An efficient multi-path 3D convolutional neural network for false-positive reduction of pulmonary nodule detection

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Pulmonary nodules in lung CT scans present diagnostic challenges due to similarities between true and false positives.
• Accurate nodule detection is crucial for early lung cancer diagnosis and treatment planning.

Purpose of the Study:

• To develop and evaluate a method for reducing false-positive pulmonary nodules in computer-aided diagnosis systems.
• To enhance the accuracy of pulmonary nodule detection in lung CT scans.

Main Methods:

• A 3D convolutional neural network (CNN) model with a hierarchical architecture was constructed to extract spatial nodule features.
• Three paths with varying receptive field sizes were incorporated and concatenated to fully extract and fuse feature information.
• The model was designed to adapt to diverse nodule shapes, sizes, and contextual information for effective false-positive reduction.

Main Results:

• The multi-path 3D CNN achieved a competitive performance metric score of 0.881 on the LUNA16 dataset.
• Excellent sensitivity was observed, with scores of 0.952 and 0.962 at 4 and 8 false positives per scan, respectively.
• The method demonstrated effective reduction of false-positive pulmonary nodules.

Conclusions:

• The developed multi-path 3D CNN accurately identifies pulmonary nodules by fully extracting candidate target features, accommodating variations in size, shape, and background.
• The proposed framework is versatile and applicable to other 3D medical image classification tasks.
• This approach significantly improves the reliability of computer-aided diagnosis systems for pulmonary nodule detection.