Ultrasound image degradation in the human body arises from the propagation and reflection of acoustical waves in a complex acoustical environment. The heterogeneous distribution of soft tissue and the variation in acoustical properties distort the ultrasonic beam causing deterioration in image quality, including loss of resolution and contrast. Here we establish a framework to construct images based on a separable (additive or multiplicative) representation of aberration, multiple reverberation, and trailing reverberation. A separable approach enables high modularity and flexibility when generating quantitatively degraded image datasets. The framework provides the capability to generate images with quantitative levels of image degradation related directly to imaging physics, thus allowing for a flexible approach for augmentation techniques in ultrasound imaging data sets, as demonstrated in the included repository code. Experimentally calibrated abdominal simulations were performed in Fullwave2 by matching relevant imaging metrics such as phase aberration, reverberation strength, speckle brightness and coherence length, to experimental measurements. Then, simulations were performed to separate and characterize the different components of image degradation. Finally, these components were scaled and combined to construct quantitatively degraded image datasets. Reverberation is shown to be depth and brightness dependent, while aberration and trailing clutter are not. This general framework was tested for values in acoustical ranges that significantly, synthetically, and independently enhance or reduce these effects compared to levels naturally occurring in the body. Identifying, quantifying, and modeling these differing and complex mechanisms of degradation can be used to develop and test rational approaches to overcome these degradation mechanisms to improve image quality, particularly for traditionally harder to image patients. Additionally, the framework to synthetically modify the effects of aberration, multiple reverberation, and trailing clutter is provided, allowing for the generation of augmented datasets with a wide range of degradation effects, based on imaging physics, to improve machine learning models.
Soulioti et al. (Wed,) studied this question.