CT Image Conversion among Different Reconstruction Kernels without a Sinogram by Using a Convolutional Neural...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Different reconstruction kernels in CT imaging can affect image quality and subsequent analysis.
• Existing methods for kernel conversion may require sinogram data, limiting their applicability.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) for converting CT images between reconstruction kernels without sinogram data.
• To assess the accuracy and speed of the proposed CNN-based kernel conversion method.
• To evaluate the clinical utility of the converted images, particularly for emphysema quantification.

Main Methods:

• A CNN architecture with six convolutional layers was designed and trained on chest CT scans reconstructed with B10f, B30f, B50f, and B70f kernels.
• Quantitative evaluation used root mean square error (RMSE) to measure conversion accuracy.
• Clinical validation involved converting images from B30f to B50f kernels and assessing emphysema quantification using Bland-Altman plots.

Main Results:

• The CNN achieved rapid image conversion at 0.065 seconds per slice.
• A significant reduction in RMSE was observed in converted images, with a mean reduction of 65.7%.
• Emphysema quantification showed good agreement between original and converted images, with narrow limits of agreement.

Conclusions:

• CNN-based CT kernel conversion demonstrates high accuracy and speed.
• The method is a promising tool for clinical applications in medical imaging.
• This technique offers a sinogram-free approach to kernel conversion, enhancing its practical utility.