Scholarlink

Echocardiography is crucial in cardiac diagnostics for assessing heart structures and functions, yet traditional interpretation methods are time-consuming and prone to observer variability. Although advancements in artificial intelligence (AI) have enhanced echocardiogram analysis, challenges persist, particularly in integrating diverse data types and maintaining temporal coherence throughout cardiac cycles. This study introduces the Multi-Modal Transformer-CNN Hybrid (MT-CNN) model, designed to synergize convolutional neural networks (CNNs) for spatial feature extraction with transformer architectures capable of capturing long-range temporal dependencies. By leveraging attention mechanisms for multi-modal feature fusion, MT-CNN enhances diagnostic accuracy and speed in detecting cardiac abnormalities. Evaluations on benchmark datasets, including CAMUS and EchoNet-Dynamic, demonstrate the model’s superior performance compared to traditional CNN and hybrid CNN-RNN models, achieving an accuracy of 94.2% and improved temporal consistency scores. These results underscore MT-CNN’s potential to revolutionize echocardiographic analysis, paving the way for more reliable and efficient clinical diagnostics in cardiovascular healthcare.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-Modal Transformer-CNN Hybrid Model for Cardiac Abnormality Detection in Echocardiograms (MT-CNN)

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Aman Mishra,
  • Shivangi Choudhary

摘要

Echocardiography is crucial in cardiac diagnostics for assessing heart structures and functions, yet traditional interpretation methods are time-consuming and prone to observer variability. Although advancements in artificial intelligence (AI) have enhanced echocardiogram analysis, challenges persist, particularly in integrating diverse data types and maintaining temporal coherence throughout cardiac cycles. This study introduces the Multi-Modal Transformer-CNN Hybrid (MT-CNN) model, designed to synergize convolutional neural networks (CNNs) for spatial feature extraction with transformer architectures capable of capturing long-range temporal dependencies. By leveraging attention mechanisms for multi-modal feature fusion, MT-CNN enhances diagnostic accuracy and speed in detecting cardiac abnormalities. Evaluations on benchmark datasets, including CAMUS and EchoNet-Dynamic, demonstrate the model’s superior performance compared to traditional CNN and hybrid CNN-RNN models, achieving an accuracy of 94.2% and improved temporal consistency scores. These results underscore MT-CNN’s potential to revolutionize echocardiographic analysis, paving the way for more reliable and efficient clinical diagnostics in cardiovascular healthcare.