Scholarlink

<p>A brain tumor (BT) is an unusual and uncontrolled development of cells in the brain. If it is not detected and treated at an early stage, it can progress to a serious condition that may result in death. The current BT classification approaches struggle to handle noisy images and inadequate feature extraction methodologies. Some methods often face issues in handling complex tumor structures, overlapping tissue regions, and intensity levels. To address these limitations, a new Hybrid Jaya Skill Optimization Algorithm (HJSOA) is proposed for classifying BTs from MRI images. This method integrates the Jaya Algorithm with the Skill Optimization Algorithm (SOA). In the preprocessing stage, a Bilateral filter is applied to medical images to remove noise and artifacts, thereby improving image quality and enhancing classification accuracy. Weighted Fuzzy K-means (WFKM) clustering is used to segment the pre-processed images. This approach enables accurate delineation of BT boundaries, thereby enhancing overall performance and minimizing false detections. In addition, Discrete Wavelet Transform (DWT) with the Local Texton XOR Patterns (LTxXORP) is utilized to derive the features. It helps to handle complex and subtle patterns in the input data, which leads to faster computation and improves the generalization of the model. Finally, the BT is classified using the Deep Neuro Fuzzy–Visual Geometry Group 16 (DNF_VGG16) model, which is formed by combining the Deep Neuro-Fuzzy Network (DNFN) with the Visual Geometric Group (VGG16) architecture. Finally, the classification of BT is implemented using the Hybrid Jaya Skill Optimization Algorithm (HJSOA) and DNF_VGG16. The effectiveness of the proposed HJSOA + DNF_VGG16 model is evaluated by using different training configurations on the BRATS 2018 and Figshare datasets. In the BRATS 2018 dataset, the developed HJSOA + DNF_VGG16 model attained maximum values of 0.930 for accuracy, 0.950 for Total Positive Rate (TPR), and 0.940 for Total Negative Rate (TNR) at the learning sets of 90. The performance enhancement achieved by the developed model based on accuracy is 17.0%, 14.45%, 11.6%, 7.84%, 4.90%, 3.50%, 1.58%, 0.75%, and 0.39% higher than the Deep Convolutional Neural Network (DCNN), Multi-task Attention Guided encoder-decoder Network (MAG-Net), Hybrid Convolutional Neural Network and support vector Machine (CNN-SVM), Deep Neural Network (DNN), conditional Aquila horse herd optimization driven deep neuro-fuzzy network (CAHO-based DNFN), ResNet101 + Grey Wolf Optimizer (GWO) + Multi-Transformer Fusion model, Deep Neuro Fuzzy-Visual Geometric Group-16 (DNF_VGG16) models, ConvNeXt Base model, and Transformer‐Guided Cross‐Scale Attention and Deep Semantic Fusion network (TCAFNet).</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

A Novel Feature Extraction Technique and Hybrid Jaya Skill Optimization with DNF_VGG16 for Brain Tumor Classification Using MRI Image

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月14日

查看全文
  • Nagaraju Arumalla,
  • Veerraju Gampala

摘要

A brain tumor (BT) is an unusual and uncontrolled development of cells in the brain. If it is not detected and treated at an early stage, it can progress to a serious condition that may result in death. The current BT classification approaches struggle to handle noisy images and inadequate feature extraction methodologies. Some methods often face issues in handling complex tumor structures, overlapping tissue regions, and intensity levels. To address these limitations, a new Hybrid Jaya Skill Optimization Algorithm (HJSOA) is proposed for classifying BTs from MRI images. This method integrates the Jaya Algorithm with the Skill Optimization Algorithm (SOA). In the preprocessing stage, a Bilateral filter is applied to medical images to remove noise and artifacts, thereby improving image quality and enhancing classification accuracy. Weighted Fuzzy K-means (WFKM) clustering is used to segment the pre-processed images. This approach enables accurate delineation of BT boundaries, thereby enhancing overall performance and minimizing false detections. In addition, Discrete Wavelet Transform (DWT) with the Local Texton XOR Patterns (LTxXORP) is utilized to derive the features. It helps to handle complex and subtle patterns in the input data, which leads to faster computation and improves the generalization of the model. Finally, the BT is classified using the Deep Neuro Fuzzy–Visual Geometry Group 16 (DNF_VGG16) model, which is formed by combining the Deep Neuro-Fuzzy Network (DNFN) with the Visual Geometric Group (VGG16) architecture. Finally, the classification of BT is implemented using the Hybrid Jaya Skill Optimization Algorithm (HJSOA) and DNF_VGG16. The effectiveness of the proposed HJSOA + DNF_VGG16 model is evaluated by using different training configurations on the BRATS 2018 and Figshare datasets. In the BRATS 2018 dataset, the developed HJSOA + DNF_VGG16 model attained maximum values of 0.930 for accuracy, 0.950 for Total Positive Rate (TPR), and 0.940 for Total Negative Rate (TNR) at the learning sets of 90. The performance enhancement achieved by the developed model based on accuracy is 17.0%, 14.45%, 11.6%, 7.84%, 4.90%, 3.50%, 1.58%, 0.75%, and 0.39% higher than the Deep Convolutional Neural Network (DCNN), Multi-task Attention Guided encoder-decoder Network (MAG-Net), Hybrid Convolutional Neural Network and support vector Machine (CNN-SVM), Deep Neural Network (DNN), conditional Aquila horse herd optimization driven deep neuro-fuzzy network (CAHO-based DNFN), ResNet101 + Grey Wolf Optimizer (GWO) + Multi-Transformer Fusion model, Deep Neuro Fuzzy-Visual Geometric Group-16 (DNF_VGG16) models, ConvNeXt Base model, and Transformer‐Guided Cross‐Scale Attention and Deep Semantic Fusion network (TCAFNet).