Generation of realistic virtual nodules based on three-dimensional spatial resolution in lung computed tomography: A...

Area of Science:

• Medical Imaging
• Radiology
• Computational Imaging

Background:

• Accurate virtual nodule generation is crucial for advancing lung computed tomography (CT) imaging analysis.
• Existing methods may lack the realism and site-specific adaptability required for comprehensive CT research.

Purpose of the Study:

• To propose and validate a novel, feasible method for generating realistic virtual lung nodules using computer algorithms.
• To enable the creation of virtual nodules that accurately mimic the spatial resolution characteristics of specific CT scanners.

Main Methods:

• Pulmonary nodule images were deconvolved using the scanner's point spread function (PSF) and slice sensitivity profile (SSP) to create nodule-like object functions.
• Virtual nodules were generated by convolving these functions with the PSF and SSP of a target scanner.
• The method was validated using physical spheres and laboratory-made nodules of varying sizes, shapes, and densities across different CT scanners.

Main Results:

• High agreement was observed between virtual nodules and real images, with low root mean square errors (RMSEs) ranging from 6.2 to 16.5 Hounsfield units (HU).
• Virtual nodules generated from different scanners showed good consistency, validating the nodule-like object functions.
• The method accurately replicated the characteristics of physical nodules, including complex shapes and heterogeneous densities.

Conclusions:

• Nodule-like object functions derived from real nodule images are effective for generating realistic virtual lung nodules.
• The proposed method is feasible for creating site-specific virtual nodules, adaptable to the spatial resolution of any CT system.
• This approach facilitates the development of standardized and reproducible virtual lung nodule datasets for CT research.