Search

• To present an advanced median filter (AMF) in image pre-processing for the removal of noise and image contrast enhancement. • To extract the significant features using convolutional neural network (CNN), Haralick Texture Features and speed-up robust features (SURF) for minimizing the complexity. • To reduce the overall processing time and retain optimal results, an improved Fennec Fox Optimization (IFFO) algorithm was used for feature selection. • To introduce Dual Attention based Resnet-50 BiGRU (DA_ResBiGRU) for accurate Brain Tumor Classification with reduced error rates. Brain tumors are a major health issue that causes enormous problems for people's health. Detection is a very complicated process of classifying tumor types, where early detection can fix the problem as soon as possible. Manual diagnosis and detection pose a risk to human life. In addition, the classification of malignant tumors and gliomas is difficult due to the asymmetric or uneven boundary of the skull. For resolving the classification issues, the Dual Attention method based on the Resnet-50 with bidirectional gated recurrent unit (DA_ResBiGRU) model is proposed. Firstly, MRI images are pre-processed using Advanced Median Filter (AMF), and then the pre-processed data is transferred to the feature extraction phase. Significant features are extracted using Convolutional Neural Network (CNN), Haralick Texture Features and Speed-up Robust Features (SURF). Using the extracted features, optimized features are retrieved further by the Enhanced Fennec Fox Optimization (IFFO) algorithm. Here, brain tumor classification datasets and Figshare datasets are used for experimental analysis. Performance metrics like accuracy, F1 score, recall, precision, etc., are used to study the model performance. The proposed model achieves 98.53% accuracy, 97.7% precision, 97% recall, and 98% F1 score in brain tumor classification datasets, and 99.3% accuracy, 98.8% precision, 98.8% recall and 98.7% at F1 score are achieved in Figshare dataset. The values ​​obtained show that the proposed model overtakes other models. [ABSTRACT FROM AUTHOR]