Biwt-UNet: lung nodule segmentation via wavelet transform and multi-scale feature fusion

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Lung nodule segmentation is vital for early lung cancer detection.
• Traditional methods struggle with diverse nodule sizes, shapes, and blurred boundaries.
• Existing approaches require enhancement for improved accuracy and robustness.

Purpose of the Study:

• To develop an advanced hybrid approach for precise lung nodule segmentation.
• To address the limitations of traditional methods in segmenting complex lung nodules.
• To improve computer-aided diagnosis systems for lung cancer.

Main Methods:

• Proposed a hybrid lung nodule segmentation model named Biwt-UNet.
• Integrated improved Haar wavelet transform with multi-scale feature information.
• Utilized a multiscale dilated fusion module (MDFA) and bi-temporal fusion module (BFM).

Main Results:

• Biwt-UNet achieved excellent segmentation performance on the LIDC-IDRI dataset.
• Demonstrated superior results compared to state-of-the-art methods in Dice, IoU, and NSD metrics.
• Ablation studies confirmed the effectiveness of individual architectural components.

Conclusions:

• Biwt-UNet offers an effective and accurate solution for lung nodule segmentation.
• The model accurately segments nodules with varying sizes, shapes, and indistinct boundaries.
• This approach provides a new perspective for automatic lung nodule segmentation in CT analysis.