Joint Brain Tumor Segmentation from Multi-magnetic Resonance Sequences through a Deep Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Manual brain tumor segmentation is time-consuming and subjective.
• Automated segmentation improves diagnostic consistency and treatment planning.
• Deep learning offers a promising approach for automated brain tumor delineation.

Purpose of the Study:

• To automate brain tumor segmentation using deep learning.
• To evaluate the accuracy of individual and combined MRI sequences (FLAIR, T1W, T2W, T1ce) for segmentation.
• To identify the optimal MRI sequence or combination for accurate brain tumor delineation.

Main Methods:

• Utilized the BraTS-2020 dataset with 370 subjects and four MRI sequences.
• Trained a residual neural network for segmentation.
• Assessed models using single-channel (individual sequences) and multi-channel (combined sequences) inputs.

Main Results:

• FLAIR sequence achieved the highest accuracy (Dice index 0.77 ± 0.10) among single sequences.
• Combining FLAIR and T2W sequences improved accuracy (Dice index 0.80 ± 0.10).
• Segmenting all four sequences yielded the highest overall accuracy (Dice index 0.82 ± 0.09).

Conclusions:

• FLAIR is the optimal single MRI sequence for brain tumor segmentation.
• Combining all four MRI sequences (FLAIR, T1W, T2W, T1ce) provides the highest accuracy for tumor delineation.