Comparative Analysis of Transfer Learning Based Deep Learning Models for Brain Tumor Detection

Detecting and classifying brain tumors is essential in medical imaging. Accurate and early diagnosis significantly impacts treatment outcomes. Deep learning (DL) models provide automated and highly accurate diagnostic tools. In this paper, we have used publicly available MRI brain tumor dataset from Kaggle. It evaluates the performance of various advanced DL architectures to classify brain tumors. We used several pre-trained deep learning models, including Inception ResNet V2, Inception V3, ResNet 50, VGG 16, Xception, and VGG 19. Transfer learning was employed to fine-tune these models on the Kaggle MRI brain tumor dataset. Each model was trained and validated to assess its classification accuracy. Transfer learning allowed the models to utilize pre-existing knowledge, enhancing their performance in this medical imaging task. The validation accuracies achieved were as follows: Inception ResNet V2 at 62.39%, Inception V3 at 74.56%, ResNet 50 at 81.34%, VGG 16 at 89.83%, Xception at 70.67%, and VGG 19 at 92.37%. Among these, VGG 19 showed the highest accuracy. It demonstrated superior capability in distinguishing between different types of brain tumors, with an improvement of 6.93% over other models.