Scholarlink

<p>Fracture detection on the cervical spine identifies breaks, cracks, or damage caused to the vertebrae in the neck region through several imaging techniques. These fractures can be caused by trauma, accidents, or degenerative conditions. Early detection is crucial for preventing further injury and ensuring proper treatment. Manual detection consumes more time and is prone to errors; thus, automated detection is essential for faster and more accurate diagnoses. Therefore, the Hybrid Pyramid Maxout Network (HyPM-Net) is designed for fracture detection. First, cervical spine images are sourced from a dataset, and images are filtered by an Adaptive Bilateral Filter (ABF). Consequently, the cervical spine region is segmented by UNet +  + with hybrid Tversky loss and Binary Cross Entropy (BCE) (TB_UNet + +). Subsequently, the moment invariant and Gray-Level Co-Occurrence Matrix (GLCM) features are excerpted. Ultimately, cervical spine fracture is detected by the HyPM-Net approach, which is the fusion of Pyramid Network (PyramidNet) and Deep Maxout Network (DMN). Results acquired by the newly proposed HyPM-Net are 92.866% of accuracy, 94.235% of True Positive Rate (TPR), 92.910% of True Negative Rate (TNR), 92.777% of precision, and 93.500% of F1-score.</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

Hybrid Pyramid Maxout Network for Fracture Detection on Cervical Spine Using Computed Tomography Image

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Sreenu Ponnada,
  • Srilakshmi Vellanki,
  • Kiran Kumar Beesetti,
  • Gopalsamy Venkadakrishnan Sriramakrishnan

摘要

Fracture detection on the cervical spine identifies breaks, cracks, or damage caused to the vertebrae in the neck region through several imaging techniques. These fractures can be caused by trauma, accidents, or degenerative conditions. Early detection is crucial for preventing further injury and ensuring proper treatment. Manual detection consumes more time and is prone to errors; thus, automated detection is essential for faster and more accurate diagnoses. Therefore, the Hybrid Pyramid Maxout Network (HyPM-Net) is designed for fracture detection. First, cervical spine images are sourced from a dataset, and images are filtered by an Adaptive Bilateral Filter (ABF). Consequently, the cervical spine region is segmented by UNet +  + with hybrid Tversky loss and Binary Cross Entropy (BCE) (TB_UNet + +). Subsequently, the moment invariant and Gray-Level Co-Occurrence Matrix (GLCM) features are excerpted. Ultimately, cervical spine fracture is detected by the HyPM-Net approach, which is the fusion of Pyramid Network (PyramidNet) and Deep Maxout Network (DMN). Results acquired by the newly proposed HyPM-Net are 92.866% of accuracy, 94.235% of True Positive Rate (TPR), 92.910% of True Negative Rate (TNR), 92.777% of precision, and 93.500% of F1-score.