Diagnostic System Automatically Analyzes TTE Images to Identify Congenital Heart Disease

Congenital heart disease (CHD) is one of the most prevalent congenital anomalies worldwide, presenting substantial health and financial challenges for affected patients. Early detection and treatment of CHD can greatly enhance the prognosis and quality of life for children. However, inexperienced sonographers often struggle to accurately identify CHD using transthoracic echocardiogram (TTE) images. Therefore, there is a pressing need for an auxiliary CHD screening system that enables inexperienced sonographers and general practitioners to conduct TTE assessments in a simple and user-friendly manner, thus enhancing the rate and reach of CHD screening.

A new CHD detection system co-developed by researchers from Anhui Medical University (Anhui, China) to identify the TTE cardiac views integrates information from various views and modalities, visualizes the high-risk region, and predicts the probability of the subject being normal, atrial septal defect (ASD), or ventricular septal defect (VSD). This was accomplished through the development of a hierarchical network structure. Initially, the model recognizes the two modalities used in TTE—2D and Doppler—and identifies the cardiac views, which include the apical four-chamber (A4C), subxiphoid long-axis view (SXLAX) of the two atria, parasternal long-axis view (PSLAX) of the left ventricle, parasternal short-axis view (PSSAX) of the aorta, and suprasternal long-axis view (SSLAX). It then processes the features for each view and each modality using the ResNet50 backbone network.

Following the basic feature embedding module, the model amalgamated the data from all five views and subsequently merged the information derived from the two modal TTEs. The final predictions for each subject were then generated by the classifier, and a visualization of the high-risk regions for each child was created using the Grad-CAM strategy. After completing the TTE exam, the auxiliary CHD diagnostic system automatically analyzed the TTE images and assessed the likelihood of the subject being normal, or having ASD or VSD. The research team demonstrated that the model effectively identified children with CHD by integrating multiple views and modalities of TTEs. The findings indicate that this model could significantly aid in broadening CHD screening and accurately distinguishing between different CHD subtypes in children.

Related Links:
Anhui Medical University