Automatic segmentation of brain glioma based on XY-Net.

Area of Science:

• * Medical image analysis
• * Artificial intelligence in healthcare
• * Neuro-oncology

Background:

• * Gliomas are aggressive brain tumors requiring timely diagnosis for effective treatment.
• * Magnetic Resonance Imaging (MRI) is crucial for glioma detection, but manual tumor segmentation is time-consuming.
• * Automated glioma segmentation from MRIs is essential for clinical decision-making.

Purpose of the Study:

• * To introduce XY-Net, a novel fully convolutional network for automated glioma segmentation in MRIs.
• * To enhance tumor outlining accuracy and efficiency compared to manual methods.
• * To provide a tool that assists physicians in patient assessment and treatment strategy development.

Main Methods:

• * Development of XY-Net, a U-Net based symmetric encoder-decoder network with dual sub-encoders and X-shaped feature transmission.
• * Implementation of a hybrid loss function combining balanced cross-entropy and Dice loss to address class imbalance.
• * Training and evaluation on MRI datasets for glioma segmentation.

Main Results:

• * XY-Net demonstrated a 2.16% improvement in Dice Coefficient (DC) over single-encoder models.
• * Achieved state-of-the-art performance with a DC of 74.49%, Hausdorff Distance (HD) of 10.89 mm, recall of 78.06%, and precision of 76.30%.
• * The novel architecture effectively performs end-to-end automatic glioma segmentation on 2D MRI slices.

Conclusions:

• * The proposed XY-Net architecture, with its dual sub-encoders and cross-transmission paths, significantly improves glioma segmentation accuracy.
• * XY-Net offers a valuable auxiliary tool for radiologists, enhancing the efficiency and precision of glioma diagnosis.
• * Automated segmentation using XY-Net can aid clinicians in better understanding patient conditions and formulating treatment plans.