Artificial Intelligence-Based Methods to Characterize Kidney Macrostructure from Pre- and Post-Nephrectomy Computed Tomography Images

PROJECT SUMMARY/ABSTRACT The prevalence and burden of chronic kidney disease (CKD) is increasing worldwide. Patients who undergo radical or partial nephrectomy for kidney cancer resemble the general population with comorbidities, but with the difference that a majority undergo pre- and post-surgery abdominal imaging. Despite successful surgical

tumor removal, there is a concern for future progressive CKD. Following nephrectomy, the unaffected kidney undergoes compensatory hypertrophy, and the degree of hypertrophy and kidney function decline depend on the comorbidity burden and the amount of the removed kidney tissue. With the recent advances in artificial

intelligence (AI)--based quantification of kidney volumes from CT scans, there is an opportunity to evaluate automated imaging biomarkers as prognostic tools. There are also algorithms that quantify the number and volume of simple parenchymal cysts and many radiomic/texture features. The Co-Principal Investigators in this

program are uniquely equipped for the proposed studies. Dr. Denic has expertise in kidney micro- and macro- anatomy and advanced biostatistical skills. Dr. Kline has expertise in AI and developing advanced image processing techniques. The central hypothesis of this proposal is that macrostructural findings on

imaging of the retained (non-operated) kidney after radical or partial nephrectomy are prognostic for progressive CKD. In Aim 1, we will determine whether the degree of compensatory hypertrophy in the retained kidney after nephrectomy predicts progressive CKD. Using a recently created deep learning algorithm

we developed, we will quantify the kidney, cortex, and medullary volumes in pre-surgery and follow-up CT scans (at median 1-year post-surgery). From these volumes, we will calculate the degree of compensatory changes in kidney volumes and assess their association with baseline comorbidities and microstructural

measures. Finally, we will develop models to predict progressive CKD. In Aim 2, we will first optimize and finalize training of the model to quantify cysts and their size in CT images and develop postprocessing steps to separate cortical from medullary cysts. We will then develop models to assess whether the number and size of

cysts (overall, cortical, medullary) in the retained kidney in pre-surgery scans, and changes in number and size of cysts (overall, cortical, medullary) in the retained kidney over 1-year post-surgery, can predict progressive CKD. In Aim 3, we will determine whether novel radiological imaging texture features on pre-surgery scans are

reflective of microstructural measures of nephron size and nephrosclerosis and whether kidney texture features on follow-up CT scans predict progressive CKD. This research program will be facilitated by Mayo Clinic’s outstanding clinical and research environment at all three sites dedicated to improving patient care.

The goal is to develop a tool that can guide clinical decision-making in everyday practice, and that can help clinicians in improving their care of patients at an individual level by assessing the future risk of CKD.