Development of Efficient Brain Tumor Classification on MRI Image Results Using EfficientNet

Abstract

Brain tumors are diseases that affect the most vital organs of the human body. Abnormal cell development causes the growth of lesions in the human brain. In visualizing the emergence of a brain tumor, MRI (Magnetic Resonance Imaging) is a relatively good method as it has no radiation compared to other methods. Artificial intelligence is expected to accelerate radiologists in detecting a tumor’s emergence. This study proposes an automatic classification using a deep learning architecture with eight EfficientNet models (BO-B7) variations to classify MRI results into a normal brain or brain with a tumor. The models perform well, in which EfficientNet-B7 achieves the highest training accuracy of 99.71% and validation accuracy of 99.67%. Compared to conventional CNN, EfficientNet is superior in the performance and the computation time. From the experimental results, the level of accuracy of conventional CNN is less than EfficienNet. This indicates that the architectural modifications presented in the EfficientNet model, by combining the layer numbers, image resolution and the channels can improve the conventional CNN in classifying the MRI results.