Scholarlink

<p>Schizophrenia is a serious brain illness caused by distinct abnormalities present within the structure of the brain. Several studies have shown the utility of deep learning techniques in detecting schizophrenia. In this paper, we introduce the orthogonal linear transformation of structural MRI followed by an ensemble of tuned deep learning models for automated and accurate detection of schizophrenia. Our proposed work employs pre-processing of structural MRI scans using normalization and filtering technique to reduce noise from the neuroimaging modalities. We perform an orthogonal linear transformation of structural MR images using <i>Principal Component Analysis (PCA)</i> to minimise the dimensionality and identify highly informative patterns. The PCA transformation improves the training of deep <i>Convolutional Neural Network (CNN)</i> models by improving data representation and minimizing computational overhead. The reduced-dimensional PCA components are then input to three modified CNN architectures- VGGNet, ResNet, and DenseNet. We employ architectural and training modifications to optimize these three deep CNN models by adjusting the number of filters and applying additional drop-out layers. The deep models are also fine-tuned by modifying the learning rate, batch size, and optimizers. The modified deep CNNs not only reduce the computational burden but also achieve improved detection performance with high accuracy and minimum loss. To increase the confidence of the results and further enhance accuracy, we employ an ensemble of VGGNet, ResNet, and DenseNet, combining the strengths of these modified CNN architectures. The ensemble approach achieves high accuracy and provides better generalization of results, indicating the exceptional potential of combining PCA with deep CNNs for the early and automated detection of schizophrenia.</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

Enhanced Schizophrenia detection using ensemble deep learning and orthogonal linear transformation of structural MRI

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Ashima Tyagi,
  • Vibhav Prakash Singh,
  • Manoj Madhava Gore

摘要

Schizophrenia is a serious brain illness caused by distinct abnormalities present within the structure of the brain. Several studies have shown the utility of deep learning techniques in detecting schizophrenia. In this paper, we introduce the orthogonal linear transformation of structural MRI followed by an ensemble of tuned deep learning models for automated and accurate detection of schizophrenia. Our proposed work employs pre-processing of structural MRI scans using normalization and filtering technique to reduce noise from the neuroimaging modalities. We perform an orthogonal linear transformation of structural MR images using Principal Component Analysis (PCA) to minimise the dimensionality and identify highly informative patterns. The PCA transformation improves the training of deep Convolutional Neural Network (CNN) models by improving data representation and minimizing computational overhead. The reduced-dimensional PCA components are then input to three modified CNN architectures- VGGNet, ResNet, and DenseNet. We employ architectural and training modifications to optimize these three deep CNN models by adjusting the number of filters and applying additional drop-out layers. The deep models are also fine-tuned by modifying the learning rate, batch size, and optimizers. The modified deep CNNs not only reduce the computational burden but also achieve improved detection performance with high accuracy and minimum loss. To increase the confidence of the results and further enhance accuracy, we employ an ensemble of VGGNet, ResNet, and DenseNet, combining the strengths of these modified CNN architectures. The ensemble approach achieves high accuracy and provides better generalization of results, indicating the exceptional potential of combining PCA with deep CNNs for the early and automated detection of schizophrenia.