Residual Neural Network for Brain Tumor Classification Based on MRI Images

abstract-Brain tumors are a dangerous disease that causes quite a lot of deaths in Indonesia and early detection plays a crucial role in improving patient outcomes. In this research, the author proposed the model based on the transfer learning ResNet50 architecture and developed a new architecture named Simple Residual Network to classify the MRI images that contain 4 types of MRI with a total 7023 MRI images that collected from Kaggle. Three metrics that are observed in this research are accuracy, model size, and training time. The model can accurately distinguish the glioma, meningioma, pituitary tumors, and no tumor images. The experimental results show that the accuracy of the transfer learning model achieve a high accuracy (99.7%), and the model created using the scratch method maintaining the high accuracy (98.1%) with model size is 32 times smaller (9.2 MB) than the model obtained from the transfer learning method (294.5 MB), and the training time took 76.7% faster than the transfer learning method.