Optimal Window Settings for Detection and Characterization of COVID19-Lung Opacities Using a Simplex Algorithm-based Approach

Purpose: To determine optimal window settings for characterization and detection of ground glass and dense consolidation on CT imaging of COVID19 lung infection using a Simplex-based approach.

Methods: Fourteen patients with PCR proven COVID19 infection and radiological lung appearance consistent with COVID19 pneumonia on conventional CT were identified and served as the reference population. Ten (71.5%) patients had only peripheral ground glass opacities (GGO), and 4 (28.5%) patients had denser consolidation. Representative images were evaluated separately by 7 radiologists in multiple different Hounsfield windows of varied width and center. Lung opacities were graded separately for adequacy of characterization and detection on a 5 point Likert scale. Simplex algorithm, an efficient mathematical tool for optimizing system parameters, was used to iteratively determine optimal window settings. Surface response maps expressing the relationship between window settings and overall reader grades were constructed.

Results: Twelve different window settings were evaluated over a total of 1,176 reads (12 window settings x 7 readers x 14 cases). Optimal characterization and detection of pure GGO was seen with a center ranging from -600 to -630 HU, and width 1400-1450HU (mean grade 4.5±0.4, and 4.3±0.4, respectively). Optimal windows produced a higher grade (3.8±0.3) for characterization than the manufacture window settings (-585HU/1,600HU; P=0.005). Optimal windowing for denser consolidation had a wider, over-lapping center (-400 to -630HU), but higher width (1,600-1,800HU) (3.6±0.9 characterization, 3.9±1 detection).

Conclusion: Characteristic COVID19 lung opacities can be best seen using CT windowing lower than standard manufacturer recommended windowing, possibly due the ground glass nature of this pathophysiology compared to denser consolidations.