PDRF-Net: a progressive dense residual fusion network for COVID-19 lung CT image segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• COVID-19 diagnosis relies on identifying lung lesions in CT scans.
• Accurate segmentation of these lesions is crucial for patient monitoring and treatment.
• Existing segmentation methods may lack efficiency or precision.

Purpose of the Study:

• To develop a novel deep learning network for precise segmentation of COVID-19 lung lesions in CT images.
• To improve the accuracy and efficiency of lesion segmentation for COVID-19 diagnosis and monitoring.

Main Methods:

• Proposed a progressive dense residual fusion network (PDRF-Net) incorporating dense skip connections and an efficient aggregated residual module.
• Integrated a visual transformer and residual block for enhanced feature extraction.
• Introduced a bilateral channel pixel weighted module for progressive fusion of multi-branch feature maps.

Main Results:

• PDRF-Net achieved superior segmentation performance on two COVID-19 datasets.
• Demonstrated significant improvements over baseline methods (11.6% and 11.1%) and other mainstream approaches.
• The model proved effective in segmenting COVID-19 lung CT images.

Conclusions:

• PDRF-Net is a high-performance, easy-to-train deep learning model for COVID-19 lung CT segmentation.
• The proposed network offers a promising tool for clinical applications in COVID-19 diagnosis and management.