Magnetic Resonance Imaging Image-Based Segmentation of Brain Tumor Using the Modified Transfer Learning Method.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate brain tumor segmentation is crucial for diagnosis and treatment planning.
• Existing segmentation methods face challenges with complex tumor structures and multimodal data.

Purpose of the Study:

• To improve the accuracy of brain tumor segmentation (BraTS) using deep learning.
• To apply a three-dimensional (3D) U-Net model with transfer learning for segmenting brain tumors on 3D MRI scans.

Main Methods:

• Utilized a 3D U-Net convolutional neural network architecture.
• Employed transfer learning by training the model on multimodal BraTS datasets (2018-2021).
• Processed 2240 studies with five MRI series (T1, contrast-enhanced-T1, Flair, T2, seg) in NIFTI format.

Main Results:

• Achieved high training accuracy (99.35%) and validation accuracy (98.93%).
• Obtained a mean Dice coefficient of 0.9875% and mean Intersection over Union (IoU) of 0.8738%.
• Demonstrated superior performance over existing tumor segmentation techniques.

Conclusions:

• The proposed 3D U-Net transfer learning method significantly enhances brain tumor segmentation accuracy.
• This approach offers a promising tool for improved neuro-oncological image analysis and clinical decision-making.