Scholarlink

Spinal Tuberculosis disease classification and prediction is avital task for identification of tuberculosis on bone as it leads to kyphosis, gibbus formation and osteomyelitis events. Those events have been classified into types and its stages on basis of various spatial and temporal properties of image morphology of the vertebral disk patterns in the symptomatic sites of the lumbar spine, thoracic spine obtained using MRI based imaging system. MRI imaging offers the excellent visualization of the vertebral disk patterns at the symptomatic sites. MRI images contain some sounds such as Gaussian noise, poisson noise, Impulse Noise and Quantization noises. Further the image contains complex appearance of the disk which makes difficult discriminate the stages and types of Spinal Tuberculosis Events. In order to manage those challenges, machine learning model using the unsupervised learning techniques becomes complex and time consuming. However deep learning architecture effectively discriminate the complex morphology structures of the vertebral disk patterns into spinal tuberculosis disease class events. In this paper, a Spatial and temporal constrained convolution Neural network (SPCCNN) is designed to classify and detect spinal tuberculosis into classes according to its severeness in levels. Initially Image noise eliminated through Wavelet Threshold based Image Denoising technique. Principle Component analysis has been employed to extract the feature of the preprocessed image. As samples of the image is enhanced using image augmentation. Image augmentation is an auxiliary convolution Neural network for increasing the size of the training set without obtaining images. A Convolution Neural network processes the features extracted by Principle component analysis at different layers of the network. The convolution layer and activation layer in particular are enhanced using spatial and temporal constraints on hyper parameters. Output layer produces the disease classes of the tuberculosis events such as kyphosis, gibbus formation and osteomyelitis with stages using cross entrophy and softmax layers for error handling and normalization of the diseases. Experimental analysis is carried out using matlab simulation on the samples of 100 MRI images were collected from Kovai Medical Center and Hospital.60% of this data was used for characterization, 20% was used for testing, and the remaining 20% ​​was taken for 5-fold cross-validation through the image confusion matrix. The proposed model performed well in terms of accuracy, specificity, and sensitivity compared to traditional approaches.

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

Classification and Prediction of Spinal Tuberculosis Disease Using Optimization of Convolution Neural Network Using Spatial and Temporal Constraints

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • K. T. Askarali,
  • E. J. Thomson Fredrik

摘要

Spinal Tuberculosis disease classification and prediction is avital task for identification of tuberculosis on bone as it leads to kyphosis, gibbus formation and osteomyelitis events. Those events have been classified into types and its stages on basis of various spatial and temporal properties of image morphology of the vertebral disk patterns in the symptomatic sites of the lumbar spine, thoracic spine obtained using MRI based imaging system. MRI imaging offers the excellent visualization of the vertebral disk patterns at the symptomatic sites. MRI images contain some sounds such as Gaussian noise, poisson noise, Impulse Noise and Quantization noises. Further the image contains complex appearance of the disk which makes difficult discriminate the stages and types of Spinal Tuberculosis Events. In order to manage those challenges, machine learning model using the unsupervised learning techniques becomes complex and time consuming. However deep learning architecture effectively discriminate the complex morphology structures of the vertebral disk patterns into spinal tuberculosis disease class events. In this paper, a Spatial and temporal constrained convolution Neural network (SPCCNN) is designed to classify and detect spinal tuberculosis into classes according to its severeness in levels. Initially Image noise eliminated through Wavelet Threshold based Image Denoising technique. Principle Component analysis has been employed to extract the feature of the preprocessed image. As samples of the image is enhanced using image augmentation. Image augmentation is an auxiliary convolution Neural network for increasing the size of the training set without obtaining images. A Convolution Neural network processes the features extracted by Principle component analysis at different layers of the network. The convolution layer and activation layer in particular are enhanced using spatial and temporal constraints on hyper parameters. Output layer produces the disease classes of the tuberculosis events such as kyphosis, gibbus formation and osteomyelitis with stages using cross entrophy and softmax layers for error handling and normalization of the diseases. Experimental analysis is carried out using matlab simulation on the samples of 100 MRI images were collected from Kovai Medical Center and Hospital.60% of this data was used for characterization, 20% was used for testing, and the remaining 20% ​​was taken for 5-fold cross-validation through the image confusion matrix. The proposed model performed well in terms of accuracy, specificity, and sensitivity compared to traditional approaches.