RNN-combined graph convolutional network with multi-feature fusion for tuberculosis cavity segmentation.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence in Medicine

Background:

• Tuberculosis (TB) is a prevalent global infectious disease.
• Accurate segmentation of TB cavities on computed tomography (CT) is crucial for assessing disease activity and guiding treatment.
• Challenges in TB cavity segmentation include blurred boundaries, irregular shapes, and similarity to other lung pathologies.

Purpose of the Study:

• To develop an advanced method for precise segmentation of tuberculosis cavities in CT images.
• To address the limitations of existing methods in handling complex imaging characteristics of TB cavities.

Main Methods:

• A novel Recurrent Neural Network (RNN)-combined Graph Convolutional Network (R2GCN) was proposed.
• The method integrates Bidirectional Recurrent Network (BRN) for contextual information and Graph Convolution Network (GCN) for topological feature extraction.
• Feature extraction was performed using VGG-16 or ResNet-50, followed by fusion of BRN and GCN features.

Main Results:

• The R2GCN method demonstrated superior segmentation accuracy compared to existing methods in experiments on patient CT images.
• The integrated approach effectively utilizes correlations and differences in feature maps for precise boundary delineation.
• The proposed method achieved the best segmentation accuracy among compared techniques.

Conclusions:

• The R2GCN method offers a robust solution for accurate tuberculosis cavity segmentation from CT scans.
• This technique holds significant potential for improving the diagnosis of TB cavities and evaluating treatment efficacy.
• The study highlights the effectiveness of combining RNN and GCN for complex medical image segmentation tasks.