EPIDSeg-Net: A Multi-Modal Fusion Framework Based on DRR Guidance in Radiotherapy is Used for Precise Segmentation of...

Area of Science:

• Medical Imaging
• Radiotherapy Technology
• Computational Biology

Background:

• Accurate tumor segmentation is crucial for effective radiotherapy delivery.
• Dynamic changes in tumor volume and shape during treatment require precise monitoring.
• Current methods may lack the precision needed for dynamic treatment plan adjustments.

Purpose of the Study:

• To develop a multimodal segmentation framework (EPIDSeg-Net) for precise lung tumor localization and morphological change quantification.
• To integrate Digitally Reconstructed Radiograph (DRR) and Electronic Portal Imaging Device (EPID) images for enhanced segmentation accuracy.
• To enable objective evaluation of treatment response and dynamic optimization of radiotherapy plans.

Main Methods:

• Proposed EPIDSeg-Net framework with a dual-branch encoder (CNN and Swin-Transformer) for local texture and global semantic feature extraction.
• Employed a Dual Attention Mechanism (DAM) for adaptive calibration of multimodal input features and improved tolerance to missing data.
• Integrated Large-Kernel Grouped Attention Gating (LKG-Gate) and Multi-Path Feature Extraction (MPFE) modules for enhanced contextual awareness and feature robustness.

Main Results:

• Achieved high-precision localization and sharp boundary delineation of lung tumors while preserving anatomical details.
• Demonstrated superior segmentation performance with DICE score of 93.2, IOU of 86.0, and HD95 of 9.42.
• Effectively preserved gradient information and regional integrity, reduced feature loss, and minimized missed segmentation rates.

Conclusions:

• The EPIDSeg-Net method successfully integrates EPID and DRR image information for enhanced lesion segmentation in radiotherapy.
• The framework enables more precise localization and segmentation of target regions, improving overall segmentation accuracy.
• This approach offers objective evidence for treatment response evaluation and dynamic radiotherapy plan optimization.