Automatic Segmentation of MRI of Brain Tumor Using Deep Convolutional Network

Area of Science:

• Medical Image Processing
• Artificial Intelligence in Medicine
• Neuroimaging Analysis

Background:

• Multimodal magnetic resonance imaging (MRI) brain tumor segmentation is crucial for diagnosis and treatment.
• Traditional algorithms struggle with grayscale similarity in MRI, hindering accurate segmentation.
• Deep learning offers a solution by leveraging complementary information across MRI modalities.

Purpose of the Study:

• To develop a deep learning model for accurate multimodal MRI brain tumor segmentation.
• To address challenges posed by grayscale similarity and category imbalance in brain tumor datasets.
• To enhance the segmentation of tumor core and substructure areas.

Main Methods:

• A fully convolutional neural network (FCNN) framework was adapted for brain tumor segmentation.
• An end-to-end training approach was employed using 2D MRI slices.
• The Dice loss function and a proposed parallel Dice loss were utilized to handle category imbalance and improve substructure segmentation.

Main Results:

• The proposed cascaded FCNN model achieved good prediction results on the BraTS 2017 dataset.
• The Dice loss function effectively addressed category imbalance issues.
• Parallel Dice loss demonstrated improved segmentation of tumor substructures.

Conclusions:

• The developed cascaded deep learning model significantly enhances multimodal MRI brain tumor segmentation accuracy.
• The proposed methods offer a robust solution for segmenting tumor core and substructure areas.
• This approach holds promise for improving computer-aided diagnosis and treatment planning in neuro-oncology.