SANet: A Slice-Aware Network for Pulmonary Nodule Detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading cause of cancer mortality globally.
• Early detection of pulmonary nodules via thoracic computed tomography (CT) is crucial for timely lung cancer diagnosis.
• Existing CT nodule datasets are limited in size and diversity, hindering diagnostic model development.

Purpose of the Study:

• To introduce the PN9 dataset, the largest and most diverse collection of pulmonary nodule CT scans to date.
• To propose a novel deep learning model, the slice-aware network (SANet), for enhanced pulmonary nodule detection.
• To evaluate the performance of SANet against existing methods using the PN9 dataset.

Main Methods:

• Development of the PN9 dataset comprising 8,798 CT scans with 40,439 annotated nodules across 9 classes.
• Introduction of SANet, featuring a slice grouped non-local (SGNL) module for capturing long-range dependencies within slice groups.
• Integration of a 3D region proposal network for candidate generation and a multi-scale feature-based false positive reduction (FPR) module.

Main Results:

• The PN9 dataset provides a substantial resource for pulmonary nodule detection research.
• SANet demonstrated effective pulmonary nodule detection capabilities.
• Experimental results validated SANet's performance compared to state-of-the-art 2D and 3D CNN-based methods on the PN9 dataset.

Conclusions:

• The PN9 dataset is a significant contribution to the field, enabling more robust pulmonary nodule detection model training.
• SANet represents an effective approach for pulmonary nodule detection, leveraging slice-aware processing and advanced deep learning modules.
• The developed methods and dataset hold promise for improving early lung cancer diagnosis through automated CT analysis.