Estimating the minimum SNR necessary for object detection in the projection domain

Area of Science:

• Medical Imaging
• Computational Imaging
• Signal Processing

Background:

• The Rose criterion (5 standard deviations above background) is a traditional but limited rule for object detectability in computed tomography (CT).
• Advanced denoising algorithms, including those using convolutional neural networks, face inherent limitations imposed by sinogram data statistics.
• There is a need for a more robust method to define object detectability in the projection domain, especially with modern CT imaging techniques.

Purpose of the Study:

• To estimate the minimum signal-to-noise ratio (SNR) required for detecting objects within a defined set in the projection domain of CT data.
• To establish a new metric, the projection SNR, based on ideal observer principles for signal detection.
• To compare the derived projection SNR requirements with the established Rose criterion.

Main Methods:

• Utilized an ideal observer model to sequentially compare objects against a null hypothesis in the projection domain.
• Reduced the detection problem to a one-dimensional signal detection scenario between two Gaussian distributions.
• Employed simulations to determine the minimum projection SNR for achieving 80% sensitivity and 80% specificity in a specific CT imaging scenario.

Main Results:

• Defined a 'projection SNR' as a measure of signal detectability in the CT projection data.
• Simulations indicated a minimum projection SNR of 5.1 is necessary for detecting a 6 mm circular lesion in a 60 mm x 60 mm region across 10 slices.
• The calculated minimum projection SNR is comparable to the Rose criterion but derived from different theoretical underpinnings.

Conclusions:

• The study provides a statistically grounded method for estimating object detectability in CT projection data, moving beyond the heuristic Rose criterion.
• The derived projection SNR offers a more precise threshold for evaluating the detectability of lesions or objects in CT imaging.
• This work has implications for optimizing CT acquisition parameters and evaluating the performance of image reconstruction and denoising algorithms.