Room: AAPM ePoster Library
Purpose: Assessment of image quality directly in clinical image data is increasingly important as phantom-based testing cannot represent image quality across all patients. A global noise index (GNI) algorithm has been previously proposed for automatic measurement of noise in clinical CT exams. This work optimizes the GNI algorithm and further validates the algorithm in a large dataset of contrast-enhanced abdomen CT exams.
Methods: The global noise index (GNI) algorithm was used to automatically measure noise in a publicly-available dataset of 82 contrast-enhanced abdomen CT exams of healthy control subjects conducted at the NIH Clinical Center. The ground truth noise value was obtained by manual measurements using multiple regions-of-interest placed in the liver parenchyma by three different observers. The accuracy of the GNI algorithm was determined in terms of RMS error compared to manual measurements. The GNI algorithm was optimized by conducting 500 trials with random algorithm parameter values of kernel size and soft tissue masks. For each trial, the root-mean-square (RMS) error was computed in comparison to the noise ground truth. The trial with the lowest RMS error was used to select the optimized GNI algorithm parameters.
Results: The range of ground truth noise across CT exams was 8.8 – 28.8 HU. The inter-observer variation was 0.8 HU. The accuracy of the automatic GNI algorithm was 5.8 HU using previously published parameters, and 0.99 HU using optimized parameters. The most sensitive parameters were 1) kernel size, and 2) upper threshold value of the soft-tissue mask. The optimal kernel size is 9 pixels. The optimal high threshold value of the soft tissue mask is 125 – 150 HU.
Conclusion: GNI algorithm was optimized and the performance was benchmarked in a large clinical dataset. The GNI algorithm measures noise within 1 HU accuracy in abdomen CT exams regardless of the noise magnitude.