A mathematical algorithm for quantification of CT image noise |
| |
Authors: | Edmund K. Kerut MD Filip To PhD Michael Turner MD James McKinnie MD Thomas Giles MD |
| |
Affiliation: | 1. Heart Clinic of Louisiana, Marrero, Louisiana;2. Department of Agricultural and Biological Engineering, School of Engineering, Mississippi State University, Mississippi State, Mississippi;3. Cardiovascular Specialists of Southwest Louisiana, Lake Charles, Louisiana;4. Department of Cardiology, West Jefferson Medical Center, Marrero, Louisiana;5. School of Medicine, Tulane University, New Orleans, Louisiana |
| |
Abstract: | Quantification of computed tomography (CT) noise helps in determination of radiation dosage requirements for adequate image quality. Clinical methods used include calculation of the standard deviation (SD) of a selected region of interest (ROI). In industry, wavelet decomposition has been used for image compression while removing high‐frequency noise. We evaluated a cohort of 74 consecutive patients referred for coronary artery calcium scoring and quantitated noise within a 16×16 ROI in the ascending aorta using the traditional SD method and also using a two‐dimensional dyadic wavelet decomposition method. Clinically, noise has been shown to be proportional to patient weight and also body mass index (BMI), which is a derived value from height and weight. Noise for both methods was plotted against patient parameters of height, weight, waist circumference and calculated BMI. A regression line was calculated and coefficient of determination (CoD) calculated for each. The CoD was better for height, weight, and waist circumference using the wavelet method as compared to the traditional SD method. The wavelet method of quantification of image noise may be an improved method as compared to the SD method. This method could help further refine an imaging system's determination of radiation dosage requirements to obtain a satisfactory quality image. |
| |
Keywords: | blur contrast quality standard deviation texture wavelet |
|
|