Brain Image Segmentation via GLCM Features and CNN Classification for Improved Image Retrieval Using Machine Learning

Authors:

Abstract
Improvement methods based on picture segmentation are commonly employed in medical image analysis. Pixels are commonly put together in groupings of varied sizes to generate an image. The available pixel intensities in an image allow for the separation of distinct image regions. Edge-based and boundary-based segmentation methods are the two primary classifications of this type of analysis. The boundary-based segmentation technique is the more widespread of the two. The image of the tumor needs to be segmented into numerous fragments, each of which can then be analyzed separately. It can annotate data and locate locations of interest in medical image processing. The issues with the existing paper are addressed in the suggested study. Feature extraction using GLCM has been used for classification of tumor images. Analysis of data in segments is routine. The CNN classifier's segments are labelled by an analytical procedure. The proposed method boosts information quality and reduces error-prone training in MRI brain tumor pictures by increasing the learning coefficient of the back propagation layer in the neural network. As a result, the current CNN Shallow Net architecture achieves a higher classification rate of 99.61% thanks to its increased specificity (98.56%) and sensitivity (98.12%). The deep web achieves a segmentation accuracy of 99.58 percent, with an additional diagnostic accuracy of 98.5 percent in identifying cancerous spots.

This publication has 19 references indexed in Scilit: