Pediatric Ventricle Function Assessment Study

Completed

• Conditions: RV - Right Ventricular Abnormality; LV Dysfunction; Pulmonary Hypertension; TOF

• Registration Number: NCT06739057
• Lead Sponsor: HBI Solutions Inc.

Brief Summary

This study aims to develop a deep learning-based framework for right ventricular (RV) segmentation, prediction of RV fractional area change (FAC), and identification of pediatric RV dysfunction. The AI model was designed to distinguish between normal pediatric hearts, pulmonary hypertension (PH), and Tetralogy of Fallot (TOF). To improve diagnostic accuracy, the investigators extended the analysis beyond the A4C view by integrating data from both A4C and PSAX views. Additionally, the framework was applied to predict left ventricular ejection fraction (LV EF), further showcasing its versatility and clinical utility.

Detailed Description

Heart defects are a leading cause of birth defect-associated illness and mortality worldwide, affecting approximately 1% of live births globally. Among these, about one-quarter1 present with critical heart defects requiring early intervention to improve survival and outcomes.Right ventricular (RV) dysfunction is a prevalent form of heart disease in pediatric patients, often arising from conditions such as prematurity, post-surgical effects of congenital heart disease, functioning as the systemic ventricle, and idiopathic pulmonary hypertension. These patients face significant risks, including RV failure, decreased quality of life, potential need for transplantation, and increased mortality. Accurate assessment of RV function is crucial but challenging due to the RV's complex geometry.

Unlike left ventricular (LV) dysfunction, RV dysfunction primarily results from pressure and volume overloads and has unique pathophysiological characteristics. While LV dysfunction mechanisms are well-studied, less is known about RV, and much of its clinical management is adapted from LV-focused research. The RV, with its thinner and more adaptable walls, remodels efficiently but often tolerate changes for extended periods before failure. Furthermore, RV dysfunction frequently leads to LV dysfunction due to strong interventricular interactions. These differences, along with the RV's irregular shape and orientation, complicate imaging and assessment, requiring advanced imaging techniques, often involving multiple modalities.

Precise evaluation of RV size and function is essential for diagnosing, managing, and predicting outcomes in pediatric cardiac conditions. Echocardiography, as a non-invasive and accessible tool, is the first-line imaging technique for monitoring RV function). However, traditional RV functional measures face limitations in pediatric populations due to significant variability in RV morphology. Among systolic parameters, fractional area change (FAC) has demonstrated stronger correlations with disease severity in advanced heart failure patients and a closer relationship with RV ejection fraction as measured by cardiovascular magnetic resonance (CMR), suggesting its reliability in assessing pediatric RV function. An FAC \<35% is considered abnormal by the guideline. Accurate FAC assessment can guide timely interventions, improve prognosis, and enhance long-term outcomes by enabling better monitoring of RV function over time.

Congenital Heart Disease (CHD) includes a wide range of structural abnormalities and conditions that affect the heart's development before birth, with examples such as Tetralogy of Fallot (TOF) and Pulmonary hypertension(PH). The RV plays a crucial role in TOF diagnosis because its outflow tract obstruction and hypertrophy are primary features of the condition. Pulmonary hypertension increases the afterload (pressure) on the right ventricle, leading to RV structural and functional changes. These changes are typically not seen in the LV unless there is severe biventricular involvement or secondary effects.

Echocardiography is a non-invasive, widely accessible, and essential first-line tool for routine follow-up of various pediatric cardiac conditions affecting the right ventricle (RV). However, assessing RV function in pediatric heart disease remains challenging due to significant variability in RV morphology and physiology. The apical four-chamber (A4C) view is a cornerstone for right ventricular (RV) function assessment due to its ability to evaluate RV size, shape, and systolic function comprehensively. The parasternal short-axis (PSAX) view is often used as a supplementary echocardiographic view alongside the A4C view for assessing right ventricular (RV) function. Advances in AI-driven echocardiography, particularly deep learning, hold promise for enhancing cardiac function assessment. Recent innovations, like EchoNet-Dynamic, have shown the utility of video-based deep learning algorithms for adult LV segmentation and ejection fraction estimation. Building on this progress, the investigators developed EchoNet-Peds , an AI model for pediatric echocardiography that automates LV segmentation and ejection fraction calculations. Beside left ventricle segmentation methods, other deep learning applications include automated quantification of left ventricular structure and function, as well as novel methods for estimating intraventricular hemodynamic parameters on a beat-to-beat basis . However, most deep learning studies focus on LV assessment, with comparatively fewer models dedicated to the RV, mainly addressing segmentation tasks. While some recent advancements have emerged, such as a model estimating RV ejection fraction from echocardiographic images in pulmonary arterial hypertension patients , significant gaps remain in applying AI to pediatric RV functional assessment.

This study aims to develop a deep learning-based framework for right ventricular (RV) segmentation, prediction of RV fractional area change (FAC), and identification of pediatric RV dysfunction. The AI model was designed to distinguish between normal pediatric hearts, pulmonary hypertension (PH), and Tetralogy of Fallot (TOF). To improve diagnostic accuracy, the investigators extended the analysis beyond the A4C view by integrating data from both A4C and PSAX views. Additionally, the framework was applied to predict left ventricular ejection fraction (LV EF), further showcasing its versatility and clinical utility.

Study Timeline

• Study Start: 2014-01-01
• Primary Completion: 2024-08-31
• Study Completion: 2024-08-31
• Status Verified: 2024-12
• Study First Submitted: 2024-12-12
• Study First Submitted QC: 2024-12-16
• Study First Posted: 2024-12-18
• Last Update Submitted: 2024-12-18
• Last Update Posted: 2024-12-20

Recruitment & Eligibility

• Status: COMPLETED
• Sex: All
• Target Recruitment: 3993

Inclusion Criteria

1. Patients aged 0-18 years old.
2. Patients with the following diagnoses (defined as abnormal RVs): premature infants with lung disease, congenital heart disease with systemic right ventricles, surgically repaired congenital heart disease resulting in pressure and/or volume load on the RV (tetralogy of Fallot, Double outlet right ventricle, etc.), and idiopathic pulmonary hypertension.

Exclusion Criteria

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Primary Outcome Measures

Name	Time	Method
RV dysfunction prediction	10 minutes	FAC \< 35% is considered RV abnormal. The automated extracted FAC values were then used to predict the probability of abnormal RV function.FAC estimation accuracy was evaluated using mean absolute error (MAE), root mean squared error (RMSE), and Pearson correlation coefficient (R). Diagnostic performance metrics included AUC, sensitivity, and specificity for RV function assessment
Congenital Heart Disease Detection	10 minutes	For CHD detection, classification models were developed to distinguish between normal, PH, and TOF cases by analyzing RV echocardiogram videos. Accuracy, sensitivity, and specificity were used to evaluate the classifications of PH versus normal and PH versus TOF
Left Ventricle Dysfunction	10 minutes	The performance is assessed by calculating the area under the receiver operating characteristic (ROC) curve , evaluating the model's accuracy in detecting abnormal EF values.

Trial Locations

Locations (1)

Stanford University; 🇺🇸 Palo Alto, California, United States