Scholarlink

Atrial fibrillation (AF) is the most common arrhythmia caused by irregular impulses in the atrial tissues, and it can potentially lead to stroke and impaired heart function in patients. Diagnosing this disease in the early stages is crucial to prevent its progression and improve the quality of life for patients. This study provides a comprehensive solution based on a multimodal model and a novel feature fusion framework to provide medical diagnosis, prediction, and early prevention capabilities for AF. The proposed multimodal model diagnoses possible AF symptoms using the features fused from signal and image sources extracted by neural network models. The training data includes electrocardiography and photoplethysmography data obtained from the MIMIC PERform AF dataset and the cardiac ultrasound images provided by collaborating hospitals and represent clinical medical images. The experimental results indicate that the proposed method achieved better classification performance than deep learning features alone, with accuracy, sensitivity, and specificity of 99.32%, 99.98%, and 98.61%, respectively.

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

A High Precision Symptom Prediction and Diagnosis of Atrial Fibrillation Using CNN and LSTM with Multimodal Feature Fusion Technique

  • 章节

  • 发表日期：

    2026年

查看全文
  • Wei-Hsiu Tseng,
  • Tse-Hsien Lu,
  • Chung-Lieh Hung,
  • Che-Lun Hung,
  • Peilun Dai

摘要

Atrial fibrillation (AF) is the most common arrhythmia caused by irregular impulses in the atrial tissues, and it can potentially lead to stroke and impaired heart function in patients. Diagnosing this disease in the early stages is crucial to prevent its progression and improve the quality of life for patients. This study provides a comprehensive solution based on a multimodal model and a novel feature fusion framework to provide medical diagnosis, prediction, and early prevention capabilities for AF. The proposed multimodal model diagnoses possible AF symptoms using the features fused from signal and image sources extracted by neural network models. The training data includes electrocardiography and photoplethysmography data obtained from the MIMIC PERform AF dataset and the cardiac ultrasound images provided by collaborating hospitals and represent clinical medical images. The experimental results indicate that the proposed method achieved better classification performance than deep learning features alone, with accuracy, sensitivity, and specificity of 99.32%, 99.98%, and 98.61%, respectively.