Related Experiment Video
Updated: Dec 5, 2025
10:26
Author Spotlight: A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules
Published on: May 19, 2023
2.3K
Multi-level 3D Densenets for False-positive Reduction in Lung Nodule Detection Based on Chest Computed Tomography
Xiaoqi Lu1,2,3, Yu Gu2,3, Lidong Yang2
1College of Information Engineering, Inner Mongolia University of Technology, Hohhot, 010051, China
Current Medical Imaging
|October 21, 2020
Summary
This study introduces a novel multi-level 3D DenseNet framework to significantly reduce false-positive lung nodules in computer-aided detection (CADe) systems. The proposed scheme achieves satisfactory performance on the LUNA16 dataset, aiding radiologists in accurate lung nodule detection.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- False-positive reduction is critical for computer-aided detection (CADe) systems in lung nodule detection.
- Accurate identification of lung nodules assists radiologists in diagnosis.
Purpose of the Study:
- To propose a novel scheme for false-positive nodule reduction using a multi-level 3D DenseNet framework.
- To enhance the accuracy of lung nodule detection in CADe systems.
Main Methods:
- Utilized multi-level 3D DenseNet models to differentiate lung nodules from false positives.
- Employed 3D cubes of varying sizes for multi-level contextual information encoding.
- Applied image rotation and flipping for data augmentation of positive samples.
- Designed 3D DenseNets to preserve low-level nodule features for improved propagation.
- Combined model scores using an optimal weighted linear combination for classification.
Main Results:
- Evaluated the scheme on the LUNA16 dataset comprising 888 thin-slice CT scans.
- Validated performance using 10-fold cross-validation.
- Achieved satisfactory detection performance in the false-positive reduction track, as indicated by FROC curves and CPM scores.
Conclusions:
- The proposed multi-level 3D DenseNet scheme is significant for false-positive nodule reduction.
- This approach can improve the reliability of CADe systems for lung nodule detection.
