Organ-At-Risk Segmentation in Brain MRI using Model-Based Segmentation: Benefits of Deep Learning-Based Boundary...

Area of Science:

• Medical imaging analysis
• Radiotherapy
• Computational anatomy

Background:

• Organ-at-risk (OAR) segmentation is crucial for accurate radiotherapy planning.
• Model-based segmentation (MBS) offers robustness to noise using shape priors.
• Integrating neural networks with MBS can enhance segmentation performance.

Purpose of the Study:

• To investigate the benefits of combining convolutional neural networks (CNNs) with model-based segmentation (MBS) for organs-at-risk (OARs).
• To evaluate CNN-based boundary detectors against traditional methods for brain radiotherapy on MRI.
• To assess the impact on segmentation accuracy for various brain structures.

Main Methods:

• Trained CNN boundary detectors using classic strong gradients and a locally adaptive regression task.
• Developed a hybrid approach combining CNNs with MBS for OAR segmentation.
• Evaluated methods using 5-fold cross-validation on T1w and T2w brain MRI from 16 cancer patients.

Main Results:

• CNN-based boundary detectors significantly improved segmentation results for all structures on T1w images (p < 0.05).
• Improvements were also observed on T2w images, with statistical significance for the auditory system.
• The hybrid approach demonstrated enhanced accuracy in OAR segmentation.

Conclusions:

• Combining CNNs with MBS provides superior OAR segmentation for brain radiotherapy planning on MRI.
• This enhanced segmentation accuracy can lead to more precise radiation delivery.
• The findings support the integration of deep learning with traditional MBS techniques.