Automatic segmentation of MRI images for brain radiotherapy planning using deep ensemble learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiotherapy

Background:

• Accurate segmentation of brain magnetic resonance imaging (MRI) is crucial for radiotherapy (RT) planning.
• Existing individual convolutional neural network (CNN) models show limitations in segmentation accuracy.

Purpose of the Study:

• To develop and evaluate an efficient method for automatic segmentation of T1- and T2-weighted brain MRI.
• To compare the segmentation performance of individual CNN models against an ensemble approach.

Main Methods:

• An ensemble deep learning (EDL) strategy integrating five independently trained 2D CNNs was developed.
• The EDL model averaged class probabilities from individual networks using a weighted-average method.
• Segmentation performance was evaluated using Dice Similarity Coefficient (DSC) and Hausdorff distance at 95% (HD95) on clinical and public datasets.

Main Results:

• The EDL model achieved superior segmentation performance on both clinical (DSC: 0.7 ± 0.2, HD95: 4.5 ± 2.5 mm) and public datasets.
• The ensemble approach significantly outperformed individual CNNs (DSC ≤0.6, HD95 ≥14 mm).

Conclusions:

• Ensemble learning, specifically the EDL model, consistently enhances segmentation accuracy over individual CNNs.
• The EDL approach shows significant potential for improving clinical applications, particularly in MRI-guided RT planning.