Volume Visualization for Improving CT Lung Nodule Detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Detection

Background:

• Modern computer-aided detection (CAD) systems for CT lung nodules often use 2D or 3D Convolutional Neural Networks (CNNs).
• Detecting large pulmonary nodules and masses with 3D CNNs presents computational challenges.

Purpose of the Study:

• To investigate the use of volume-visualized CT thin-slab images with 2D CNNs to reduce computational complexity and enhance CAD performance.
• To evaluate different image visualization techniques for improved lung nodule and mass detection.

Main Methods:

• Tested four image types: original 2D CT, 2D projection of thin slabs, color channel mixture, and pixelwise maximum intensity mixture.
• Evaluated images on a dataset of 30 CT scans with 30 nodules/masses using GoogLeNet with transfer learning and cross-validation.

Main Results:

• 2D projection visualization alone showed better or equal area-under-curve scores for all nodule and mass sizes.
• The mixture of original CT and projection by maximum intensity yielded superior performance, achieving a 0.8 true positive rate and 0.046 false positive rate for large nodules/masses.

Conclusions:

• Visualizing CT thin-slab images, particularly using maximum intensity projection, can improve the performance of 2D CNN-based CAD systems for lung nodule and mass detection.
• This approach offers a computationally efficient alternative to 3D CNNs for detecting challenging large pulmonary lesions.