Fully automated, fast, global noise level calculation for abdomino-pelvic adult CT images

Purpose The Global Noise Level (GNL) is an established objective CT noise metric evaluated directly on patient clinical images. A free, open-source, customisable tool to acceptance-test and validate commercial GNL software applications, monitor patient image noise, optimise protocols and which permits flexible input parameters for research purposes is lacking. This working developed such a tool and investigated the impact of various user-supplied input parameters on the GNL. Method The tool was developed in Python and followed standard published methods; it was validated against manual calculation. The impact of user-supplied input parameters on GNL values, namely the upper/lower CT-number threshold-fences, kernel size and shape (square and novel disc-shape) and histogram bin-widths, was studied by applying the tool to two abdomino-pelvic multi-centric, multi-system and multi-phasic image datasets (n = 357). Results The tool accurately calculated GNL (Bland-Altman plot of automatic vs manual calculation showed bias = 0 and standard deviation = 0) with a processing rate of 14 slices/s. GNL values changed significantly as a function of threshold-fence. GNL values increased monotonically with kernel size for both square and disc kernels with the latter consistently giving lower values. GNL values also increased with bin-width. Conclusions A free, fully-automated flexible tool for GNL calculation for adult abdomino-pelvic CT was developed. Since GNL values are dependent on choice of input parameters, these should therefore be reported alongside GNL values. Recommended parameters are threshold fences of − 300:0HU for adipose and 0:170HU for soft tissue (higher when contrast is present), 3 mm radius disc-kernel, 1.0HU bin-width.