Brain Tumor Detection Using Convolutional Neural Network

Abstract

Early and accurate diagnosis of brain tumors, a lethal disease caused by the abnormal growth of cells in the brain, is imperative to increase survival rates. A popular method for detection, diagnosis, and treatment is magnetic reasoning imaging (MRI) because it is non-invasive and provides high-quality visuals. Unfortunately, analyzing them manually can often be time-consuming and requires medical expertise. Image classification, a subset of computer vision, is a computer’s ability to classify and interpret objects within images. It can support a doctor’s diagnosis and serve as an entry-level screening system for brain tumors.

This study aims to build an accurate machine learning model to predict the existence of brain tumors from magnetic resonance images. We used the Br35H dataset to build two different convolutional neural network (CNN) models: Keras Sequential Model (KSM) and Image Augmentation Model (IAM). First, images from our dataset were preprocessed, augmented, and standardized to improve efficiency and reduce inaccuracies. Then, the data was normalized, and our models were trained. Lastly, aside from the validation accuracy and loss observed while training, we cross-referenced the accuracy of our model using the accuracy validation dataset. Of our two models, the IAM outperformed the KSM. The IAM had a validation accuracy of 97.99% and a validation loss of 4.94% on the Br35H dataset, and a 100% accuracy when classifying MRIs from the accuracy validation dataset.