Scholarlink

This study presents a deep learning approach for automated lumbar vertebrae segmentation from magnetic resonance imaging (MRI) data. The proposed pipeline employs a U-Net architecture trained on the SPIDER dataset using a composite loss that combines Cross-Entropy and Intersection over Union (IoU) terms to enhance both pixel-wise accuracy and spatial consistency. Model evaluation through 5-fold cross-validation achieved a mean IoU of 0.76 and a Dice coefficient of 0.86, confirming robust generalization across diverse samples. The system demonstrates potential as a computer-aided diagnosis (CAD) tool for spinal assessment.

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-Based Lumbar Vertebrae Segmentation in MRI Scans

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Maria Enache,
  • Andreea-Nicola Moisoiu,
  • Otilia Zvorișteanu

摘要

This study presents a deep learning approach for automated lumbar vertebrae segmentation from magnetic resonance imaging (MRI) data. The proposed pipeline employs a U-Net architecture trained on the SPIDER dataset using a composite loss that combines Cross-Entropy and Intersection over Union (IoU) terms to enhance both pixel-wise accuracy and spatial consistency. Model evaluation through 5-fold cross-validation achieved a mean IoU of 0.76 and a Dice coefficient of 0.86, confirming robust generalization across diverse samples. The system demonstrates potential as a computer-aided diagnosis (CAD) tool for spinal assessment.