Image Quality Comparison of Post-Processed True Axial Head CT Reconstructions with Direct Reconstructions on Scanner
Abstract
Purpose
To investigate the impact of post-processed true axial reconstructions of axial head CT scans.
Methods
40 head CT exams with 3mm original axial series (recon-O) reconstructed from the scanner (Siemens) under IRB approval, were collected from PACS. Each reconstruction was post-processed for true axial (recon-P) on Syngo.via (Siemens) for comparison. For each reconstruction, global noise was calculated using a kernel size of 7x7 pixels, a soft tissue mask upper threshold of 100HU, and a histogram bin width of 0.1HU. The global noise of each series was reported as the median of all slices. Additionally, fourteen dual energy exams without contrast, thirteen single energy exams without contrast, and thirteen single energy exams with contrast underwent manual true axial reconstructions (recon-M) to determine if manual reconstructions outperformed recon-P.
Results
The noise of recon-P (3.3±.4HU) is statistically (P<0.001, paired-T-Test) higher than recon-O (3.2±.4HU). The noise of recon-P (4.2±2.1HU) is statistically (P<1e-6, paired-T-Test) higher than recon-M (3.9±2.0HU). When selecting dual energy systems, the noise of recon-P (3.3±.4HU) is statistically (P<0.001, paired-T-Test) higher than recon-O (3.1±.3HU) and (P<0.005, paired-T-Test) recon-M (3.0±.3HU). The noise difference between recon-P and recon-O does not correlate with pixel spacing (linear fit R-Squared<.05) or rotation angle (linear fit R-Squared<.05).
Conclusion
This work indicates that the global noise of recon-P is statistically larger than recon-O and recon-M, even when using dual energy systems. Additionally, the global noise in recon-P does not correlate with pixel spacing or rotation angle. Further investigations will focus on gathering more data on the effect of contrast on reconstruction methods and selecting exams with greater slice overlap.