© Springer Nature Singapore Pte Ltd. 2019.Misdiagnosis of brain tumor types will prevent effective response to medical intervention and decrease the chance of survival among patients. One conventional method to differentiate brain tumors is by inspecting the MRI images of the patient’s brain. For large amount of data and different specific types of brain tumors, this method is time consuming and prone to human errors. In this study, we attempted to train a Convolutional Neural Network (CNN) to recognize the three most common types of brain tumors, i.e. the Glioma, Meningioma, and Pituitary. We implemented the simplest possible architecture of CNN; i.e. one each of convolution, max-pooling, and flattening layers, followed by a full connection from one hidden layer. The CNN was trained on a brain tumor dataset consisting of 3064 T-1 weighted CE-MRI images publicly available via figshare Cheng (Brain Tumor Dataset, 2017 [1]). Using our simple architecture and without any prior region-based segmentation, we could achieve a training accuracy of 98.51% and validation accuracy of 84.19% at best. These figures are comparable to the performance of more complicated region-based segmentation algorithms, which accuracies ranged between 71.39 and 94.68% on identical dataset Cheng (Brain Tumor Dataset, 2017 [1], Cheng et al. (PLoS One 11, 2017 [2]).

Brain tumor classifications,Convolutional neural network,Convolutional Neural Networks (CNN),Medical intervention,Overfitting,Region-based segmentation,Training accuracy,Validation accuracy

Overfitting,Training accuracy,Training loss,Validation accuracy,Validation loss

https://doi.org/10.1007/978-981-10-9035-6_33