Scholarlink

With the rapid advancements in technology, the medical field has witnessed significant progress, particularly in diagnostics, where accurate preoperative planning is essential for successful treatment. Liver segmentation is critical for understanding its complex structure and localizing tumors, enabling precise clinical diagnosis and treatment planning. Image segmentation, a process of dividing images into regions with homogeneous features, has evolved from traditional handcrafted methods to deep learning-based approaches. This paper proposes a two-stage deep learning framework: an Encoder-Decoder Convolutional Neural Network (EDCNN) for liver segmentation and a UNet with Res2Net and Squeeze-and-Excitation (SE) modules for tumor segmentation. The framework effectively addresses challenges such as intensity variations and liver complexity by incorporating data augmentation, windowing, and Cumulative Distribution Function (CDF) pre-processing. Experimental results on the 3DIRCADb dataset demonstrate the proposed method’s superiority, achieving Dice scores of 0.968 for the liver and 0.855 for tumors, paving the way for improved clinical decision-making and patient care.

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

Enhanced Liver and Tumor Segmentation in Multi-modal Medical Images Using Deep Learning Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Wail M. Idress,
  • Yuqian Zhao,
  • Khalid A. Abouda,
  • Aala A. Ahmed

摘要

With the rapid advancements in technology, the medical field has witnessed significant progress, particularly in diagnostics, where accurate preoperative planning is essential for successful treatment. Liver segmentation is critical for understanding its complex structure and localizing tumors, enabling precise clinical diagnosis and treatment planning. Image segmentation, a process of dividing images into regions with homogeneous features, has evolved from traditional handcrafted methods to deep learning-based approaches. This paper proposes a two-stage deep learning framework: an Encoder-Decoder Convolutional Neural Network (EDCNN) for liver segmentation and a UNet with Res2Net and Squeeze-and-Excitation (SE) modules for tumor segmentation. The framework effectively addresses challenges such as intensity variations and liver complexity by incorporating data augmentation, windowing, and Cumulative Distribution Function (CDF) pre-processing. Experimental results on the 3DIRCADb dataset demonstrate the proposed method’s superiority, achieving Dice scores of 0.968 for the liver and 0.855 for tumors, paving the way for improved clinical decision-making and patient care.