MD-NDNet: a multi-dimensional convolutional neural network for false-positive reduction in pulmonary nodule detection

Area of Science:

• Medical imaging analysis
• Artificial intelligence in radiology
• Computer-aided diagnosis

Background:

• Pulmonary nodule detection in low-dose computed tomography (LDCT) lung cancer screening faces challenges with false positives.
• Individual nodule variations and similarities to soft tissues complicate accurate detection.
• Current methods can be time-consuming and carry risks.

Purpose of the Study:

• To develop an automated system for reducing false positives in pulmonary nodule detection.
• To enhance the accuracy and efficiency of lung cancer screening using LDCT.
• To propose a deep convolutional neural network (DCNN) based approach for nodule detection.

Main Methods:

• A multi-dimensional nodule detection network (MD-NDNet) integrating 3D CNNs and 2D CNNs with an attention module was developed.
• The network extracts volumetric and spatial correlation features from multiple planes (sagittal, coronal, axial).
• A multi-scale ensemble strategy was employed for probability aggregation to handle diverse nodule sizes and shapes.

Main Results:

• The MD-NDNet achieved a classification performance with a CPM score of 0.9008 on the LUNA16 dataset.
• The method demonstrated effective reduction of false positives in pulmonary nodule detection.
• Ten-fold cross-validation confirmed the robustness and reliability of the proposed framework.

Conclusions:

• The proposed MD-NDNet offers an efficient and accurate solution for pulmonary nodule detection.
• This deep learning approach significantly improves false-positive reduction in LDCT analysis.
• The method holds promise for enhancing clinical diagnosis in lung cancer screening programs.