Automated objective determination of the noise power spectrum (NPS) for quality control of CT scanners

Abstract

Background: Existing Noise Power Spectrum (NPS) calculation tools face limitations such as inaccessible source code, lack of automation, platform dependency, inefficiency in handling large image volumes, and inadequate API support. These limitations hinder transparency, flexibility, usability, and scalability in NPS analysis, impacting fields reliant on precise image characterization. Objectives: The primary aim of this research is to develop a software tool for determining noise standard deviation in terms of the Noise Power Spectrum (NPS) for CT images. The NPS provides a comprehensive characterization of noise texture and intensity as a function of spatial frequency. The study focuses on creating an objective, repeatable tool that eliminates observer variability and integrates seamlessly into automated quality control workflows. Research Methodology: The research employs a cross-sectional observational design, combining quantitative and qualitative approaches to evaluate the NPS of CT images using the Siemens water phantom. A sample of 150 CT images was collected to ensure comprehensive NPS evaluation. Python code was developed for NPS evaluation, aiming for a flexible framework that can accommodate future image quality analysis metrics. Results: The study found that the B46 kernel achieved the highest frequency and detail with the lowest noise among the three kernels evaluated. Increasing the number of sample images significantly improved the accuracy of NPS calculation. Recommendations: For optimal results, it is recommended to take multiple scans and use multiple sub-ROIs (e.g., 9) when working out the NPS curve. Future Recommendations: Future research should replicate this study using different phantom materials, such as polyethylene or acrylic, and consider anthropomorphic phantoms for more comprehensive results. Investigating the impact of iterative reconstruction techniques on NPS and developing standardized protocols for NPS calculation are also suggested. Additionally, exploring the use of artificial intelligence to differentiate between noise and actual signals, and to develop adaptive noise filtering techniques, could significantly enhance noise reduction and image quality.