Brain Cancer Detection and Classification using Transfer Learning

Brain tumors are abnormal cells that grow inside the brain. If not diagnosed or treated early, such incidents can be fatal. Types of brain tumors include meningioma, glioma, and pituitary tumors. Magnetic resonance imaging (MRI) is an essential resource in diagnosis due to the clear expression of brain tissues, enabling professionals to analyze abnormal developments in the brain. The detection and classification of brain tumors using MRIs can be automated using machine learning techniques that can help to identify and diagnose brain tumors in their early stages, and subsequent treatment may result in a lower death rate. Transfer Learning allows the transfer of knowledge learned from large-scale datasets to other smaller datasets, thus avoiding wide-ranging training requirements with high accuracy and making the system converge much faster. This substantially minimizes computational resources and time in the training process. Transfer Learning also enables the effective MRI image abstraction of complex features, improved accuracy in tumor detection and classification, minimizes overfitting, especially with small amounts of medical data. In this work, we have used Transfer Learning with two pre-trained deep models, VGG16 and ResNet-50 on 3064 images available in the dataset to classify brain tumors. Fine-tuning for VGG16 and ResNet-50, we achieved validation accuracies of 98.85% and 97.72%, respectively. The results show that transfer learning can be used for the classification of brain tumors to enhance efficiency and precision for early detection and treatment.