Brain Tumor Detection from MRI Images using Enhanced Transfer Learning

Abstract

Brain tumor detection is vital for the diagnosis and treatment of one of the most lethal diseases in both children and adult. Traditional ways of analyzing brain tumour include reviewing Magnetic Resonance Imaging (MRI) images and interpreting results manually by radiologists; this technique has several problems including; complexity of the tumour type, size and location of the tumours may complicate the results of the analysis. Due to this, the above challenges are even more challenging to overcome in areas of scarcity of qualified human resource in healthcare. This study aims at suggesting a novel approach to resolve these challenges together with the utilization of Convolutional Neural Network (CNN) models such as EfficientNet-B2, Xception, Xception with Attention Mechanism. This paper is presented based on how these models can be employed in the classification and detection of the presence of brain tumours; these findings has seen the Xception with Attention Mechanism model register the highest accuracy at 95% among all the employed models. The refinement of the researchers’ strategy is as follows, notably Synthetic Minority Over-sampling Technique (SMOTE) which caters for the issues of class imbalance in the set data. In addition to performing better than conventional non-automated techniques, the system provides a reliable solution for the detection of brain tumors that can be applied globally, benefiting physicians, particularly those to from developing countries, determine the patient’s condition and choose the right treatment plan. Automated medical testing also means that the results are delivered quickly and with greater accuracy also resulting to positive outcomes for the patients.