Paper Key : IRJ************477
Author: Dr.v.jeyalakshmi,Muthu Lakshman N,S.s.akash Varsa Raja College
Date Published: 01 Jul 2024
Abstract
MRI is the best and most commonly used option for diagnosing brain tumors. Brain tumor analysis often involves high-dimensional data, where each voxel or region represents a feature in the existing works. Clustering high-dimensional data is more challenging due to the curse of dimensionality and it leads to computational complexity and loss of interpretability. In the existing brain tumor segmentation and classification techniques, it does not classify the tumors with each stage identification and there is a no adequate kernels to convolve feature map and softmax for pixel-wise classification. To overcome these issues, an innovative Unsupervised Feature Selection via Mean Shift Clustering based WOA and Adaptive filter with DCT segmentation is proposed. In this proposed approach, a Mean Shift Clustering based WOA is developed to overcome the variations in dynamic imaging modalities by dividing the data into non-hierarchical groups and reconstructs the selected data from the reduced feature set. To overcome the limitations in the segmentation and classification process, an innovative PNN-based DCT segmentation is introduced, which segments the brain tumor image by utilizing the ICD-11 tool to identify the tumors based on each stages and PNN approach obtain the appropriate kernel by applying adaptive filters to convolve feature map thereby improved the pixel-wise classification, and enhanced the detection accuracy. As a result, the proposed method achieves the high accuracy, precision, precision, recall, F1-score, sensitivity and specificity. Hence, the proposed novel achieves a better unsupervised feature selection-based clustering process for brain tumor detection and classification.
DOI Requested