Clinical evaluation of motion robust reconstruction using deep learning in lung CT.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Image Reconstruction

Background:

• Motion artifacts from cardiac and respiratory movements are prevalent in lung CT imaging.
• These artifacts can degrade image quality and hinder accurate diagnosis.
• CLEAR Motion is a novel deep learning-based reconstruction method designed to address these issues.

Purpose of the Study:

• To quantitatively assess the clinical utility of CLEAR Motion in lung CT.
• To evaluate the effectiveness of CLEAR Motion in reducing motion artifacts.
• To correlate image quality improvements with patient characteristics.

Main Methods:

• 129 lung CT scans were analyzed, with images reconstructed both with and without CLEAR Motion.
• Quantitative evaluation used Variance of Laplacian (VL) and Peak Signal-to-Noise Ratio (PSNR).
• Correlation analysis was performed between Body Mass Index (BMI), heart rate, and the difference in VL (DVL); visual assessments by radiologists were also conducted.

Main Results:

• CLEAR Motion generally resulted in higher VL values, indicating improved image quality.
• Significant improvements in image quality (large DVL) were predominantly observed in the lower lung regions.
• A positive correlation was found between BMI and DVL (r=0.55, p<0.05), while heart rate showed no significant impact.
• Radiological technologists preferred CLEAR Motion reconstructions in most cases (average preference score 0.96).

Conclusions:

• CLEAR Motion effectively reduces motion artifacts in lung CT imaging.
• The method demonstrates particular efficacy in the lower lung fields.
• CLEAR Motion offers a valuable tool for enhancing diagnostic accuracy in lung CT examinations.