A Benchmark for automatic noise measurement in clinical computed tomography

Area of Science:

• Medical Imaging
• Radiology
• Image Quality Assessment

Background:

• Clinical image quality assessment is crucial, as phantom-based testing has limitations.
• Automatic noise measurement algorithms for computed tomography (CT) require validation against clinical data.
• The Global Noise (GN) algorithm offers automated noise assessment in CT images.

Purpose of the Study:

• To benchmark the accuracy of the Global Noise (GN) algorithm for automatic noise measurement in contrast-enhanced abdomen CT exams.
• To optimize the GN algorithm's parameters for improved performance.
• To compare automated noise measurements with precise reference measurements from clinical data.

Main Methods:

• Established reference noise values using manual region-of-interest measurements in liver parenchyma from 82 CT exams.
• Averaged noise measurements from six observers to enhance reference precision.
• Applied the GN algorithm for automated noise measurement and determined accuracy via Root Mean Square (RMS) error.
• Optimized GN algorithm parameters through 1000 trials with random values.

Main Results:

• The noise range in CT image sets was 8.8–28.8 HU, with reference measurements precise to ±0.78 HU.
• The GN algorithm achieved an RMS error of 0.93 HU for automatic noise measurement.
• Accuracy remained consistent across varying noise levels in image sets.
• Optimized GN parameters: kernel size 7 pixels, soft tissue thresholds 0–170 HU.

Conclusions:

• The GN algorithm's performance was validated in a large clinical CT dataset.
• This study establishes a framework for validating automated clinical image quality measurement tools.
• The optimized GN algorithm is suitable for automatic soft-tissue noise measurement in abdomen CT exams.