Techniques for virtual lung nodule insertion: volumetric and morphometric comparison of projection-based and...

Area of Science:

• Medical Imaging
• Radiology
• Computational Imaging

Background:

• Standardized databases of hybrid CT images with known lesions are crucial for developing robust lung nodule detection algorithms.
• Virtual nodule insertion offers a promising approach to generate such datasets by creating artificial lesions within real CT scans.

Purpose of the Study:

• To evaluate three virtual lung nodule insertion techniques (one established, two new) for their accuracy in replicating nodule volume and shape.
• To compare the CT-derived characteristics of physically inserted synthetic nodules with those generated by virtual insertion methods.

Main Methods:

• Physically inserted 24 synthetic lung nodules into a chest phantom, imaged using a commercial CT scanner at two dose levels.
• Reconstructed CT data and created 3D virtual nodule models, then inserted them into nodule-free images using projection-based (Technique A), image-based (Technique B), and region-cropping (Technique C) methods.
• Quantified nodule volume using segmentation and deformation using Hausdorff distance, comparing results between CT-derived and virtual nodules via linear mixed effects regression.

Main Results:

• Virtual nodule volumes closely matched CT-derived volumes, with percent differences under 3% across all techniques and no statistical significance in most cases.
• Correlation coefficients between CT-derived and virtual nodule volumes exceeded 0.97.
• Hausdorff distances indicated similar deformation patterns between CT-derived and virtual nodules, with minimal statistical significance for all tested techniques.

Conclusions:

• Both projection-based and image-based virtual nodule insertion techniques produce realistic renderings with statistical similarity to synthetic nodules in terms of volume and deformation.
• These validated techniques can be utilized to construct hybrid CT image databases featuring nodules with precisely known characteristics (size, location, morphology).