A Deep Transfer Learning Approach for Automated Detection of Brain Tumor Through Magnetic Resonance Imaging

Nowadays, brain tumor is a fatal disease among humans. Early brain tumor detection is a significant challenge in the medical field to save a patient's life. This paper aims to present a unique segmentation and classification framework based on effective transfer learning. In the proposed framework, threshold and fast bounded box techniques are used for segmentation. The two pre-trained convolutional neural networks AlexNet and VGG-19 are used for classification using transfer learning. In pre-trained models, two transfer learning techniques stochastic gradient descent with restarts (SGDR) and fine-tune are applied through magnetic resonance images using Kaggle and Figshare datasets. The results demonstrated that the proposed transferred VGG-19 model achieved high accuracy of 99.75% and the transferred AlexNet model achieved 98.89% accuracy on the Kaggle repository. The transferred VGG-19 model achieved 98.50% accuracy and the transferred AlexNet model achieved 97.25% accuracy on the Figshare dataset. The outcomes demonstrated that the transferred VGG-19 model is obtained better performance than the AlexNet model. The presented models have accurately and efficiently segmented and classified brain tumor. Moreover, we have also compared our proposed models with existing state-of-the-art methods. The proposed transfer learning models provide competitive performance and can be successfully applied in clinical applications.

Automated summary

This study introduces a unique segmentation and classification framework for brain tumors based on transfer learning, achieving high accuracy in tumor classification with pre-trained convolutional neural networks.