Optimizing Organ-Specific Noise Measurement In Clinical CT Using a Global Noise Index Framework
Abstract
Purpose
Image noise is a fundamental determinant of CT image quality and directly impacts diagnostic performance. Conventional noise metrics are typically global and do not account for spatial or organ-specific variations introduced by modern CT techniques such as tube current modulation and iterative reconstruction. This study extends the Global Noise Index (GNI) framework to develop and optimize an automated Organ Noise Index (ONI) for multiple organs in clinical body CT and validates the approach against manually measured reference noise.
Methods
Sixteen contrast-enhanced adult body CT exams were retrospectively analyzed. Organ segmentation for lungs, liver, spleen, kidneys, and urinary bladder was performed using TotalSegmentator. Manual noise measurements were obtained by five medical physicists using a custom MATLAB GUI. Automated noise measurement employed a GNI-based sliding-window approach constrained by organ masks. Three key parameters—ROI size, HU thresholding limits, and histogram bin width—were systematically varied. ONI measurements were compared to reference noise across parameter combinations, and optimal values were determined by minimizing root mean square error (RMSE).
Results
Manual noise measurements showed greater variability for heterogeneous organs compared to more homogeneous organs. Optimized ROI sizes were 5 pixels for liver, spleen, and urinary bladder, and 3 pixels for lungs and kidneys. Optimal histogram bin widths were 1 HU (1.5–1.6 HU) for heterogeneous organs. Minimum RMSE values ranged from 0.43 HU (liver) to 0.76 HU (kidneys). Bland–Altman analysis demonstrated good agreement between ONI and reference noise, with ONI confidence intervals generally narrower than manual measurements for most organs.
Conclusion
The proposed ONI framework enables robust, organ-specific noise quantification in clinical CT and shows strong agreement with expert manual measurements. This approach provides a practical tool for organ-level image quality assessment and has potential applications in protocol optimization and quality control.