Scholarlink

<p>In multimodality imaging, image segmentation is crucial, particularly when combining structural images provided by computed tomography (CT) or magnetic resonance imaging (MRI) with functional images acquired by optical technologies or other innovative imaging options. Since changes in tissue volume and form allow for the diagnosis and monitoring of neurological disorders, medical professionals and researchers have taken an interest in accurate brain tissue segmentation in MRI. Traditional machine learning techniques and methods based on convolutional neural networks (CNNs) have formed the basis of multiple suggestions developed throughout the years. Automatic brain tissue segmentation into grey matter (GM), white matter (WM), and cerebrospinal fluid (CSF) from&#xa0;MRI&#xa0;allows visualisation of brain anatomy, analysis of brain changes, and delineation of diseased regions. In order to train a DNN, this paper suggests a semi-supervised learning framework. Using a new augmented target loss function that aims to enhance accuracy in regions of tissue border, we train a CNN&#xa0;to create a model for tissue segmentation. An improved chimp optimisation algorithm (IChOA) is used to fine-tune the DNN model's weight and bias values, leading to even greater accuracy. The initial stage involves normalising all input images in order to identify possible regions of the existing tissues. The next step is to choose the most beneficial features by utilising the IChOA. The last step in segmenting brain tissues is feeding the best-extracted features into the optimised DNN model. Performance metrics such as sensitivity 93.07%, specificity 94.26%, accuracy 97.42%, and dice score 97.54% show that the experimental results on the OASIS and IBSR dataset outperform the current frameworks. Applying the trained model to data from other acquisitions with lesser resolution and fewer gradient directions, as well as the OASIS and IBSR datasets, we show good tissue segmentation accuracy.</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

Accurate Human Brain Tissue Segmentation in MRI Scans Using Optimized Deep Neural Network

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Mohammed Razia Alangir Banu,
  • K. Purushottama Rao,
  • Arpita Gupta

摘要

In multimodality imaging, image segmentation is crucial, particularly when combining structural images provided by computed tomography (CT) or magnetic resonance imaging (MRI) with functional images acquired by optical technologies or other innovative imaging options. Since changes in tissue volume and form allow for the diagnosis and monitoring of neurological disorders, medical professionals and researchers have taken an interest in accurate brain tissue segmentation in MRI. Traditional machine learning techniques and methods based on convolutional neural networks (CNNs) have formed the basis of multiple suggestions developed throughout the years. Automatic brain tissue segmentation into grey matter (GM), white matter (WM), and cerebrospinal fluid (CSF) from MRI allows visualisation of brain anatomy, analysis of brain changes, and delineation of diseased regions. In order to train a DNN, this paper suggests a semi-supervised learning framework. Using a new augmented target loss function that aims to enhance accuracy in regions of tissue border, we train a CNN to create a model for tissue segmentation. An improved chimp optimisation algorithm (IChOA) is used to fine-tune the DNN model's weight and bias values, leading to even greater accuracy. The initial stage involves normalising all input images in order to identify possible regions of the existing tissues. The next step is to choose the most beneficial features by utilising the IChOA. The last step in segmenting brain tissues is feeding the best-extracted features into the optimised DNN model. Performance metrics such as sensitivity 93.07%, specificity 94.26%, accuracy 97.42%, and dice score 97.54% show that the experimental results on the OASIS and IBSR dataset outperform the current frameworks. Applying the trained model to data from other acquisitions with lesser resolution and fewer gradient directions, as well as the OASIS and IBSR datasets, we show good tissue segmentation accuracy.