[Diagnosis of congenital heart disease using Deep Learning in pediatric chest X-rays: A proof of concept].

Background: In Mexico, congenital heart diseases (CHD) are the most common birth defects. Despite their high mortality rate, many CHD are not detected early by general practitioners and pediatricians at the primary care level. Echocardiography, the diagnostic standard for CHD, is only available at tertiary care facilities with specialized equipment and personnel. In contrast, chest X-rays are an inexpensive and accessible test, and their analysis using artificial intelligence (AI) could allow for the presumptive diagnosis of CHD. Objective: Create a database of pediatric chest X-rays and develop a proof of concept that evaluates the feasibility of applying AI algorithms for the presumptive detection of CHD from these images. Material and methods: A retrospective cross-sectional study was conducted based on the analysis of pediatric chest X-ray images using a deep convolutional neural network implemented under a Residual Network (ResNet) architecture. Results: Among the radiographs included in the study, 426 (65%) corresponded to patients with CC and 230 (35%) to patients without CC. A deep learning algorithm for binary classification (Healthy/Cardiopath) applied to this set achieved a diagnostic precision of 75%. Conclusions: According to available records, this database represents the largest collection of chest X-rays from Mexican pediatric patients with CHD confirmed by clinical experts. This resource enabled the training of an AI-based model with sufficient diagnostic performance to support its potential utility as a presumptive screening tool in healthcare settings with limited access to pediatric cardiology subspecialists.