Multi-Class Brain Tumor Detection Using CNN-Based Medical Imaging Analysis

The detection of brain tumours is a long-term persistent challenge in medical imaging where accurate and efficient diagnostic tools remain an absolute necessity for clinicians to detect in early stages and to determine the appropriate treatment. Rapid developments in DL have opened new avenues to automate and improve the accuracy of brain tumour classification from MRI scans. This study proposes to develop a robust Convolutional Neural Network (CNN) based classification method for various types of brain tumours, including glioma, meningioma, pituitary tumours, and healthy brain tissue. For this purpose, the paper utilises the following deep learning models: DenseNet169, ResNet50, VGG16, and VGG19. The dataset used was comprehensive, with labelled MRI images-70% for training and 30% for testing data. Accuracies, precisions, recalls, and F1 scores are used to assess model performance. DenseNet169 proved to be the most effective model, with an accuracy of 97.45<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>. ResNet50 came in second at 92.34%, followed by VGG16 at 88.65% and VGG19 at 90.23%. The confusion matrices also show that in the case of each model, DenseNet169 tended to reduce the number of false positives, particularly for glioma and pituitary tumours. These results demonstrate a great possibility of deep learning to develop automatisation for accurate high-throughput brain tumour detection. Since DenseNet169 has superior results, it implies usability in real-time clinical applications that can improve the whole process of diagnosis, which would increase early detection or treatment for the majority of patients with brain tumours.