How Accurate Is in Vivo Noise Estimation From Patient CT Images? Validation of Methods Against Ensemble Noise Gold Standard
F Ria1,2,3,*, H Setiawan2,3, E Abadi2,3, J Solomon1,2,3,, E Samei1,2,3, (1) Clinical Imaging Physics Group Duke University Health Systems, Durham, NC, (2) RAI Labs Duke University, Durham, NC, (3) Center for Virtual Imaging Trials Duke University, Durham, NC
Presentations
MO-I430-BReP-F1-4 (Monday, 7/11/2022) 4:30 PM - 5:30 PM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 1
Purpose: One of the main indicators of CT image quality is noise magnitude. Inaccurate estimations can misguide protocol design and optimization, and negatively impact procedures and technology assessment. Several methods have been implemented to estimate noise magnitude in patient CT images, in vivo. Conclusive validation of such methods should ideally be based on ensemble noise values obtained from repeated patient images which is ethically infeasible. We deployed Virtual Imaging Trials (VITs) to obtain repeated CT images of virtual patients and test two different noise magnitude estimation across multiple of clinical scenarios and populations.
Methods: Using an established VIT platform (DukeSim), Chest and Abdominopelvic XCAT-phantoms were imaged 50 times each at three dose levels, and reconstructed with three kernels, using both FBP and IR algorithms, for a total of 1800 image datasets. Applying [-300,100]HU and HU<-900 thresholds, noise magnitudes were calculated in soft tissues (GN) and in the air surrounding the patient (AN) and compared with ensemble noise in soft tissue (EN) per each dose level, kernel, and reconstruction technique, for a total of 36 different comparisons.
Results: Across all image datasets and for FBP-reconstructed images, the median difference between GN and EN was 2% (min: -4%; max: 12%); and between AN and EN was -45% (min: -61%; max -35%). For IR-reconstructed images, the median difference between GN and EN was -3% (min: -12%; max: 9%); and between AN and EN was -49% (min: -60%; max -31%).
Conclusion: Noise magnitudes in the air surrounding the patient largely underestimated the ensemble noise and cannot represent noise magnitude in soft tissues. The presented results can enable the definition of adjustment factors to better represent image quality in vivo informing optimization actions’ design and effective technology assessment.
Funding Support, Disclosures, and Conflict of Interest: This work was funded in part by the Center for Virtual Imaging Trials, NIH/NIBIB P41-EB028744, and in part by NIH/NIBIB R44-EB031658-01.
Keywords
Taxonomy
IM- CT: Quality Control and Image Quality Assessment
Contact Email