Scholarlink

<p>Cervical Spine Fractures (CSF) are an acute medical condition that might end in death or lifelong paralysis. Conventional methods for detecting fractures, such as manually interpreting radiography images, are tedious and susceptible to human error. In medical imaging applications, like bone fracture detection, Deep Learning (DL) techniques have demonstrated encouraging outcomes. This study proposes a new CSF detection model. Initially, Anisotropic filtering is applied in the input image to improve the image resolution. Then, the Log Transform Applied Deep Joint approach (LTA-DJS) is proposed for segmentation. Further, features like intensity weight-based MBP (IW-MBP), Local Gabor XOR Pattern (LGXP) and Multi-Texton (MT) features are extracted. In IW-MBP, Sobel-filtered edge intensity enhances descriptor accuracy. The final stage is detection that employs a Hybrid classifier combining batch-normalised weighted ReLU with Weighted Pooling-based Parallel Convolution Neural network (BNWRWP-PCNN) and Bi-GRU models. The outcomes resulting from the average of Bi-GRU and Batch Normalized Weighted ReLU with Weighted Pooling-Based Parallel Convolution Neural Network (BNWRWP-PCNN) provide the absolute outcomes on CSF detection. From the study, the proposed BNWRWP-PCNN + Bi-GRU grasped a better sensitivity of 0.941%, while existing approaches scored a lesser sensitivity.</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

Improved Deep Joint Segmentation with Enhanced Feature Set for Cervical Spine Fracture Classification

  • 期刊论文

  • 发表日期：

    2025年12月18日

查看全文
  • K Goutham Raju,
  • S Ravikumar

摘要

Cervical Spine Fractures (CSF) are an acute medical condition that might end in death or lifelong paralysis. Conventional methods for detecting fractures, such as manually interpreting radiography images, are tedious and susceptible to human error. In medical imaging applications, like bone fracture detection, Deep Learning (DL) techniques have demonstrated encouraging outcomes. This study proposes a new CSF detection model. Initially, Anisotropic filtering is applied in the input image to improve the image resolution. Then, the Log Transform Applied Deep Joint approach (LTA-DJS) is proposed for segmentation. Further, features like intensity weight-based MBP (IW-MBP), Local Gabor XOR Pattern (LGXP) and Multi-Texton (MT) features are extracted. In IW-MBP, Sobel-filtered edge intensity enhances descriptor accuracy. The final stage is detection that employs a Hybrid classifier combining batch-normalised weighted ReLU with Weighted Pooling-based Parallel Convolution Neural network (BNWRWP-PCNN) and Bi-GRU models. The outcomes resulting from the average of Bi-GRU and Batch Normalized Weighted ReLU with Weighted Pooling-Based Parallel Convolution Neural Network (BNWRWP-PCNN) provide the absolute outcomes on CSF detection. From the study, the proposed BNWRWP-PCNN + Bi-GRU grasped a better sensitivity of 0.941%, while existing approaches scored a lesser sensitivity.