Implementing a Novel Low Complexity CNN Model for Brain Tumor Detection

All over the world, thousands of people die yearly from brain tumor disease. It is considered a life-threatening, prominent, and serious disease. Thus, saving people-life requires automated detection and well-timed analysis of brain tumors. Accurate detection is the first and most important stage in diagnosing and treating brain tumors. Recently, the deep learning approach has been widely used in the medical sector, especially in diagnosing brain tumors. This paper introduces a novel low-complexity CNN model for detecting brain tumors using MRI images. A freely available dataset from Kaggle is employed. Initially, to select the baseline model, four famous CNN models, involving AlexNet, VGG-16, GoogLeNet, and ResNet-50, are examined and trained from scratch. The AlexNet model showed the best performance and got the highest accuracy of 98.8%. After improving it by inserting different layers, the proposed model reached an accuracy of 99.4%. To attain a low-complexity model, the final model has twenty-two layers, including nine convolutional blocks and four dropout layers. Other performance metrics include 99.67% precision, 99.02% recall, and 99.35% F1 score.