Advanced ultrasound methods to improve chronic kidney disease diagnosis

Abstract

Chronic kidney disease (CKD) affects 850 million people worldwide and is associated with significant cardiovascular morbidity and mortality. Routine laboratory tests do not reflect early stages of microcirculatory changes and vascular rarefaction that characterise kidney fibrosis, the common endpoint of CKD. Imaging techniques that detect CKD in early stages could promote timely treatment with new drugs like SGLT2 inhibitors, thus, decreasing CKD progression and the cardiovascular disease burden. Ultrasound is the most used imaging modality in CKD, as it is non-invasive and radiation free. Initially, ultrasound imaging was applied to assess kidney macro-morphology and to rule out ureteral obstruction. The development of higher frequency probes allowed for more detailed imaging of kidney parenchyma, and advances in Doppler ultrasound provided insights into segmental arterial flow patterns including resistive indices as an indirect measure of microcirculatory impedance, elevated values of which correlated with progressive organ failure and fibrosis. Today, low-flow detection methods and matrix probes better resolve organ parenchyma and smaller vascular beds, and contrast-enhanced ultrasound allows perfusion measurement. Particularly, super-resolution ultrasound imaging, a technology currently being in clinical translation, can characterise the microcirculation morphologically and functionally in unrivalled detail. This is accompanied by rapid developments in radiomics and machine learning supporting ultrasound image acquisition and processing, as well as lesion detection and characterisation. This perspective article introduces emerging ultrasound methods for the diagnosis of CKD and discusses how the promising technical and analytical advancements can improve disease management after successful translation to clinical application.

Fig. 1. CKD grading, risk and measurement blind spot.

A Longitudinal sections of healthy and CKD mouse kidney. B Prognosis of CKD by GFR and Albuminuria Categories: KDIGO 2012. C Hyperbolic relationship of serum creatinine and eGFR.

Fig. 2. Evolution of Doppler mode in detection of smaller vascular beds.

A Colour Doppler, (B) advanced dynamic flow, (C) SMI imaging. Insets: Cortex area and degree of measured vasculature with flow.

Fig. 3. Workflow for ultrasound localisation microscopy.

A B-mode images of a contrast bolus are acquired along with either contrast-enhanced imaging or as shown here with a slow-flow Doppler mode (SMI, Canon) to better detect microbubbles. B The image of tissue background is used for motion estimation and compensation. Here motion estimation was performed with the function imregdemons provided by Matlab (Mathworks, Natick, MA, USA). C Microbubbles are detected and localised in the contrast image. D Tracking of microbubbles and visualizing the tracks results in the images of super-resolved vessel. Here, an image of microbubble counts is shown. Images are from a study at RWTH Aachen University with ethical approval by the local review board.

Fig. 4. Super-resolution ultrasound microscopy images of the mouse kidney.

A B-mode, (B) maximum intensity persistence image, (C) microbubble counts, (D) microbubble velocities.

Fig. 5. AI-based decision modelling workflow.

Ultrasound data acquired by established (B-mode, colour Doppler, SMI) and emerging (ULM) ultrasound methods can be used for building AI-based decision models. First a region of interest (ROI) is selected. This is usually done by segmenting the image either automatically with e.g., deep learning or manually by delineating the ROI. Then, features are extracted from the selected ROI. Here, the features can be calculated automatically (deep learning) or by using predefined radiomics features (histogram, texture and wavelet). In the final step, the most descriptive features are selected and used to build a decision model that facilitates the diagnosis of a disease from the image.