Enhancing brain tumor MRI classification with an ensemble of deep learning models and transformer integration

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Brain tumors are a leading cause of cancer mortality worldwide.
• Early and accurate detection is crucial for improving patient survival rates.
• Manual analysis of MRI data for brain tumor identification is time-consuming and challenging.

Purpose of the Study:

• To develop a precise and efficient deep learning methodology for brain tumor diagnosis.
• To leverage transfer learning and transformer encoder mechanisms for enhanced diagnostic accuracy.
• To address the limitations of manual MRI analysis in early brain tumor detection.

Main Methods:

• A hybrid deep learning model integrating transfer learning and a transformer encoder was developed.
• An ensemble of three pre-trained models (DenseNet201, GoogleNet, InceptionResNetV2) was used for feature extraction.
• The transformer encoder incorporated a Shifted Window-based Self-Attention mechanism and a multilayer perceptron.

Main Results:

• The hybrid model achieved high accuracy across three public datasets: 99.34% (Cheng), 99.16% (BT-large-2c), and 98.62% (BT-large-4c).
• The proposed model demonstrated superior performance compared to existing techniques.
• Consistent results were observed across datasets with varying sample numbers, planes, and contrasts.

Conclusions:

• The hybrid deep learning model offers a dependable solution for precise brain tumor diagnosis.
• This approach significantly improves upon current methods for early tumor detection and classification.
• The findings highlight the potential of advanced AI in reducing cancer-related mortality.