Advancing real-time echocardiographic diagnosis with a hybrid deep learning model

This research focuses on developing a novel hybrid deep learning architecture designed for real-time analysis of ultrasound heart images. The object of the study is the diagnostic accuracy and efficiency in detecting heart pathologies such as atrial septal defect (ASD) and aortic stenosis (AS) from ultrasound data. The problem is the insufficient accuracy and generalizability of existing models in real-time cardiac image analysis, which limits their practical clinical application. To solve this, the convolutional neural networks (CNNs), combining local feature extraction was integrated with global contextual understanding of cardiac structures. Additionally, a YOLOv7 for precise segmentation and detection was utilized. The results demonstrate that the hybrid model achieves an overall diagnostic accuracy of 92 % for ASD detection and 90 % for AS detection, representing a 7 % improvement over the standard YOLOv7 model. These improvements are attributed to the hybrid architecture's ability to simultaneously capture fine-grained anatomical details and broader structural relationships, enhancing the detection of subtle cardiac anomalies. The findings suggest that combination of CNNs enhances pattern recognition and contextual analysis, leading to better detection of cardiac anomalies. The key features contributing to solving the problem include the hybrid architecture's ability to capture detailed local features and broader structural context simultaneously. In practical terms, the model can be applied in clinical settings that require real-time cardiac assessment using standard medical imaging equipment. Its computational efficiency and high accuracy make it suitable even in resource-constrained environments, reducing analysis time for clinicians, supporting personalized treatment plans, and potentially improving patient outcomes in cardiology