Author: Lavsen Dahal, Francesco Ria, Ehsan Samei, Justin B. Solomon, Liesbeth Vancoillie, Yakun Zhang 👨🔬
Affiliation: Duke University, Carilion Clinic, Clinical Imaging Physics Group, Department of Radiology, Duke University Health System 🌍
Purpose: Clinical diagnostic task-based optimization of CT procedures require precise and organ-specific assessments. This study investigates inter-organ noise variability to highlight the limitations of global image quality measures and thus leading to optimized organ-specific imaging protocols.
Methods: A dataset of 3133 clinical CT image series from two vendors (A, B) was analyzed. The dataset included Chest Abdomen Pelvis exams and Abdomen Pelvis exams from one large academic medical center. The DICOM images were converted to NIfTI format for segmentation. Organ masks for large organs—lungs, liver, spleen, stomach, pancreas, kidneys, urinary bladder, and small bowel—were generated using the validated segmentation tool "TotalSegmentator" (https://doi.org/10.1148/ryai.230024). Organ noise properties, global noise indices, Hounsfield Unit (HU) values, and patient water-equivalent diameters were calculated using an in-house pre-developed patient image quality analysis program. Coefficient of variation (COV) was used to quantify inter-organ noise variability for each image series. Noise profiles were compared across vendors, models, slice thicknesses, reconstruction methods and patient sizes.
Results: The COV increased with slice thickness for both Vendor A and Vendor B, although the magnitude of increase differed. For Vendor A, COV was highest in medium-sized patients, while for Vendor B, it was highest in small-sized patients. Organs with similar HU values, such as the liver and spleen, exhibited consistent noise changes across different patient sizes and slice thicknesses for both vendors. Conversely, organs with divergent HU values displayed distinct noise behaviors depending on slice thickness and patient size. Notably, IR sometimes reduced COV despite increased slice thickness.
Conclusion:
The application of iterative reconstruction affects noise smoothing differently across organs for one vendor. Additionally, inter-organ noise variability does not always increase with patient size, challenging assumptions about uniform noise behavior. These findings underscore the need for organ-specific protocols to achieve optimal diagnostic image quality.