MTRFU-Net: a lung nodule segmentation model based on improved U-Net architecture with spatial-frequency fusion

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate pulmonary nodule segmentation is crucial for early lung cancer diagnosis.
• Existing deep learning models face challenges with nodule heterogeneity, blurred boundaries, and multi-scale variations.

Purpose of the Study:

• To propose MTRFU-Net, an improved U-Net based model for enhanced pulmonary nodule segmentation.
• To integrate spatial-frequency features and multi-module collaboration for dynamic feature fusion.

Main Methods:

• Developed MTRFU-Net with a ResNet50 encoder featuring a Spatial-Frequency Fusion (SFF) module.
• Integrated a Transformer encoder with an optimized Atrous Spatial Pyramid Pooling (ASPP) module in the bottleneck.
• Employed residual connections and a dynamically weighted scSE attention mechanism in the decoder.

Main Results:

• MTRFU-Net demonstrated excellent performance on the LIDC-IDRI dataset.
• Achieved high Dice Similarity Coefficient (DSC) and mean Intersection over Union (mIoU) scores.
• Validated the effectiveness of frequency-domain information in segmentation tasks.

Conclusions:

• MTRFU-Net offers a robust solution for pulmonary nodule segmentation.
• The integration of spatial and frequency domain features significantly improves segmentation accuracy.
• Provides a valuable reference for developing clinically applicable segmentation models.