Scholarlink

<p>The accurate identification of structural and neurodevelopmental abnormalities in infant brain MRI is critical for early diagnosis, timely clinical intervention, and improved developmental outcomes. In this study, a deep learning–based framework utilizing pre-trained convolutional neural networks (CNNs) is proposed for the multi-class classification of infant brain abnormalities using magnetic resonance imaging (MRI) scans, addressing the growing demand for automated and reliable diagnostic support in pediatric neuroimaging. Four state-of-the-art CNN architectures, Inception-ResNet-v2, ResNet-50, DenseNet-121, and VGG19, were optimized using transfer learning to enable efficient feature extraction while reducing dependence on large labeled datasets. All the MRI images were resized to 224 × 224 × 3 pixels to satisfy the model input requirements. Training was conducted for 50 epochs using the Adam optimizer with an initial learning rate of 0.0001 and batch size of 32, with categorical cross-entropy employed as the loss function. The model performance was comprehensively evaluated using accuracy, precision, recall, F1-score, specificity, sensitivity, and area under the receiver operating characteristic curve (AUC). Among the evaluated architectures, Inception-ResNet-v2 achieved the highest accuracy (98.31%), followed by ResNet-50 (98.04%), DenseNet-121 (97.86%), and VGG19 (97.42%). The results demonstrate the ability of transfer-learning-based CNN models to capture subtle structural variations in infant brain MRI scans with high discriminative power across severity-based classes. These findings indicate that the proposed framework provides a robust, noninvasive, and clinically practical tool to support early abnormality detection and informed clinical decision-making in pediatric neuroimaging.</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

Infant Brain Abnormality Classification Using Transfer Learning and Deep CNN Models from MRI

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • B. N. Yashasvi,
  • N. P. Kavya

摘要

The accurate identification of structural and neurodevelopmental abnormalities in infant brain MRI is critical for early diagnosis, timely clinical intervention, and improved developmental outcomes. In this study, a deep learning–based framework utilizing pre-trained convolutional neural networks (CNNs) is proposed for the multi-class classification of infant brain abnormalities using magnetic resonance imaging (MRI) scans, addressing the growing demand for automated and reliable diagnostic support in pediatric neuroimaging. Four state-of-the-art CNN architectures, Inception-ResNet-v2, ResNet-50, DenseNet-121, and VGG19, were optimized using transfer learning to enable efficient feature extraction while reducing dependence on large labeled datasets. All the MRI images were resized to 224 × 224 × 3 pixels to satisfy the model input requirements. Training was conducted for 50 epochs using the Adam optimizer with an initial learning rate of 0.0001 and batch size of 32, with categorical cross-entropy employed as the loss function. The model performance was comprehensively evaluated using accuracy, precision, recall, F1-score, specificity, sensitivity, and area under the receiver operating characteristic curve (AUC). Among the evaluated architectures, Inception-ResNet-v2 achieved the highest accuracy (98.31%), followed by ResNet-50 (98.04%), DenseNet-121 (97.86%), and VGG19 (97.42%). The results demonstrate the ability of transfer-learning-based CNN models to capture subtle structural variations in infant brain MRI scans with high discriminative power across severity-based classes. These findings indicate that the proposed framework provides a robust, noninvasive, and clinically practical tool to support early abnormality detection and informed clinical decision-making in pediatric neuroimaging.