[Multi-scale 3D convolutional neural network-based segmentation of head and neck organs at risk]

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiotherapy Planning

Background:

• Accurate segmentation of organs at risk (OARs) in head and neck CT images is crucial for effective radiotherapy.
• Existing segmentation methods often face challenges in accuracy and efficiency.

Purpose of the Study:

• To develop and validate a novel 3D convolution neural network algorithm for automated segmentation of head and neck OARs.
• To enhance the performance of deep learning models for medical image segmentation.

Main Methods:

• A V-Net based 3D convolution neural network incorporating squeeze-and-excitation (SE) and residual convolution modules was developed.
• A multi-scale strategy with two cascade models was employed for location and fine segmentation.
• Input images were preprocessed with multi-resolution resampling to optimize feature extraction.

Main Results:

• The proposed algorithm demonstrated superior segmentation accuracy, achieving a 9% average improvement over existing methods.
• Segmentation efficiency was significantly enhanced, reducing average test time from 33.82s to 2.79s.
• Experiments were conducted on the segmentation of 22 OARs in the head and neck region.

Conclusions:

• The developed 3D convolution neural network with a multi-scale strategy effectively improves OAR segmentation accuracy and efficiency.
• This algorithm shows potential for clinical application in head and neck cancer treatment planning and could be adapted for other organs.