Whole Tumor Segmentation from Brain MR images using Multi-view 2D Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Brain tumor segmentation is crucial for diagnosis and treatment planning.
• Accurate segmentation of gliomas from MRI data remains a challenge.
• Existing methods often struggle with variations in tumor appearance and location.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated brain tumor segmentation.
• To assess the performance of a U-Net based 2D CNN on the BraTS 2019 dataset.
• To investigate the effectiveness of fusing predictions from orthogonal planes for improved segmentation accuracy.

Main Methods:

• Utilized the BraTS 2019 dataset comprising 259 high-grade glioma (HGG) and 76 low-grade glioma (LGG) patient data with four MR modalities.
• Employed a U-Net architecture based 2D convolutional neural network (CNN) applied to sagittal, coronal, and axial planes.
• Fused predictions from orthogonal planes and minimized Dice loss; incorporated test-time augmentation and 7-fold cross-validation.

Main Results:

• The model demonstrated strong performance in segmenting brain tumors, achieving high sensitivity, specificity, accuracy, and Dice scores.
• Training was conducted on 222 HGG samples, with testing on 37 HGG samples.
• Test-time augmentation and cross-validation indicated robust segmentation capabilities.

Conclusions:

• The proposed U-Net based CNN approach effectively segments brain tumors, specifically gliomas, from multimodal MRI data.
• Fusion of predictions from orthogonal planes enhances segmentation accuracy.
• The method shows promise for clinical application in neuro-oncology.