The glomerular filtration rate (GFR) is the best measure of kidney function. And the single-kidney GFR is required for preoperative evaluation in patients with atrophic kidney or hydronephrosis. The split renal function (SRF) represents the relative contribution of single-kidney GFR to overall GFR and is important for decision-making in urological surgery. The single-photon emission computed tomography (SPECT) is the gold standard for measuring single-kidney GFR and SRF. Compared with SPECT, computed tomography (CT) is more commonly used due to its lower cost, faster imaging speed, and reduced radiation exposure. Previous investigations have used contrast-enhanced CT (CECT) to estimate renal function. However, the administration of iodine contrast media poses an elevated risk of nephrotoxicity. To avoid the use of iodinated contrast media, some studies found that the non-contrast CT based renal parenchymal volume (RPV) is associated with single-kidney GFR, and the percent RPV could be used to estimate SRF. However, the RPV alone is insufficient for a comprehensive assessment of the health status of renal tissues, resulting in moderate performance. Besides, in previous studies, the calculation of RPV relied on the delineation of the renal parenchymal region, which mainly achieved by labor-intensive manual segmentation, with limited potential for clinical application. Artificial intelligence (AI) methods, including deep learning and radiomics, play an important role in medical image processing. Using deep learning algorithms, organ segmentation can be accomplished in an efficient fully automated manner. Besides, converting medical images into quantitative features through radiomics has great potential to extract more representative disease-related image features. The renal parenchyma serves as the primary site for urine filtration, making its condition directly correlated with GFR. Furthermore, studies have indicated that increased severity of hydronephrosis is related to worse GFR. Therefore, we hypothesized that the radiomics features extracted from renal parenchyma and hydronephrosis regions could be used to estimate single-kidney GFR and SRF. In this study, we aimed to develop and validate an AI-based automatic estimation method of single-kidney GFR and SRF using non-contrast CT for patients with hydronephrosis and atrophic kidney. Additionally, we assessed the clinical utility of estimations in distinguishing between kidneys with varying health statuses.

245 patients with atrophic kidney or hydronephrosis were included from two centers (Training set: 128 patients from Center I; Test set: 117 patients from Center II). The renal parenchyma and hydronephrosis regions in non-contrast CT were automatically segmented by deep learning. Radiomic features were extracted and combined with clinical characteristics using multivariable linear regression (MLR) to obtain a radiomics-clinical-estimated GFR (rcGFR). The relative contribution of single-kidney rcGFR to overall rcGFR, the percent renal parenchymal volume, and the percent renal hydronephrosis volume were combined by MLR to generate the estimation of SRF (rcphSRF). The Pearson correlation coefficient (r), mean absolute error (MAE), and Lin's concordance coefficient (CCC) were calculated to evaluate the correlations, differences, and agreements between estimations and SPECT-based measurements, respectively.

Compared to manual segmentation, the deep learning-based automatic segmentation could reduce the average segmentation time by 434.6 times to 3.4 seconds. Compared to single-kidney GFR measured by SPECT, the rcGFR had a significant correlation of r=0.75 (p<0.001), MAE of 10.66 mL/min/1.73 m2, and CCC of 0.70. Compared to SRF measured by SPECT, the rcphSRF had a significant correlation of r=0.92 (p<0.001), MAE of 7.87 %, and CCC of 0.88.

For patients with atrophic kidney or hydronephrosis, the non-contrast CT and artificial intelligence methods can be used to estimate single-kidney glomerular filtration rate and split renal function, which may minimize the radiation risk, enhance diagnostic efficiency, and reduce costs.