Pathological brain detection based on AlexNet and transfer learning

By Siyuan Lu, Zhihai Lu, Yudong Zhang. Journal of Computer Science, 2019

360 citations recorded by the source corpus.

Abstract The aim of this study is to automatically detect pathological brain in magnetic resonance images (MRI) based on deep learning structure and transfer learning. Deep learning is now the hottest topic both in academia and industry. However, the volume of brain MRI datasets are usually too small to train the entire deep learning structure. The training can be easily trapped into overfitting. Therefore, we introduced transfer learning to train the deep neural network. Firstly, we obtained the pre-trained AlexNet structure. Then, we replaced parameters of the last three layers with random weights and the rest parameters served as the initial values. Finally, we trained the modified model with our MRI dataset. Experiment results suggested that our method achieved accuracy of 100.00%, which outperformed state-of-the-art approaches.

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition, Neurology

View Paper: https://www.semanticscholar.org/paper/8ad27112e9c4a90d1fbaec9385d6676fb7f9a097

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

1. Yudong Zhang
2. Jian Wang
3. Shui-hua Wang

In Computer Vision and Pattern Recognition

1. Yudong Zhang
2. Jian Wang
3. Shui-hua Wang
4. Hengde Zhu

Citations in the corpus

• A Review of Deep Learning on Medical Image Analysis (2020, methodology): “[39] 2019 Pathological brain detection Fine-tuning on AlexNet”
• Accurate brain tumor detection using deep convolutional neural network (2022, background): “They are more biased towards classes with a larger number of samples [36][38][56].”
• Automatic Detection of Coronavirus Disease (COVID-19) in X-ray and CT Images: A Machine Learning Based Approach (2020, neutral): “Transfer learning strategies havevarious advantages, such as avoiding the overfitting issue when the number of training samples is limited, reducing the computational resources, and also speeding up the convergence of the network [39,40].”