Weighted Average Ensemble Learning Model for Improving Brain Tumor Classification

The detection of Brain Tumors (BTs) is of great benefit in identifying the growth of dangerous cells in the brain that pose a threat to the patient's life. Early detection of BTs remains a significant challenge due to subtle differences among tumor types and the complexity of brain anatomy. Manual diagnosis by radiologists is often time-consuming and subject to variations, which may delay treatment. Among available imaging techniques, Magnetic Resonance Imaging (MRI) is widely used because of its high-resolution capabilities. Deep Learning (DL) has achieved remarkable success in image classification tasks through Transfer Learning (TL), which utilizes pre-trained models. Therefore, this study proposes a new model, namely Weighted Average Ensemble Learning (WAEL), based on TL for BT classification. Convolutional Neural Network (CNN) architectures, including Visual Geometry Group (VGG16), Visual Geometry Group (VGG19), and Inception_v3, were employed as classifiers within the WAEL model for automatic BT prediction. The model was trained using the Figshare BT dataset, containing 3,064 images from three tumor classes (meningioma, glioma, and pituitary tumors). To enhance generalizability, benign (no-tumor) images were incorporated into the dataset. Experimental results demonstrated superior performance, with the WAEL model achieving an accuracy of 0.9908, compared to 0.9809, 0.9786, and 0.9832 for VGG16, VGG19, and Inception_v3, respectively. Furthermore, the proposed model achieved strong precision, recall, and F1-score values, confirming its robustness. The WAEL model also outperformed previous studies using the same dataset. Thus, the WAEL model can be a secondary tool for radiologists to discover tumors from brain MRI images.