Technical Development and Validation of Synthetic Images in Multiphasic CT Examinations of the Kidneys - PROJECT SUMMARY CT urography is a three-phase CT technique used clinically in the evaluation of blood in the urine, known as hematuria. Each of the three phases in CT urography—the non-contrast, nephrographic, and urographic phases—provide unique information that aid in diagnosing the various causes of hematuria. Unfortunately, three- phase CT urography requires approximately ~2x the examination time and 3x the radiation dose of a standard single-phase CT. Previously developed methods for improving the CT urography technique have lower diagnostic accuracy, are more challenging to interpret, and/or have no savings in examination time. Our group has developed an innovative and novel technique termed SNICT (synthetic nephrographic phase images in CT). The SNICT technique uses image processing to reconstruct the nephrographic phase images from the non-contrast and urographic phases. Our preliminary results with a transformer-based deep learning SNICT technique have shown high fidelity in synthesizing these nephrographic phase images. As a result, the nephrographic phase acquisition could be eliminated—effectively reducing the CT urography acquisition from a three-phase to a two-phase study and reducing radiation dose by 33%. In this R01 project, we seek to further improve and validate the SNICT technique. We will expand upon our SNICT technique by developing a diffusion-based deep learning model (Aim 1A), which will provide further improvements in image quality. Furthermore, we will develop a SNICT technique with dual-energy (DE) CT urography studies that can provide additional benefits, including a 2x reduction in overall examination time and a radiation dose reduction of 66% (Aim 1B). Moreover, we will perform a rigorous clinical validation of the SNICT technique (Aim 2), in which the diagnostic accuracy of the reconstructed images will be assessed and the savings in examination time will be quantified. Our criteria for success will be to show that synthesized nephrographic phase images are (a) of equivalent image quality compared to ground truth images, (b) are of equivalent diagnostic quality compared to ground truth images, and (c) provide significantly reduced examination time compared to the standard image acquisition. These criteria will ensure that the SNICT technique is clinically actionable as a robust and reliable image reconstruction technique in CT urography examinations that ultimately provides up to a 66% reduction in radiation dose and 2x reduction in examination time.
