Journal of Medical Diagnostic Methods
Open Access

Abstract

CAD-BT: A Computer Aided Diagnostic System Based on Optimized MobileNet for Detection and Classification of Brain Tumor

Muhammad Zaheer Sajid*, Muhammad Imran Sharif, Ayman Youssef, Ayesha Butt, Muneeba Daud, Muhammad Fareed Hamid

Statistically, about seven hundred thousand patients suffer brain tumor disease every year around the world causing the condition unstable and even, dangerous to live with. The condition happens when the brain cells that have undergone mutation or exposure to a radiation source can no longer control their growth relatively. Herein, we introduce CAD-BT, an intelligently automated detection and classification system that is dedicated to determine the severity of brain tumors. The system capitalizes on the MobileNet architecture with a subnet-enhanced version along with residual blocks to increase performance. The CAD-BT preprocessing is done by means of an approach which consists of MSR-enhanced, CLAHE-enhanced, histogram equalization, Gaussian blur and mixed-enhanced techniques that help to optimize tumor visibility, suppress surrounding noisy brain tissue and enhance overall brain tumor examining classification accuracy. To deal with the under-representation of the bottom layer classes, data augmentation methods are used in order to improve generalization on the model. Additionally, the grouping of residual blocks in the MobileNet architecture outperforms the classification task for 4 brain tumor severity levels. In regard to CAD-BT, the outcome is a top-notch quality with a simplified model range and fewer manipulations to carry out. This research evaluates the performance of the CAD-BT system through mixed datasets of various kinds. This is done through the use of a pre-trained model to the purpose of extraction of the semantically important aspect of brain tumor images. The final step involves categorizing the images into one of four classes: Glioma, meningioma, lack of tumor and pituitary. This categorization is brought out by the fact that an SVM classifier layer with linear activation function is included. Astonishingly it achieves a 99.5% accuracy and confirms on a difficult brain tumor dataset which we used to validate the algorithm. Moreover, in this research, five models, namely, InceptionV3, Efficient-Net, VGG19 and Xception, achieve accuracies of 88%, 85%, 87% and 82%, respectively. The analyses and comparison of the results brought forth the effectiveness of the CAD-BT tool in increasing model performance and learning. Therefore, with that said, the CAD-BT system allows neurologists to have a helping tool that provides an appropriate frame for classifying brain tumors of different types in their early stage.

Published Date: 2025-04-21; Received Date: 2024-05-10