Enhancing Brain Tumor Classification Using Efficientnetb3 And Squeeze-And-Excitation Attention Mechanisms

Abstract

The presence of brain tumors is one of the major problems for modern medicine, mainly because it is a rather complex phenomenon. Accurate and rapid tumor identification plays an important role in the treatment and subsequent recovery of the patient. It is particularly informative in the detection of brain tumors as an MRI is utilized to yield clear images of the structures present in the brain. However, the assessment of MRI scans by human radiologists is time-consuming, subjective, and prone to various errors that affect the accuracy and timeliness of diagnosis. Considering all these factors, it can be stated that the need for the creation of artificial intelligence-driven diagnostic tools can enhance the speed and accuracy of brain tumor identification. The focus of this paper is thus the development of an automated system for categorizing brain tumors using the EfficientNetB3 deep learning model, which may be improved through the addition of Squeeze and Excitation blocks. The SE Block, an attention mechanism, assists in achieving the feature extraction with a major focus on the significant zones of the image. The proposed model underwent training and testing based on the 7,023 MRI image data. The model met the result of 99.24%, and it showed that it offered 8-10% better performance than present benchmarks. The good performance exhibited here corroborates that deep learning with attention models can enhance the identification of brain tumors to be at par with other benchmark systems. This will be a valuable contribution towards improved medical workflows and overall better healthcare systems for Future Computer-aided Diagnosis Systems (CDS).