Abstract:

Abstract Brain tumor is an abnormal tissue growth which may lead to cancer and it is characterized by the excessive cell proliferation in certain parts of the brain. One of the current reliable technologies that may be employed to identify brain tumor is to apply Magnetic Resonance Imaging (MRI) scans. The scanned MRI images are then monitored and conventionally examined by the medical specialists to observe the existence of tumor. As the number of people suffering brain tumor is very much increased and their corresponding mortality rate has reached 18,600 in the year 2021, research on devising more effective and efficient tools to assist the medical specialists on the identification of brain tumor is considered very urgent. In the previous studies, the Convolutional Neural Network (CNN) based models demonstrate their capability to detect brain tumor with 96% classification accuracy and they are found to be more reliable than other machine learning based models. In an attempt to obtain the best classification accuracy on both binary and multi-class MRI brain images, some powerful pretrained deep CNN models namely VGG16, VGG19, ResNet50, ResNet101, and InceptionResNetV2 are computationally experimented using publicly open MRI datasets. State of the art accuracy of the pretrained models is achieved by fine tuning the parameters of the convolutional layers of the base models and followed by feeding the high-level feature maps extracted from each corresponding base model either into flatten layers or into global average pooling layers prior to classifying the tumor by fully connected layers. The highest testing accuracy score as high as 99% is achieved by the VGG16 and InceptionResnetV2 on binary MRI image classification and a little higher than 99% is obtained by all the pretrained models on multi-class MRI brain image classification.

Keywords: