A Robust EfficientNet Architecture for Brain Tumor Classification and Identification Using MRI Image

The most common method for classifying brain tumors is by using biopsy, but this procedure is only carried out after surgery in which a tiny piece of brain tissue is takeout as well as analyzed under an electronic microscope to see a tumor. Since technology has advanced, particularly in ML and AI, we are now able to identify and categorize tumors without undergoing surgery by feeding the machine's classification engine the current data from various MRI images. By doing our study, we present a approach for the identification of tumors utilizing EfficientNet, an architecture that has already been trained utilizing the transfer learning methodology. This study emphasizes on the EfficientNet-B2, EfficientNet-B3, as well as EfficientNet-B4 algorithms from the group of EfficientNet approach. The suggested approach makes use of thresholds to enhance the dataset for more effective accuracy as well as data augmentation to improve the variety of brain images in the data in addition to the previously-trained approach to enhance the effectiveness of training an improved the model. Because it utilizes the compound coefficient to scale all three dimensions of an image with a fixed value of ratio, depth, width, and resolution—EfficientNet relatives of models outperform earlier CNN architecture, according to preliminary results. The outcomes also showed that scaling the previous approaches, enabled the model to capture complex features, improving the accuracy of the model.