Scholarlink

<p>Tetralogy of Fallot (TOF) is a complex congenital heart defect requiring precise diagnosis. Echocardiography is essential for detection, but a global shortage of pediatric cardiologists can hinder timely diagnosis. Our study aims to develop an Artificial intelligence (AI) model using deep learning to automate TOF diagnosis from echocardiograms, serving as a screening tool in regions lacking specialists. We trained an AI model with echocardiograms from 174 pediatric patients at Seoul Asan Medical Center (2018–2023). Utilizing Detectron2 and Mask R-CNN, we processed, labeled, and trained the model to detect four characteristic defects of TOF in images and videos. Model performance was assessed via PR curves, accuracy, and F1 scores. The developed model achieved an AUC of approximately 1.00 and F1 score of 96.8%, demonstrating high sensitivity and precision. It reliably distinguished TOF with over 97% accuracy across multiple videos, showing promise for clinical application. Deep learning-based automated feature level detection of TOF offers a promising way to improve diagnosis, increase accessibility, and reduce the workload of pediatric cardiologists globally. Future research should focus on expanding datasets and refining models to apply to other congenital heart disease, enhancing patient care overall.</p>

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

Real-time deep learning interpretation of echocardiographic video for automated detection of anatomical features associated with tetralogy of fallot in pediatric patients : a feasibility study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月5日

查看全文
  • Mi Jin Kim,
  • Jeong Jin Yu,
  • Seulgi Cha,
  • Jae Suk Baek,
  • Dongha Yang,
  • Yeon Jin Jang

摘要

Tetralogy of Fallot (TOF) is a complex congenital heart defect requiring precise diagnosis. Echocardiography is essential for detection, but a global shortage of pediatric cardiologists can hinder timely diagnosis. Our study aims to develop an Artificial intelligence (AI) model using deep learning to automate TOF diagnosis from echocardiograms, serving as a screening tool in regions lacking specialists. We trained an AI model with echocardiograms from 174 pediatric patients at Seoul Asan Medical Center (2018–2023). Utilizing Detectron2 and Mask R-CNN, we processed, labeled, and trained the model to detect four characteristic defects of TOF in images and videos. Model performance was assessed via PR curves, accuracy, and F1 scores. The developed model achieved an AUC of approximately 1.00 and F1 score of 96.8%, demonstrating high sensitivity and precision. It reliably distinguished TOF with over 97% accuracy across multiple videos, showing promise for clinical application. Deep learning-based automated feature level detection of TOF offers a promising way to improve diagnosis, increase accessibility, and reduce the workload of pediatric cardiologists globally. Future research should focus on expanding datasets and refining models to apply to other congenital heart disease, enhancing patient care overall.