Schlunk, Siegfried; Byram, Brett. (2024 五一茶馆儿).听.听Journal of Medical Imaging, 11(5), 057001.听
Evaluating the quality of ultrasound images has become more challenging as newer image-processing methods no longer meet the assumptions behind many traditional image quality metrics. Some advanced聽beamforming聽techniques (algorithms that improve image formation) can produce better scores on existing metrics without actually making images more useful for diagnosis. To address this issue, the authors developed聽generalized signal-to-noise ratio (gSNR), a new image quality metric based on the聽generalized contrast-to-noise ratio (gCNR), which measures how well a lesion can be distinguished from the surrounding tissue. Mathematical analyses showed that gSNR is equivalent to the traditional signal-to-noise ratio (SNR) for conventional聽delay-and-sum (DAS)聽beamforming but is more reliable for modern beamforming methods that do not follow the assumptions of older metrics. Tests using simulated lesions and real ultrasound images demonstrated that gSNR more accurately reflected how easily lesions could be detected. For example, while the traditional SNR suggested that some advanced beamforming methods or image transformations produced much better images, gSNR produced results that more closely matched the images’ actual visual quality. These findings suggest that gSNR provides a more robust and clinically meaningful measure of lesion detectability and may offer a better way to evaluate image quality in modern ultrasound systems.
Fig.听1
Comparing the analytic relationship to simulated lesion results for converting among CR,聽饾惗饾湏, and gCNR using Eqs.听(5) and (12) (a)聽and Eq.听(13) (b). The simulated lesions include varying radii (2 to 5聽mm) across a range of CRs, showing good agreement between the analytic equations and the simulated results. Note that simulations were performed at amplitude increments of 5聽dB, which produces the gaps in the figure. (c)聽Using Eq.听(1) and the Rose criterion threshold (3.54) to predict the threshold values of聽饾惗饾湏聽and gCNR. For a given value of聽饾憖聽(the number of correlation or resolution cells), having a measured聽饾惗饾湏聽or gCNR above the corresponding line indicates that the lesion is above the Rose criterion.