Automatic pulmonary ground-glass opacity nodules detection and classification based on 3D neural network

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Pulmonary nodule detection

Background:

• Pulmonary ground-glass opacity (GGO) nodules present diagnostic challenges due to indistinct boundaries.
• Accurate detection and classification of GGO nodules are crucial for timely cancer diagnosis.

Purpose of the Study:

• To develop an automated deep learning framework for detecting and classifying pulmonary GGO nodules.
• To improve the accuracy and efficiency of GGO nodule diagnosis in clinical settings.

Main Methods:

• A two-stage framework utilizing a pretrained 3D U-Net for lung parenchyma extraction.
• Adaptation of Mask region-based convolutional neural networks (RCNN) for 3D medical image analysis.
• Implementation of a class-balanced loss function and a feature-based weighted clustering (FWC) scheme for enhanced accuracy.

Main Results:

• The detection phase achieved a mean average precision of 0.5182.
• The feature-based weighted clustering (FWC) algorithm effectively controlled false positives, reaching a competition performance metric (CPM) of 0.817.
• Comparative analyses demonstrated the proposed method's effectiveness against other deep learning approaches.

Conclusions:

• An automated deep learning framework for pulmonary GGO nodule detection and classification has been developed.
• The proposed method offers accurate localization and classification of nodules, serving as a valuable tool for clinicians.