AG-MSTLN-EL: A Multi-source Transfer Learning Approach to Brain Tumor Detection

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Medicine
• Neuro-oncology

Background:

• Accurate brain tumor classification from MRI is crucial for diagnosis and treatment planning.
• Existing models face challenges in achieving high accuracy and interpretability.
• Medical image analysis requires robust computational models for early disease detection.

Purpose of the Study:

• To introduce AG-MSTLN-EL, an attention-aided multi-source transfer learning ensemble learning model.
• To enhance brain tumor classification accuracy and reliability using advanced AI techniques.
• To improve the interpretability of medical image analysis models for clinical applications.

Main Methods:

• Utilized multi-source transfer learning with Visual Geometry Group ResNet and GoogLeNet.
• Incorporated an attention mechanism to focus on critical MRI regions.
• Employed ensemble learning combining k-nearest neighbor, Softmax, and support vector machine classifiers.

Main Results:

• AG-MSTLN-EL demonstrated superior performance across all classification measures compared to state-of-the-art models.
• The model achieved robust and accurate classification on a dataset of 3064 brain tumor MRI images.
• The attention mechanism enhanced model interpretability and focused on relevant image features.

Conclusions:

• AG-MSTLN-EL offers a reliable and accurate solution for brain tumor classification.
• The integration of transfer learning, attention, and ensemble methods significantly improves classification outcomes.
• This model serves as a valuable tool for clinicians and researchers in medical image analysis and neuro-oncology.