A Fusion-Based Deep Approach for Enhanced Brain Tumor Classification

Abstract

A brain tumor is an aberrant proliferation of living cells in the brain that grows uncontrolled, posing a significant risk to the human body. Precise segmentation and classification of tumors are essential for future prognosis and therapy planning. Due to its susceptibility to errors and time-consuming nature, radiologists are required to use an automated approach for brain tumor identification. This study presents the development of an integrated and fully automated classification system for categorizing brain tumors in MRI images. The system combines the deep representations of features obtained from two distinct deep learning models, ResNet18 and ResNet50, to create feature vectors that more effectively distinguish between various classes. The vectors of features are then inputted into the machine learning layer to categorize them into four distinct classes. The model's performance was tested using a publicly available dataset from the internet. The testing findings showed that the fusion model suggested attained an accuracy rate for classification of 92.47%. Ultimately, the findings were compared to existing approaches, and the suggested model demonstrated superior performance considerably.