A Benchmark for automatic noise measurement in clinical computed tomography

• DOI: 10.1002/mp.14635
• Publication Date: 2020-12-07T09:56:39Z
• Authors: Moiz Ahmad; Megan C. Jacobsen; M. Allan Thomas; Henry S. Chen; Rick R. Layman; A. Kyle Jones
• Keywords: Image noise

Purpose Assessment of image quality directly in clinical data is an important control objective as phantom‐based testing does not fully represent across patient variation. Computer algorithms for automatically measuring noise computed tomography (CT) images have been introduced, but the accuracy these unclear. This work benchmarks global (GN) algorithm automatic measurement contrast‐enhanced abdomen CT exams comparison to precise reference measurements. The GN was further optimized compared previous report literature. Methods Reference values were established a public dataset 82 exams. obtained by manual regions‐of‐interest measurements pixel standard deviation liver parenchyma according instruction protocol. Noise taken six observers averaged together improve statistical precision. used measure each set. determined terms RMS error noise. conducting 1000 trials with random parameter values. trial lowest select optimum parameters. Results range sets 8.8–28.8 HU. made precision ±0.78 HU (95% confidence interval). 0.93 (0.77–1.19 95% equally accurate varying magnitude. Optimum were: kernel size 7 pixels, and soft tissue lower upper thresholds 0 170 HU, respectively. Conclusions performance benchmarked large dataset. study provides framework thorough validation methods. validated soft‐tissue