Detection and Classification of Brain Tumor using Deep Transfer Learning Approach

Abstract

Abstract : Precise and timely recognition of brain tumor is crucial for treatment planning to prolong the life span of the patient. Over the past decades many research efforts have been focused on classification and detection of brain tumors. However, it is a daunting research problem due to the tumor being different in shape, size, and location. Magnetic Resonance Imaging (MRI) is a commonly prescribed diagnostic test for diagnosing brain tumors. This study investigates the use of EfficientNetV2S, a pre-trained model, to diagnose brain tumors. Furthermore, the placement and order of adding layers in the top final classifier was also studied. Data augmentation was incorporated to expand the dataset size. Finally fine-tuning was applied on the best performing model in the previous experiment to improve the performance and accuracy of the model. The tumors were categorized into four classes: glioma, meningioma, pituitary, and no-tumor. The final classification layer included a dense layer along with the SoftMax activation function. It was concluded that the highest performing model was the fine-tuned EfficientNetV2S model with no pooling layer and batch normalization layer added after the dense layer. The proposed model was able to score 99.98% in training f1-score, while scoring 99.47% in testing fl-score. It was also concluded that the placement and order in which layers are in the final classifier has an impact on the model performance. Keywords: brain tumor classification, EfficientNetV2S, MRI, BraTS, transfer learning, fine- tuning