Scholarlink

<p>This paper presents a new framework for classifying brain tumours, using a self-adaptive technique to improve image processing. The framework utilises a carefully selected dataset of T1-weighted contrast-enhanced pictures from 233 individuals, which includes meningioma, glioma, and pituitary tumours. It employs advanced pre-processing and augmentation techniques to boost the performance of the model. The novelty resides in the utilisation of Adaptive Contrast Limited Histogram Equalisation (CLAHE) and Self-Adaptive Spatial Attention techniques in a collaborative manner. The CLAHE algorithm is designed to improve grayscale images by adjusting contrast based on local image characteristics. Meanwhile, the Self-Adaptive Spatial Attention mechanism enhances sensitivity to subtle tumour features by assigning dynamic weights to spatial locations in critical regions. The model architecture incorporates cutting-edge transfer learning models, including DenseNet169, DenseNet201, ResNet152, and InceptionResNetV2. It also utilises techniques such as batch normalisation, dropout, layer normalisation, and an adaptive learning rate strategy to mitigate overfitting and improve flexibility. The model achieves exceptional classification performance by employing the Adam optimizer and softmax activation function, resulting in an accuracy of 94.85%, precision of 95.16%, and recall of 94.60%. This method signifies a notable progress in computer-assisted diagnosis for neuroimaging, showcasing improved precision and applicability in the detection of brain tumours.</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

NeuroInsight: a revolutionary self-adaptive framework for precise brain tumor classification in medical imaging using adaptive deep learning

  • 期刊论文

  • 发表日期：

    2025年1月3日

查看全文
  • Sonia Arora,
  • Gouri Sankar Mishra

摘要

This paper presents a new framework for classifying brain tumours, using a self-adaptive technique to improve image processing. The framework utilises a carefully selected dataset of T1-weighted contrast-enhanced pictures from 233 individuals, which includes meningioma, glioma, and pituitary tumours. It employs advanced pre-processing and augmentation techniques to boost the performance of the model. The novelty resides in the utilisation of Adaptive Contrast Limited Histogram Equalisation (CLAHE) and Self-Adaptive Spatial Attention techniques in a collaborative manner. The CLAHE algorithm is designed to improve grayscale images by adjusting contrast based on local image characteristics. Meanwhile, the Self-Adaptive Spatial Attention mechanism enhances sensitivity to subtle tumour features by assigning dynamic weights to spatial locations in critical regions. The model architecture incorporates cutting-edge transfer learning models, including DenseNet169, DenseNet201, ResNet152, and InceptionResNetV2. It also utilises techniques such as batch normalisation, dropout, layer normalisation, and an adaptive learning rate strategy to mitigate overfitting and improve flexibility. The model achieves exceptional classification performance by employing the Adam optimizer and softmax activation function, resulting in an accuracy of 94.85%, precision of 95.16%, and recall of 94.60%. This method signifies a notable progress in computer-assisted diagnosis for neuroimaging, showcasing improved precision and applicability in the detection of brain tumours.