Scholarlink

This research presents a structured approach to the accurate segmentation of Cardiac MRI images, crucial for precise diagnosis, particularly in delineating the left ventricle. Utilizing various Encoder-Decoder architectures, including Spatial Attention Encoder-Decoder Architecture, Encoder-Decoder Architecture, and U-Net, we systematically developed and refined models through architectural adjustments, dataset augmentation, and hyperparameter tuning, using the curated Sunny Brook Left Ventricle dataset. Ethical considerations guided the meticulous dataset curation process. Preliminary results showed promising improvements, with a Dice coefficient of 0.4307 and a Jaccard coefficient of 0.2834. By implementing data augmentation, preprocessing images and masks, and incorporating a spatial attention mechanism into the Modified Spatial Attention U-Net architecture, we achieved significant improvements, attaining a Dice coefficient of 0.8036 and a Jaccard coefficient of 0.70. Our model holds potential for even higher performance, leveraging GPU computation.

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

Quantification and Classification of Cardiac MRI Data Using Deep Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Anupama Bhan,
  • Dhruv Raj Iyer,
  • Faizan Nabi,
  • Abraham Richard,
  • Abdullah Zuber Patel,
  • Deepa Parasar

摘要

This research presents a structured approach to the accurate segmentation of Cardiac MRI images, crucial for precise diagnosis, particularly in delineating the left ventricle. Utilizing various Encoder-Decoder architectures, including Spatial Attention Encoder-Decoder Architecture, Encoder-Decoder Architecture, and U-Net, we systematically developed and refined models through architectural adjustments, dataset augmentation, and hyperparameter tuning, using the curated Sunny Brook Left Ventricle dataset. Ethical considerations guided the meticulous dataset curation process. Preliminary results showed promising improvements, with a Dice coefficient of 0.4307 and a Jaccard coefficient of 0.2834. By implementing data augmentation, preprocessing images and masks, and incorporating a spatial attention mechanism into the Modified Spatial Attention U-Net architecture, we achieved significant improvements, attaining a Dice coefficient of 0.8036 and a Jaccard coefficient of 0.70. Our model holds potential for even higher performance, leveraging GPU computation.