indication for medical devices, is a key indicator of cardiac function. The “bullet method” or the “area length 5/6 method”, which derives left ventricular volumes from apical and parasternal short-axis views, is the ejection fraction method used for the pediatric data set. The ratio of left ventricular end-stroke volume (ESV) to left ventricular end-diastolic volume (EDV), which is calculated as (EDV – ESV)/EDV, is the ejection fraction, given as a percentage.

The way doctors identify and track children’s heart problems has the potential to change with the application of AI in medical imaging. EchoNet-Pediatric can ease stress on cardiologists and increase the accuracy and consistency of diagnoses by automating the image processing process. This can help ensure that children receive the best care available and can improve patient outcomes.

However, it is essential to remember that the application of AI in medicine is still in its infancy and further studies are required to identify its advantages and disadvantages. It is also crucial to ensure that AI systems like EchoNet-Pediatric are used ethically and responsibly and are subject to appropriate oversight and regulation.

challenges

• Ensuring the accuracy and reliability of the results is one of the main difficulties in employing AI in medical imaging. Deep learning models like EchoNet-Pediatric may need to be corrected, especially if they were trained on a small or skewed data set or applied outside the context in which they were developed. This means that AI systems like EchoNet-Pediatric must undergo substantial validation and testing before they can be deployed in a clinical setting.
• Another difficulty is ensuring that AI is used in medical imaging in an ethical and legal manner. For example, there are concerns about data ownership and privacy, and the potential for artificial intelligence systems to be used in ways that discriminate against specific groups or have unintended repercussions. Clear rules and regulations should be established to control the use and development of AI systems like EchoNet-Pediatric to ensure their ethical and responsible use.

Key factors:

• The first AI model trained in pediatrics to assess ventricular function is EchoNet-Peds.
• Like human specialists, EchoNet-Peds accurately calculate ejection fractions.
• Many ages and pediatric sizes can use EchoNet-Peds quickly and consistently.
• Using pediatric training data, EchoNet-Peds performed better than adult models.

EchoNet-Pediatric, which has the potential to dramatically improve the diagnosis and treatment of pediatric heart disease, is a fascinating example of how AI could be applied to medical imaging. To ensure that these systems are used responsibly and ethically, it is crucial to approach AI in medicine with caution.

review the Paper and project page. All credit for this research goes to the researchers of this project. Also, don’t forget to join our 14k+ ML SubReddit, discord channel, and electronic newsletterwhere we share the latest AI research news, exciting AI projects, and more.