Challenges on Deep Learning Models to Improve Magnetic Resonance Imaging Brain Tumor Images Classification: A Review

Brain tumor is one of the mortal cancers in the world. A brain tumor can be diagnosed by invasive or non-invasive approaches. Nowadays, magnetic resonance imaging or MRI is the most non-invasive approach to do anatomical neuroimaging. Brain tumor diagnosis using deep learning models has robust improvement in classification. This review provides the recent research topics in last the five years. There are some developments in deep learning models: augmentation, architecture modification, detection, segmentation, and ensemble. Augmentation helps generating more data to train a model, architecture modification adapts to solve feature extraction and selection, and detection or localization assists feature extractor with segmented images. Those methods sometimes work with image processing on MRI images, to boost feature extraction process. According to literature review, architecture modification is the most popular approach. After all, the main purpose of all models is improving classification accuracy. By classifying many deep learning methods and datasets, both help to analyze challenge and opportunity on brain tumor classification based deep learning methods. In conclusion, the best start to the research in MRI brain tumor classification using deep learning is developing a model. Hopefully, this paper can be a brief summary for novice researchers.