A brain tumor segmentation enhancement in MRI images using U-Net and transfer learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate brain tumor segmentation in Magnetic Resonance Imaging (MRI) is crucial for diagnosis and treatment planning.
• Existing segmentation methods face challenges with complex tumor structures and variations in imaging data.

Purpose of the Study:

• To develop and evaluate a novel transfer learning approach for enhanced brain tumor segmentation in MRI.
• To leverage a VGG19-based U-Net architecture for improved segmentation performance.

Main Methods:

• Utilized Fluid-Attenuated Inversion Recovery (FLAIR) abnormality segmentation masks and MRI scans from The Cancer Genome Atlas (TCGA) lower-grade glioma dataset.
• Implemented a U-Net architecture incorporating a VGG19 network with fixed pre-trained weights as the encoder.
• Employed transfer learning to enhance the model's ability to segment brain tumors.

Main Results:

• Achieved high performance metrics: Area Under the Curve (AUC) of 0.9957, F1-Score of 0.9679, Dice Coefficient of 0.9679, Precision of 0.9541, Recall of 0.9821, and Intersection-over-Union (IoU) of 0.9378.
• The VGG19-powered U-Net demonstrated superior performance compared to conventional U-Net models and other variants with different pre-trained backbones.
• Validated the effectiveness of the proposed transfer learning framework for brain tumor segmentation.

Conclusions:

• The proposed VGG19-based U-Net transfer learning approach offers a highly effective method for brain tumor segmentation in MRI.
• This framework shows significant potential for clinical applications in neuro-oncology, improving diagnostic accuracy and treatment planning.
• The study highlights the advantages of transfer learning with pre-trained VGG19 weights for medical image segmentation tasks.