| dc.description.abstract | Ultrasound imaging is a popular medical diagnostic tool for many applications due to its low cost and non-destructive nature compared to many alternative imaging methods. However, despite its popularity, image quality tends to suffer in high dynamic range scenarios and additional imaging, such as with computed tomography (CT), is often required to achieve a confident diagnosis. High dynamic range scenarios are situations where the imaging target and background have significantly different intensities, which results in artifacts and general image quality degradation. Targets at a lower intensity are referred to as hypoechoic and include general cysts and blood vessels. These often appear smaller than they should and are more susceptible to imaging artifacts and noise sources. In comparison, higher intensity targets are hyperechoic, such as kidney stones, and appear larger. This can complicate diagnosis when using ultrasound alone, as image quality is an important factor and accurate sizing is critical in many clinical applications. We have developed different methods for tackling these scenarios based on the model-based beamforming method ADMIRE, or aperture domain model image reconstruction.
In hypoechoic cases, image quality, and therefore artifact and noise reduction are crucial. However, many advanced denoising ultrasound beamformers are at risk of incorrectly removing diagnostic information that is weaker in amplitude compared to the noise and artifacts present in the image. For these cases, we developed iterative ADMIRE, the purpose of which is to denoise images while maintaining accurate amplitude measurements. We demonstrate that iterative ADMIRE removes sources of reverberation and sidelobe clutter without loss of diagnostic information in cyst simulations and in vivo blood vessels, which simultaneously improves contrast accuracy compared to alternative methods.
For hyperechoic cases, we consider the challenge of accurately sizing kidney stones. Ultrasound is notorious for overestimating stones, which results in the need for CT for accurate diagnosis in most cases. We consider two approaches for increasing sizing accuracy: combining ADMIRE with alternative beamforming methods such as minimum variance (MV) and tuning the model used in ADMIRE specifically for sizing applications. We show that ADMIRE and MV are strong complements for each other, as MV improves lateral resolution which improves sizing accuracy, while ADMIRE helps to denoise images improving the performance of MV. We demonstrate that the combination of these approaches successfully improves sizing accuracy in a preliminary study of patients suffering from kidney stone disease, improving diagnostic accuracy.
Finally, we investigate novel and popular image quality methods used with ultrasound images. Robust and accurate methods are important for our work, to ensure that our beamformer solutions are achieving our goals. Specifically, we consider the contrast ratio dynamic range (CRDR) for measuring contrast accuracy, generalized contrast-to-noise ratio (gCNR) for measuring target detectability, and histogram matching as method for generally creating fair comparisons between different beamformers. We discuss the value of these methods, as well as their strengths and weaknesses. | |
