Scholarlink

<p>Schizophrenia (SC) is a complex mental disorder with diverse symptoms that make diagnosis challenging. This study aims to enhance diagnostic accuracy for SC using deep learning techniques applied to resting-state functional MRI (rsfMRI) data, capturing both spatial and temporal features of brain activity. We introduced a novel slice-wise classification approach using convolutional neural networks (CNNs) to analyze brain activity from rsfMRI images. Preprocessing included normalization, noise reduction, and contrast enhancement. The study utilized data from 158 subjects, including 83 schizophrenia patients and 75 healthy controls. We combined CNN-extracted features with traditional machine learning models such as support vector machines (SVM), random forest (RF), and gradient boosting (GB) to boost classification performance. Transfer learning using pre-trained models like VGG16, ResNet50, and Xception was applied to leverage advanced feature extraction. Model performance was evaluated based on precision, accuracy, recall, and F1 score. The CNN-based approach achieved significant improvements in classification accuracy, with a peak accuracy of 98.67%. The hybrid models combining CNN features with SVM, RF, and GB achieved accuracies of 97.01%, 98.44%, and 98.84%, respectively. The CNN model alone achieved a precision of 0.9826, recall of 1.0, and F1 score of 0.9912. Pre-trained models also demonstrated high performance, with ResNet50 achieving an accuracy of 98.71%. Brain regions such as the frontal lobe (slices 5 and 9) and temporal lobe (slices 15 and 25) were identified as key areas with significant differences between schizophrenia patients and healthy controls. This study demonstrates the effectiveness of deep learning models, particularly CNNs and hybrid approaches with traditional machine learning models, in enhancing schizophrenia diagnosis using rsfMRI data. Identifying key brain regions provides insights into schizophrenia’s neurobiological underpinnings. Future research should focus on larger datasets and interpretable models to advance diagnosis.</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

Enhancing schizophrenia diagnosis through deep learning: a resting-state fMRI approach

  • 期刊论文

  • 发表日期：

    2025年5月22日

查看全文
  • Zahra Raeisi,
  • Mehri Mehrnia,
  • Reza Ahmadi Lashaki,
  • Fatemeh Abedi Lomer

摘要

Schizophrenia (SC) is a complex mental disorder with diverse symptoms that make diagnosis challenging. This study aims to enhance diagnostic accuracy for SC using deep learning techniques applied to resting-state functional MRI (rsfMRI) data, capturing both spatial and temporal features of brain activity. We introduced a novel slice-wise classification approach using convolutional neural networks (CNNs) to analyze brain activity from rsfMRI images. Preprocessing included normalization, noise reduction, and contrast enhancement. The study utilized data from 158 subjects, including 83 schizophrenia patients and 75 healthy controls. We combined CNN-extracted features with traditional machine learning models such as support vector machines (SVM), random forest (RF), and gradient boosting (GB) to boost classification performance. Transfer learning using pre-trained models like VGG16, ResNet50, and Xception was applied to leverage advanced feature extraction. Model performance was evaluated based on precision, accuracy, recall, and F1 score. The CNN-based approach achieved significant improvements in classification accuracy, with a peak accuracy of 98.67%. The hybrid models combining CNN features with SVM, RF, and GB achieved accuracies of 97.01%, 98.44%, and 98.84%, respectively. The CNN model alone achieved a precision of 0.9826, recall of 1.0, and F1 score of 0.9912. Pre-trained models also demonstrated high performance, with ResNet50 achieving an accuracy of 98.71%. Brain regions such as the frontal lobe (slices 5 and 9) and temporal lobe (slices 15 and 25) were identified as key areas with significant differences between schizophrenia patients and healthy controls. This study demonstrates the effectiveness of deep learning models, particularly CNNs and hybrid approaches with traditional machine learning models, in enhancing schizophrenia diagnosis using rsfMRI data. Identifying key brain regions provides insights into schizophrenia’s neurobiological underpinnings. Future research should focus on larger datasets and interpretable models to advance diagnosis.