Efficient Deep Learning Fusion-Based Approach for Brain Tumor Diagnosis

Technology has advanced to the point where it can influence every facet of human existence.Here, we look at how technology can help treat brain tumors, one of the most frequent malignancies and a leading cause of death.Many people lose their lives each year because of brain tumors.In the United States, roughly 85,000 new cases are diagnosed each year, bringing the total number of people with primary brain tumors to an estimated 700,000.Artificial intelligence has helped medicine and people overcome this challenge.Most brain cancers are detected via magnetic resonance imaging.Medical imaging and image processing make extensive use of magnetic resonance imaging for diagnosing anatomical differences.In this paper, we investigate the performance of various convolutional neural network (CNN) models like AlexNet, GoogleNet, VGGnet11, VGGnet13, VGGnet16, VGGnet19, ResNet18, ResNet34, ResNet50, ResNet101, ResNet152, DenseNet121, DenseNet161, DenseNet169, and DenseNet201 for brain tumor diagnosis tasks.On a dataset of 3264 MRI images, we perform experiments for healthy meningioma, glioma, and pituitary brain tumor classification.Our tests reveal that the ResNet and DenseNet models yield the highest accuracy (82%).Furthermore, we investigate the potential of a fusionbased approach where we test for different combinations of fusion of CNN models.The results show that fusing many CNN features improves accuracy even more.Classification accuracy is improved to 86% when ResNet50 and ResNet101 are fused and improves to 84% when DenseNet161 and DenseNet169 are fused.