Scholarlink

X-ray Computed Tomography (XCT) is widely used in industrial inspection for detailed 3D visualization of multi-material components. However, segmenting XCT data remains challenging due to material heterogeneity, noise, and artifacts. This study presents a deep learning pipeline based on the U-Net architecture to automate segmentation tasks in multi-energy XCT scans. Applied to a complex test object, the model achieved high performance, with high accuracy and Dice scores across 17 distinct object classes—a level of granularity rarely addressed in existing studies. This work establishes a reproducible framework for XCT analysis and lays the foundation for more advanced architectures such as ResNet-based U-Nets and future 3D segmentation using models like 3D U-Net or V-Net.

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

Deep Learning Approaches for Automatic Segmentation of 3D X-Ray Computed Tomography Images

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fatima Zahra Oujebbour,
  • Valerie Kaftandjian-Doudet,
  • Yassine Fdil,
  • Houda Hassouane

摘要

X-ray Computed Tomography (XCT) is widely used in industrial inspection for detailed 3D visualization of multi-material components. However, segmenting XCT data remains challenging due to material heterogeneity, noise, and artifacts. This study presents a deep learning pipeline based on the U-Net architecture to automate segmentation tasks in multi-energy XCT scans. Applied to a complex test object, the model achieved high performance, with high accuracy and Dice scores across 17 distinct object classes—a level of granularity rarely addressed in existing studies. This work establishes a reproducible framework for XCT analysis and lays the foundation for more advanced architectures such as ResNet-based U-Nets and future 3D segmentation using models like 3D U-Net or V-Net.