DMCT-Net: dual modules convolution transformer network for head and neck tumor segmentation in PET/CT

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiotherapy Planning

Background:

• Accurate segmentation of head and neck (H&N) tumors in FDG-PET/CT images is crucial for effective radiotherapy.
• Existing segmentation methods struggle to integrate local/global information, semantic/contextual features, and spatial/channel data.

Purpose of the Study:

• To propose a novel Dual Modules Convolution Transformer Network (DMCT-Net) for enhanced H&N tumor segmentation.
• To improve the integration of diverse feature types for more accurate tumor delineation.

Main Methods:

• Developed DMCT-Net incorporating Convolution Transformer Blocks (CTB), Squeeze and Excitation (SE) pool module, and Multi-Attention Fusion (MAF) module.
• CTB captures remote dependencies and multi-scale receptive fields; SE pool extracts semantic and contextual features; MAF fuses global context, channel, and spatial information.
• Utilized up-sampling auxiliary paths to supplement multi-scale information.

Main Results:

• DMCT-Net demonstrated superior or competitive segmentation performance against advanced methods on three datasets.
• Achieved high segmentation metric scores: DSC of 0.781, HD95 of 3.044, precision of 0.798, and sensitivity of 0.857.
• Bimodal input (FDG-PET/CT) proved more effective than single modal input; ablation studies confirmed module significance.

Conclusions:

• The proposed DMCT-Net offers a novel and highly accurate approach for 3D H&N tumor segmentation in FDG-PET/CT images.
• The network effectively integrates local, global, semantic, contextual, spatial, and channel features for improved segmentation accuracy.
• This method holds significant potential for advancing radiotherapy planning and treatment for head and neck cancer patients.