Detecting brain tumors can be a time-consuming and error-prone process, heavily reliant on the expertise of radiologists. With the increasing number of patients, traditional methods have become inefficient and costly. Digital medical images have emerged as crucial tools, not only for diagnosis but also for training and research purposes. The demand for electronic medical images is soaring. Manual evaluation of these images is slow, inaccurate, and susceptible to mistakes. Brain tumors pose a significant health threat, ranking among the top causes of death in India. Researchers have been exploring various algorithms, with a focus on Deep Learning (DL), to improve the speed and accuracy of brain tumor detection and classification. DL techniques, particularly Convolutional Neural Networks (CNNs), have gained popularity for automating the diagnosis and segmentation of brain tumors. The proposed Brain Tumor Classification Model based on CNN (BCM-CNN) optimizes CNN hyperparameters, including those defining the network structure and those involved in training the network.
In this study, our primary goal was to develop a CNN model capable of classifying MRI images into two categories: malignant brain tumors and healthy brains with no tumors. We utilized a publicly available dataset from Kaggle, consisting of contrast-enhanced MRI images. Our CNN model achieved impressive results on various subsets of the dataset, with the following accuracy metrics on the testing data:
Test Loss: 0.39058661460876465 Test Accuracy: 0.9032257795333862
These results highlight the robust generalization and high speed of our proposed network. Consequently, our model could serve as an effective decision-support tool for radiologists in medical diagnostics.