The objective of this study was to investigate the influence of kidney volume measured at the initial visit on subsequent eGFR decline in individuals with mild eGFR reduction (G2 category). We found that the adjusted HR associated with low kidney volume for a progression of eGFR categories was 1.64 (95% CI: 1.09–2.45, p = 0.02), indicating that low kidney volume was a significant risk factor even after adjusting for eGFR values per se and other known risk factors.

eGFR is known to decrease for various reasons, and early interventions for high-risk groups with declining kidney function may contribute to improved patient outcomes by preventing disease progression [1, 5, 17, 18]. While previous studies have identified various factors influencing CKD progression such as sociodemographic, behavioral, genetic, cardiovascular, and metabolic factors [5, 16, 19], our findings add a new dimension to this established list by suggesting the potential of imaging factors such as kidney volume, as prognostic indicators for eGFR decline. Several studies have investigated factors associated with kidney volume in the general population. Kidney volume has been weakly to moderately correlated with factors such as sex, height, weight, BMI, and age [7, 13, 14], and there is a potential association between low birth weight and this measure in adulthood [20]. However, the relationship between kidney volume and longitudinal health outcomes remains poorly understood. Furthermore, while marked atrophy of the kidneys in end-stage CKD is well-known, little attention has been paid to the imaging evaluation of the intermediate conditions leading up to it. This highlights a gap in research on the progression of kidney volume changes and their impact on health over time. Our central research question was whether a relatively low kidney volume is a mere manifestation of functional decline or an indication of future renal outcomes. To address this question, our study employed PS matching to ensure a balanced comparison of risk factors for eGFR decline between the LKV and control groups. After matching, neither group showed significant differences in variables, indicating controlled variations in baseline characteristics. The LKV group had a consistently higher cumulative incidence curve than its matched control group, and this difference was statistically significant according to the log-rank test (p = 0.03). Upon conducting a multivariate Cox regression analysis, it was evident that being in the LKV group (HR: 1.64, 95%CI: 1.09–2.45) and having an eGFR value ranging 60–69 (HR: 56.27, 95%CI: 7.81–405.35) were significant risk factors for migration to G3. Following this, as a direction for future research, we believe that investigating the effects of early and intensive interventions or follow-ups for such high-risk groups for CKD progression could lead to the accumulation of insights that are beneficial for the preservation of renal function.

The significance of knowing the volume of areas of interest in medical imaging has been demonstrated in various contexts: volume of kidneys as highlighted in this study; relationship between Alzheimer’s disease and hippocampus volume [21]; correlation between liver cirrhosis and liver volume [22]; and detailed evaluations of tumors [23, 24] and aneurysms [25, 26]. Volumetric analysis offers a more precise three-dimensional assessment than traditional linear measurements [27]. Conventionally, the analysis has been labor-intensive, requiring manual delineation of a certain workstation/software for each individual case, making large-scale analysis and evaluation challenging. However, deep-learning-based segmentation is now streamlining this process [28, 29]. By training or utilizing appropriate deep-learning models, we can efficiently and swiftly extract areas of interest from medical images, circumventing the cost and time of manual methods. Our study is the largest to date to use CT volumetry of the kidney for analytical research [7, 10, 13, 14, 20, 30, 31]. In addition to volumetry, the extraction of regions of interest from medical images yields diverse quantitative data such as CT density in CT scans, signal intensity in MRI, standardized uptake value (SUV) in PET, and even radiomic features [32]. This technology facilitates large-scale image analysis and is likely to be increasingly integrated with epidemiological research, aiding in the generation of high-quality evidence.

This study has several limitations. First, it was based on a single ethnic group from one institution. These participants underwent fee-based checkups, possibly indicating a health-conscious cohort. Therefore, the generalizability of these results is limited. Second, a comprehensive verification of the segmentation results for all the cases was not conducted. In particular, CT scans were performed with the arms down, and streak artifacts may have affected the quality of renal imaging and segmentation results. However, given the similarity in kidney volumetry results with those of prior studies [7, 31, 33, 34] and their near-normal distribution, we posit that the overall impact of segmentation errors is not significant. As examples, CT images of kidney slices of three randomly selected cases used in the analysis (cases segmented by the deep-learning model) and their segmentation results are attached in Additional file 1: Appendix 3. Third, our study overlooked the impact of small lesions. Those with a maximum diameter of up to 4 cm were included; if assumed to be spherical, this represents an ignored volume increase of up to 33.5 mL. It cannot be definitively stated that this has no impact on the results. Furthermore, as a future direction, if a system capable of segmenting both lesions and normal renal tissue were simultaneously developed allowing for the measurement of “volume of normal part of the kidney,” this could potentially yield more detailed and novel insights. Fourth, certain background factors may not have been adjusted for, and the presence of unmeasured confounding factors that strongly influence kidney volume and eGFR cannot be ruled out.

In conclusion, low kidney volume defined as below the sex-specific mean minus 1 standard deviation emerged as a significant predictor of a progression of eGFR categories within 5 years, with an adjusted hazard ratio of 1.64 (95% CI: 1.09–2.45, p = 0.02). This underscores the prognostic value of imaging in the early detection of CKD risk. Further research is warranted to explore whether targeted interventions in individuals with low kidney volume can arrest CKD progression.