Brain tumor classification for MR images using transfer learning and fine-tuning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Accurate brain tumor classification from MRI is crucial for patient treatment.
• A key challenge is the semantic gap between low-level image data and high-level interpretation.
• Traditional methods struggle with feature extraction, while deep learning requires large datasets.

Purpose of the Study:

• To address the challenge of training deep learning models on small medical imaging datasets.
• To propose a block-wise fine-tuning strategy based on transfer learning for brain tumor classification.
• To evaluate the proposed method on a T1-weighted contrast-enhanced MRI dataset.

Main Methods:

• Utilized a pre-trained deep convolutional neural network (CNN) model.
• Implemented a block-wise fine-tuning strategy leveraging transfer learning.
• Evaluated the method on a benchmark CE-MRI dataset using five-fold cross-validation.

Main Results:

• Achieved an average accuracy of 94.82% on the CE-MRI dataset.
• The proposed method requires minimal preprocessing and no handcrafted features.
• Outperformed traditional machine learning and existing deep learning CNN methods.

Conclusions:

• The proposed transfer learning strategy effectively classifies brain tumors from CE-MRI, even with limited data.
• This approach overcomes the limitations of training deep CNNs from scratch on small medical datasets.
• The method demonstrates superior performance compared to state-of-the-art techniques.