DBPNDNet: dual-branch networks using 3DCNN toward pulmonary nodule detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Pulmonary nodule detection is challenging due to visual similarities with anatomical structures and noise.
• Accurate detection and segmentation of pulmonary nodules are crucial for early lung cancer diagnosis.

Purpose of the Study:

• To develop an efficient deep learning model for automated pulmonary nodule detection and segmentation.
• To improve the accuracy and efficiency of computer-aided diagnosis systems for lung nodules.

Main Methods:

• Proposed DBPNDNet, a dual-branch 3D convolutional neural network architecture.
• One branch for nodule candidate region extraction, another for semantic segmentation.
• Incorporated a 3D attention-weighted feature fusion module to enhance detection using segmentation features.

Main Results:

• Achieved 91.33% sensitivity with an average of 1 FP per CT scan.
• Reached 97.14% sensitivity with 8 FPs per scan.
• Demonstrated superior performance compared to existing mainstream approaches.

Conclusions:

• DBPNDNet effectively automates pulmonary nodule detection and segmentation.
• The dual-branch architecture with attention-based fusion enhances diagnostic accuracy.
• The framework shows significant potential for clinical application in lung cancer screening.