A Generalized SNR to Quantify Lesion Detectability for Modern Adaptive beamformers

It is well known that the enveloped data after delay-and-sum beamforming of an ultrasound image follows a Rayleigh distribution. Many image quality metrics, such as signal-to-noise ratio (SNR) proposed by Smith et al., are designed specifically with these statistics in mind. However, many modern adaptive beamformers or post-processing techniques produce images where the data no longer follows a Rayleigh distribution. This makes the applicability of those image metrics questionable. Metrics like the generalized contrast-to-noise ratio (gCNR) have been proposed explicitly for this reason, as these metrics are designed to be transform-invariant, and measure the information of the image more accurately. We propose to apply gCNR and other robust techniques to create a generalized version of Smith et al.’s SNR, which will likewise be robust to transformations but will also retain much of the subjective clinical context that SNR was designed around.