Combined radiological approach for characterization of solid renal masses: a proposed strategy for integrating the RENAL nephrometry score into the clear cell likelihood score

Abstract

Objective

To systematically evaluate whether the combined application of RENAL nephrometry and clear cell likelihood score (ccLS) could improve the diagnostic reliability of the ccLS in distinguishing clear cell renal cell carcinoma (ccRCC).

Methods

This retrospective single-center study encompassed 140 histopathologically confirmed solid renal masses in 130 patients who underwent preoperative MRI between March 2011 and January 2025. Two radiologists independently assigned the RENAL nephrometry and ccLS for each lesion. The diagnostic performance of the ccLS and the combined radiological approach was assessed using receiver operating characteristic curve analysis in the whole study cohort and two subgroups stratified by lesion diameter (≤ 4 cm vs. >4 cm). Interreader consistency was evaluated using the kappa coefficient.

Results

Overall, 74 (52.9%) ccRCC and 66 (47.1%) non-ccRCC lesions were enrolled. Diagnostic performance analysis yielded a negative predictive value of 90.9%, a specificity of 30.3%, and an AUC of 0.638 for excluding ccRCC when lesions with a ccLS of 2 or lower were accepted as non-ccRCC. With the combined approach, considering low-complexity lesions per the nephrometry score with ccLS of 3 as non-ccRCC improved diagnostic performance using a ccLS cut-off of ≤ 2 for excluding ccRCC (AUC = 0.684, p = 0.022). The combined approach exhibited better performance in lesions ≤ 4 cm for excluding ccRCC among ccLS 3 lesions (p = 0.042). Interobserver consistency was found to be near-perfect (κ = 0.927) and substantial (κ = 0.70) for the nephrometry and ccLS, respectively.

Conclusion

In this preliminary single-center cohort, the combined radiological approach was systematically evaluated for both predicting and excluding ccRCC. Its benefit was concentrated in excluding ccRCC among indeterminate ccLS 3 lesions ≤ 4 cm, reducing false-positive outcomes and enhancing diagnostic confidence and suggesting a role in risk stratification rather than detection; external validation is required before clinical adoption.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02631-6.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02631-6.

Introduction

The incidental detection of solid renal masses (SRMs) in abdominal imaging has been escalating in recent decades, mainly due to the widespread employment of imaging technologies in patients with various clinical conditions [1, 2]. The majority of SRMs were documented to be renal cell carcinoma (RCC) with several subtypes; however, a minority of resected masses measuring ≤ 4 cm were demonstrated to exhibit aggressive biological behavior [3]. On the other hand, the clinical course varies depending on the histological characteristics of RCC subtypes, with clear cell RCC (ccRCC) exhibiting a more aggressive nature and a higher risk of a dismal prognosis [4], underscoring the importance of accurate preoperative assessment and subtyping for optimal clinical management.

In this context, multiparametric magnetic resonance imaging (mpMRI), which provides a non-invasive multistep evaluation of SRMs, has become a pivotal imaging technique for subtyping of RCC and in distinguishing malignant lesions from benign renal masses [5]. Based on mpMRI indices, a novel Likert-scale-based allocation scheme, termed the clear cell likelihood score (ccLS), was devised and has recently been updated to provide a standardized, stepwise approach for interpreting imaging findings and stratifying the probability of an SRM to represent the most prevalent subtype of RCC [6–8]. Furthermore, this algorithm facilitates the prediction of other potential alternative diagnoses within each score category, often narrowing the differential diagnoses and thereby assisting in clinical decision-making and personalized management regarding intervention or active surveillance [8, 9]. In a similar vein, another imaging-based scoring system (i.e., the RENAL nephrometry score) influencing the therapeutic decision process has been developed to quantify surgical complexity based on key anatomical features of renal masses, thereby providing crucial preoperative information guiding surgical planning [10, 11]. Moreover, the nephrometry score and histological features have also been reported to be highly correlated; in particular, ccRCC was found to have higher nephrometry scores than papillary RCC (pRCC) and benign renal masses [12, 13]. Beyond histological subtype, RENAL-defined anatomical complexity has also been linked to tumor biological behavior: higher nephrometry scores correlate with increased Ki-67 proliferative index in ccRCC [14], and individual anatomical components of the RENAL score independently predict malignant histology across a large multi-study meta-analysis [15], collectively suggesting that its anatomical descriptors capture biologically meaningful tumor behavior rather than surgical geometry alone. Hence, integrating the RENAL nephrometry system with ccLS may contribute to a broader perspective for characterizing SRMs and, in turn, may guide more individualized management decisions.

Several studies have demonstrated the promising diagnostic performance of the updated MRI ccLS algorithm in predicting ccRCC, with interreader reliability ranging from moderate to near-perfect [16–22]. Nevertheless, certain limitations remain within this workflow, such as the risk of false-positive outcomes in ccLS 4 and the equivocal assessments in lesions assigned to ccLS 3. A few studies have attempted to enhance the ccLS framework to overcome these shortcomings by utilizing radiomic analysis, considering cystic degeneration/necrosis as an additional criterion, or simplifying major features [23–25]. However, no study to date has incorporated the RENAL nephrometry score into the ccLS framework for predicting/excluding ccRCC, although each method has shown value individually. Therefore, this study primarily aimed to systematically analyze whether the combined application of RENAL nephrometry and ccLS would enhance diagnostic reliability in lesions with ccLS of 3 and 4.

Materials and methods

This study, designed as a retrospective cohort analysis, received approval from the Institutional Review Board of a tertiary university hospital (decision number: 25T-8/53), with informed consent waived.

Study population

The institutional pathology database was scrutinized to distinguish adult patients with histopathologically proven renal masses obtained through either percutaneous biopsy or partial/total nephrectomy, from March 2011 to January 2025. A radiologist who did not participate in image interpretation subsequently searched the picture archiving and communication system (PACS) to identify renal masses exhibiting a ≥ 25% enhancing solid component, without harboring macroscopic fat, and with a maximum time interval of six months between histopathological confirmation and mpMRI. Patients who underwent MRI examination at an outside institution, had inadequate MR images with poor resolution or motion artifacts affecting evaluation, had treatment or biopsy prior to mpMRI, or had neoplasms of non-renal origin, including urothelial carcinoma, were excluded. The flowchart for patient selection is displayed in Fig. 1.

Fig. 1.

Flowchart illustrating the enrollment process of the study cohort. AML, angiomyolipoma; RCC, renal cell carcinoma

Imaging technique

MRI examinations were conducted on either 1.5 T or 3 T whole-body scanners (Magnetom Symphony, Amira, and Verio; Siemens Medical Solutions, Erlangen, Germany) equipped with a body phased-array coil at the study institution. Although the protocols differed in detail between scanners, each examination still comprised the key sequences required to assess the updated ccLS algorithm [8] and RENAL nephrometry score [10], as follows: axial and coronal T2-weighted non-fat-saturated two-dimensional single-shot fast spin-echo sequences; axial T1-weighted dual gradient-echo images; axial fat-saturated T2-weighted fast-spin-echo sequence; axial echo planar DWI with at least two b values; and axial dynamic multiphasic T1-weighted gradient-echo sequences with fat suppression; complete acquisition parameters for both scanner field strengths are provided in Tables S1–S2.

Image interpretation and assignment of the RENAL nephrometry and ccLS

Two radiologists, with 3 and 7 years of experience, independently interpreted mpMRI studies on the institutional PACS, without access to clinical data and histopathology results. Prior to image evaluation, both radiologists reviewed the provided articles in detail, which depicted the application of the RENAL nephrometry and ccLS frameworks [8, 11, 26, 27]. During the image analysis process, both radiologists were also delivered a summary diagram of the nephrometry and ccLS systems, which included written descriptions and corresponding graphical illustrations.

The radiologists recorded all major and ancillary MRI features necessary to designate the ccLS based on the relevant guideline [26]. Among these features, the T2-weighted signal intensity and homogeneity of SRMs, as well as the presence of segmental enhancement inversion (SEI) or diffusion restriction, were visually assessed. Conversely, the corticomedullary phase (CMP) enhancement, presence of microscopic fat, and arterial-to-delayed enhancement ratio (ADER) were quantitatively analyzed. The percentage of CMP enhancement was also noted, along with its category. In contrast, the ADER and presence of microscopic fat were listed in a binary fashion (i.e., ≥ 1.5 or < 1.5, and present or absent, respectively). Based on these findings, a separate score was assigned to each SRM, ranging from 1 (very unlikely) to 5 (very likely).

Similarly, the RENAL nephrometry score was determined based on the five anatomical features on mpMRI, including maximal tumor diameter (radius, R), tumor location within the parenchyma (exo/endophytic status, E), the vicinity of the tumor’s deepest portion to the renal sinus/collecting system (nearness, N), the location of tumor along the ventral or dorsal aspects of the kidney (anterior/posterior/neither, A), and the tumor’s position in relation to the polar lines (location, L), which were rated on a scale of 1 to 3, apart from the “A” descriptor, per reference articles [10, 27]. The assigned scores for each of the four anatomical features were summed to designate the final nephrometry score, which is used to grade the complexity of renal masses as follows: low (< 7 points), moderate (7–9 points), and high complexity (10–12 points).

Finally, any differences between the two radiologists that caused discrepancies in designating the nephrometry or ccLS category were resolved in a consensus meeting, and the final consensus score was used in the statistical analysis.

Establishing the combined radiological approach

Initially, the performance metrics of ccLS, including sensitivity, specificity, positive and negative predictive values (PPV and NPV, respectively), and accuracy, were assessed at a threshold of 4, indicating SRMs with ccLS of 4–5 were considered as ccRCC. Thereafter, to incorporate the RENAL nephrometry score into ccLS, lesions classified as ccLS 3, together with high complexity per nephrometry score, were deemed to indicate ccRCC. Subsequently, similar diagnostic metrics were recalculated to analyze the diagnostic contribution of the combined approach in predicting ccRCC at the same threshold (Analysis 1). Secondly, the diagnostic performance was re-evaluated when lesions with a ccLS of 4 and low complexity were considered non-ccRCC (Analysis 2). Thirdly, the diagnostic performance of the combined approach was reassessed when both aforementioned assumptions coincided (Analysis 3). Additionally, metrics of diagnostic performance for the ccLS algorithm were appraised when a ccLS of ≥ 3 indicated a positive result for ccRCC. Lastly, to analyze whether the combined approach enhances diagnostic accuracy in excluding ccRCC in equivocal outcomes at this cut-off, lesions designated as ccLS 3 and low-complexity were categorized as non-ccRCC (Analysis 4). These assessments were performed for the entire cohort and for subgroups stratified by SRM diameter (≤ 4 cm vs. >4 cm), given size-related differences in diagnostic complexity, treatment, and prognosis. All four analyses were pre-specified prior to data analysis, based on previously established RENAL complexity grade cut-offs [10] and ccLS diagnostic thresholds [8], rather than being derived empirically from the study cohort, thereby minimizing the risk of overfitting.

Histopathological assessment

An experienced genitourinary histopathologist reviewed all specimens of involved SRMs to ascertain their subtype per the 2022 WHO classification [28]. Tumor stage was defined based on imaging findings for patients who underwent biopsy, and pathological evaluation was used for those who had extirpative procedures.

Statistical analysis

All statistical analyses were performed using IBM SPSS Statistics for Windows, Version 25.0 (IBM Corp., Armonk, NY, USA) and MedCalc Statistical Software, Version 23.3.7 (MedCalc Software Ltd., Ostend, Belgium). The distribution of continuous variables was assessed using the Kolmogorov–Smirnov and Shapiro–Wilk tests. Data not following a normal distribution are presented as median (Q1–Q3), whereas normally distributed variables are expressed as mean ± standard deviation (SD). Categorical variables are summarized as counts and percentages [n (%)].

For group comparisons, the Mann–Whitney U test was used for two groups, and the Kruskal–Wallis H test for comparisons involving more than two groups. When significant differences were detected, post-hoc pairwise comparisons with Benjamini–Hochberg correction were applied to control for type I error. For categorical data, comparisons were performed using the Pearson Chi-square test, while Fisher’s Exact test or Fisher–Freeman–Halton test was applied when appropriate. Interobserver agreement was evaluated using different reliability indices. For categorical variables, Cohen’s kappa coefficient (κ) was calculated, and for multi-categorical data, the weighted Cohen’s kappa was applied. For continuous variables, reliability was assessed using the Intraclass Correlation Coefficient (ICC). A two-way random-effects model with absolute agreement was used for interobserver reliability. The diagnostic accuracy of the ccLS system and combined radiological approach for predicting ccRCC was further systematically assessed. Performance metrics comprised accuracy rate, F1-score, Matthews Correlation Coefficient (MCC), area under the ROC curve (AUC), and standard error (SE). The Youden index (J) was used to determine the optimal cut-off point, and differences between ROC curves were evaluated using the DeLong test. In addition, sensitivity, specificity, PPV, and NPV were reported. A two-sided p-value < 0.05 was considered statistically significant in all analyses.

Results

Overall, 140 SRMs were detected in 130 patients [89 (68.5%) men and 41 (31.5%) women; median age at the time of initial diagnosis, 60 (IQR, 53–68 years)], of whom seven (5.4%) presented with multiple lesions. The median largest tumor diameter was 35 mm (IQR, 22.5–53 mm), with 85 lesions (61%) measuring ≤ 4 cm in diameter. A significant proportion of the lesions (89.3%, 125/140) had a histopathological diagnosis after partial or radical nephrectomy, while US-guided percutaneous biopsy rendered the diagnosis in the remaining SRMs, resulting in a cohort of 74 (52.9%) ccRCC and 66 (47.1%) non-ccRCC. No significant differences were observed in patient age, gender, tumor size, or tumor stage between ccRCC and non-ccRCC groups. Demographic features of patients, along with tumor characteristics, are outlined in Table 1.

Table 1.

Demographic features and baseline tumor characteristics of the study cohort

	Total	ccRCC	non-ccRCC	p
Number of Patients	130	66 (50.8)	64 (49.2)
Number of Lesions	140	74 (52.9)	66 (47.1)
Age [median (Q1-Q3), year]	60 (53–68)	59.5 (54–65)	61.5 (50–71)	0.326 ᵘ
Sex, n (%)				0.999 ᶜ
Female	41 (31.5)	21 (31.8)	20 (31.3)
Male	89 (68.5)	45 (68.2)	44 (68.8)
Mass size [median (Q1-Q3), mm]	35 (22.5–53)	35 (25–56)	35.5 (21–48)	0.516 ᵘ
Side of tumor, n (%)				0.866 ᶜ
Right	68 (48.6)	35 (47.3)	33 (50.0)
Left	72 (51.4)	39 (52.7)	33 (50.0)
Pathology Source, n (%)				0.222 ᶜ
Biopsy	15 (10.7)	5 (6.8)	10 (15.2)
Partial Nephrectomy	65 (46.4)	34 (45.9)	31 (46.9)
Radical Nephrectomy	60 (42.9)	35 (47.3)	25 (37.9)
T stage, n (%)				0.204 ᶠᶠ
T1	116 (82.9)	61 (82.4)	55 (83.3)
T2	8 (5.7)	4 (5.4)	4 (6.1)
T3	13 (9.3)	9 (12.2)	4 (6.1)
T4	3 (2.1)	0 (0.0)	3 (4.5)

ccRCC: clear cell renal cell carcinoma; ᵘ Mann Whitney U Test (Monte Carlo); ᶜ Pearson Chi Square test (Monte Carlo); ᶠᶠ Fisher Freeman Halton Test (Monte Carlo); Post Hoc test: Benjamini-Hochberg correction; n: number (percentage); Q1: first quartile (25th percentile); Q3: third quartile (75th percentile)

The probability of ccRCC diagnosis increased with higher ccLS, as follows: 5.9% in ccLS 1; 20% in ccLS 2; 18.9% in ccLS 3; 57.1% in ccLS 4; and 97.8% in ccLS 5. Thus, there was a predominance of ccRCC in ccLS categories 4 and 5. Conversely, the most frequent histopathology in ccLS categories 1 and 3 was pRCC, with 76.5% and 37.8% of SRMs in those categories, respectively. Moreover, most of the renal masses with a ccLS of 2 were diagnosed as fat-poor angiomyolipoma (fpAML), accounting for 60% of these lesions. In addition, lesions confirmed as oncocytoma were mainly assigned a ccLS of 4, whereas chromophobe (chr) RCC was primarily designated as ccLS 3, with additional cases categorized within ccLS 4. When using the nephrometry score, the highest values were noted in lesions diagnosed as oncocytoma, while those identified as fpAML demonstrated the lowest values (p = 0.009). Additionally, exploratory comparison of individual RENAL components between ccRCC and non-ccRCC lesions showed a nominally significant difference in exo/endophytic status between groups (p = 0.015; not significant after correction for multiple comparisons across the five components, corrected p = 0.077), a non-significant trend for nearness (p = 0.078), and no significant difference for the remaining components (radius, anterior/posterior/neither, location) (Table S3). In contrast, no significant difference was detected among RCC subtypes in terms of nephrometry grade, as most exhibited moderate complexity. However, lesions with low complexity were more frequently encountered in pRCC compared to those with high complexity, whereas an opposite pattern was observed in ccRCC. Moreover, the distribution of lesions with low and high complexity was equal in tumors shown to be chrRCC. Histopathological distribution of the nephrometry score/grade and ccLS categories is presented in Table 2.

Table 2.

Histopathological distribution of the RENAL nephrometry score/grade and the clear cell likelihood score categories

	Histopathological subtypes						p
	ccRCC	pRCC	chrRCC	Oncocytoma	fpAML	other RCC
	(n = 74)	(n = 27)	(n = 13)	(n = 8)	(n = 5)	(n = 13)
	median(Q1-Q3)	median(Q1-Q3)	median(Q1-Q3)	median(Q1-Q3)	median(Q1-Q3)	median(Q1-Q3)
Nephrometry score (Consensus)	9 (7–9)	8 (6–9)	9 (6–10)	9.5 (8-10.5) ᵉ	6 (5–6)	8 (7–9)	0.009ᵏ
	n (%)	n (%)	n (%)	n (%)	n (%)	n (%)
Nephrometry grade							0.106 ᶠᶠ
Low complexity	13 (17.6)	8 (29.6)	4 (30.8)	0 (0.0)	4 (80.0)	3 (23.1)
Moderate complexity	44 (59.4)	15 (55.6)	5 (38.5)	4 (50.0)	1 (20.0)	7 (53.8)
High complexity	17 (23.0)	4 (14.8)	4 (30.8)	4 (50.0)	0 (0.0)	3 (23.1)
Nephrometry grade							0.316 ᶠᶠ
Others	57 (77.0)	23 (85.2)	9 (69.2)	4 (50.0)	5 (100.0)	10 (76.9)
High complexity	17 (23.0)	4 (14.8)	4 (30.8)	4 (50.0)	0 (0.0)	3 (23.1)
Nephrometry grade							0.023ᶠᶠ
Others	61 (82.4)	19 (70.4)	9 (69.2)	8 (100.0)ᵉ	1 (20.0)	10 (76.9)
Low complexity	13 (17.6)	8 (29.6)	4 (30.8)	0 (0.0)	4 (80.0)ᵈ	3 (23.1)
ccLS (Consensus)							< 0.001ᶠᶠ
Very unlikely	1 (1.4)	13 (48.1) ᵃ	1 (7.7)	0 (0.0)	0 (0.0)	2 (15.4)
Unlikely	1 (1.4)	0 (0.0)	0 (0.0)	0 (0.0)	3 (60.0)ᵃᵇᶜ	1 (7.7)
Equivocal	7 (9.4)	14 (51.9)ᵃ	7 (53.8)ᵃ	2 (25.0)	1 (20.0)	6 (46.1)ᵃ
Likely	20 (27.0)ᵇ	0 (0.0)	5 (38.5)ᵇ	6 (75.0)ᵇ	1 (20.0)	3 (23.1)
Highly likely	45 (60.8)ᵇᶜᵈᶠ	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	1 (7.7)

ccRCC, clear cell renal cell carcinoma; pRCC, papillary renal cell carcinoma; chrRCC, chromophobe renal cell carcinoma; fpAML, fat-poor angiomyolipoma

ᵏ Kruskal-Wallis H Test (Monte Carlo); ᶠᶠ Fisher Freeman Halton Test (Monte Carlo); Post Hoc test: Benjamini-Hochberg correction; Q1:first quartile (25th percentile); Q3:third quartile (75th percentile)

^a–f Indicates statistical significance compared with the histopathological subtypes of ccRCC, pRCC, chrRCC, oncocytoma, fpAML, and other RCC, respectively

Diagnostic accuracy for predicting ccRCC using a ccLS cut-off of ≥ 4 was 82.1%, with corresponding sensitivity of 87.8%, PPV of 80.2%, specificity of 75.8%, NPV of 84.7%, F1-score of 0.84, MCC of 0.64, and the AUC of 0.818. Among the 81 lesions designated as ccLS of 4–5, 16 (19.8%) were ultimately diagnosed as non-ccRCC, constituting false-positive results, whereas 9 (15.3%) out of 59 lesions categorized as ccLS 1–3 represented false-negative outcomes. Conversely, of the 22 lesions assigned to ccLS 1–2, only two (9.1%) were confirmed as ccRCC. Therefore, applying a ccLS threshold of 2 to exclude ccRCC resulted in higher NPV and sensitivity (90.9% and 97.3%, respectively), though it compromised other performance metrics due to the misclassification of 46 (39%, 46/118) lesions with ccLS ≥ 3 as ccRCC. The diagnostic performance of the ccLS algorithm and the combined radiological approach for the whole study cohort is detailed in Table 3; Fig. 2. Regarding the combined approach, when lesions classified as ccLS 3 and having high complexity were considered as ccRCC with a ccLS cut-off of ≥ 4 (Analysis 1), a minor improvement in sensitivity and NPV was observed; however, performance in the other parameters declined significantly (p = 0.035). Conversely, when lesions designated as ccLS 4 and having low complexity were deemed non-ccRCC with the same threshold (Analysis 2), all performance metrics, except for sensitivity, NPV, and F1-score, showed improvement, although this did not reach statistical significance (p = 0.38). Moreover, the concurrence of these assumptions (Analysis 3) was associated with an insignificant reduction in diagnostic performance compared to that of the ccLS workflow (p = 0.346). Lastly, when lesions with a ccLS of 3 and low complexity were categorized as non-ccRCC (Analysis 4, Fig. 3), except for sensitivity and NPV, the remaining metrics improved significantly when using a ccLS threshold of ≤ 2 to exclude ccRCC (p = 0.022).

Table 3.

The diagnostic performance of the clear cell likelihood score (ccLS) algorithm and the combined radiological approach for the whole study cohort

	Clear cell renal cell carcinoma		Accuracy rate (%)	F1 - Score	Matthews Correlation Coefficient (MCC)	AUC (SE)	P	Pairwise comparison of ROC curves
	Negative (Absent)	Positive (Present)						Variables 1 vs. Variables 2	p
	n (Row%) (Column%)	n (Row%) (Column%)
Analysis 1			77.9	0.81	0.57	0.771 (0.034)	< 0.001	Analysis 1 vs. Analysis 2	0.024
Negative	42 (85.7) ⁿᵖᵛ (63.6) ˢᵖ	7 (14.3) (9.5)						Analysis 1 vs. Analysis 3	0.380
Positive	24 (26.4) (36.4)	67 (73.6) ᵖᵖᵛ (90.5) ˢˢ						Analysis 1 vs. Analysis 4	0.003
Analysis 2			83.6	0.84	0.67	0.836 (0.031)	< 0.001	Analysis 1 vs. ccLS accuracy 1	0.035
Negative	55 (82.1) ⁿᵖᵛ (83.3) ˢᵖ	12 (17.9) (16.2)						Analysis 1 vs. ccLS accuracy 2	< 0.001
Positive	11 (15.1) (16.7)	62 (84.9) ᵖᵖᵛ (83.8) ˢˢ
Analysis 3			79.3	0.82	0.59	0.788 (0.035)	< 0.001	Analysis 2 vs. Analysis 3	0.035
Negative	47 (82.5) ⁿᵖᵛ (71.2) ˢᵖ	10 (17.5) (13.5)						Analysis 2 vs. Analysis 4	< 0.001
Positive	19 (22.9) (28.8)	64 (77.1) ᵖᵖᵛ (86.5) ˢˢ						Analysis 2 vs. ccLS accuracy 1	0.380
Analysis 4			70.0	0.77	0.45	0.684 (0.046)	< 0.001	Analysis 2 vs. ccLS accuracy 2	< 0.001
Negative	27 (90.0) ⁿᵖᵛ (40.9) ˢᵖ	3 (10.0) (4.1)
Positive	39 (35.5) (59.1)	71 (64.5) ᵖᵖᵛ (95.9) ˢˢ						Analysis 3 vs. Analysis 4	0.002
ccLS accuracy 1			82.1	0.84	0.64	0.818 (0.033)	< 0.001	Analysis 3 vs. ccLS accuracy 1	0.346
Negative	50 (84.7) ⁿᵖᵛ (75.8) ˢᵖ	9 (15.3) (12.2)						Analysis 3 vs. ccLS accuracy 2	< 0.001
Positive (ccLS ≥ 4)	16 (19.8) (24.2)	65 (80.2) ᵖᵖᵛ (87.8) ˢˢ
ccLS accuracy 2			65.7	0.75	0.38	0.638 (0.030)	< 0.001	Analysis 4 vs. ccLS accuracy 1	< 0.001
Negative	20 (90.9) ⁿᵖᵛ (30.3) ˢᵖ	2 (9.1) (2.7)						Analysis 4 vs. ccLS accuracy 2	0.022
Positive (ccLS ≥ 3)	46 (39.0) (69.7)	72 (61.0) ᵖᵖᵛ (97.3) ˢˢ						ccLS accuracy 1 vs. ccLS accuracy 2	< 0.001

ROC curve Analysis (Youden index J - Delong); AUC: Area Under the Curve; SE: Standard Error; ˢˢ Sensitivity; ˢᵖ Specificity; ⁿᵖᵛ Negative Predictive Value; ᵖᵖᵛ Positive Predictive Value; vs.: versus Analysis 1: ccLS = 3 & nephrometry grade = 3 → accepted as ccRCC; Analysis 2: ccLS = 4 & nephrometry grade = 1 → designated as non-ccRCC; Analysis 3: combination of Analysis 1 and 2; Analysis 4: ccLS = 3 & nephrometry grade 1 assigned as non-ccRCC

Fig. 2.

The receiver operating characteristic curves of the combined radiological approach (Analysis 1–4) and clear cell likelihood score (ccLS) in the whole study cohort for predicting and excluding clear cell renal cell carcinoma (ccRCC). Analysis 1: ccLS = 3 and nephrometry grade = 3 → accepted as ccRCC; Analysis 2: ccLS = 4 and nephrometry grade = 1 → designated as non-ccRCC; Analysis 3: combination of Analysis 1 and 2; Analysis 4: ccLS = 3 and nephrometry grade 1 assigned as non-ccRCC; ccLS accuracy 1: using a ccLS cut-off of ≥ 4 in predicting ccRCC; ccLS accuracy 2: applying a ccLS threshold of 2 in excluding ccRCC

Fig. 3.

Multiparametric MRI scans of an 18-mm right renal mass arising from the lower pole (A–I). A and B, Coronal and axial T2-weighted single-shot fast spin-echo images depict a completely exophytic mass (thick arrow) with predominantly hypointense signal intensity compared to the renal cortex. Small areas of cystic degeneration are also present (A, thin arrows). Axial fat-saturated T1-weighted contrast-enhanced images in corticomedullary (C) and delayed (D) phases demonstrate avid enhancement without marked washout, resulting in an arterial-to-delayed enhancement ratio of 1.3 (< 1.5). The lesion is located below the lower polar line on coronal fat-saturated T1-weighted contrast-enhanced images in the delayed phase (E), and the distance between the innermost edge of the mass and the renal sinus is measured to be 6 mm. No microscopic fat is detected on axial T1-weighted in- and opposed-phase images (F and G, respectively). Axial diffusion-weighted image (H) displays high signal intensity at a b value of 800 s/mm², and the corresponding apparent diffusion coefficient (I) verifies diffusion restriction. The lesion was assigned a clear cell likelihood score of 3 and exhibited low complexity, as per the RENAL nephrometry score. Based on the combined radiological approach, the mass was considered as non-clear cell renal cell carcinoma, which was compatible with the histopathological diagnosis of fat-poor angiomyolipoma

Subgroup analyses revealed superior diagnostic performance of the ccLS system and combined approach in lesions ≤ 4 cm compared to those > 4 cm; however, the difference was significant only for Analyses 1, 3, and 4 (p = 0.005, 0.001, and 0.042, respectively). The combined approach exhibited a similar effect on the diagnostic metrics in Analysis 1, 2, and 4 in lesions ≤ 4 cm compared to the whole study cohort, whereas Analysis 3 achieved higher diagnostic accuracy. Conversely, for lesions larger than 4 cm, the combined approach had no impact on diagnostic performance in Analyses 2 and 4, while deteriorating the outcomes in the remaining analyses (Table 4).

Table 4.

The diagnostic performance of the clear cell likelihood score (ccLS) system and the combined radiological approach stratified by the diameter of the renal mass

	Clear cell renal cell carcinoma		Accuracy rate (%)	F1 - Score	Matthews Correlation Coefficient (MCC)	AUC (SE)	P	Pairwise comparison of ROC curves
	Negative (Absent)	Positive (Present)						Variables 1 vs. Variables 2	p
	n (Row%) (Column %)	n (Row%) (Column %)
Mass diameter ≤ 4 cm
Analysis 1			84.7	0.860	0.697	0.845 (0.039)	< 0.001	Analysis 1 vs. Analysis 2	0.249
Negative	32 (88.9) ⁿᵖᵛ (78.0) ˢᵖ	4 (11.1) (9.1)						Analysis 1 vs. Analysis 3	0.404
Positive	9 (18.4) (22.0)	40 (81.6) ᵖᵖᵛ (90.9) ˢˢ						Analysis 1 vs. Analysis 4	0.004
Analysis 2			88.2	0.881	0.769	0.884 (0.035)	< 0.001	Analysis 1 vs. ccLS accuracy 1	0.317
Negative	38 (84.4) ⁿᵖᵛ (92.7) ˢᵖ	7 (15.6) (15.9)						Analysis 1 vs. ccLS accuracy 2	< 0.001
Positive	3 (7.5) (7.3)	37 (92.5) ᵖᵖᵛ (84.1) ˢˢ
Analysis 3			87.1	0.871	0.743	0.872 (0.036)	< 0.001	Analysis 2 vs. Analysis 3	0.317
Negative	37 (84.1) ⁿᵖᵛ (90.2) ˢᵖ	7 (15.9) (15.9)						Analysis 2 vs. Analysis 4	0.001
Positive	4 (9.8) (9.8)	37 (90.2) ᵖᵖᵛ (84.1) ˢˢ						Analysis 2 vs. ccLS accuracy 1	0.404
Analysis 4			74.1	0.792	0.525	0.733 (0.043)	< 0.001	Analysis 2 vs. ccLS accuracy 2	< 0.001
Negative	21 (91.3) ⁿᵖᵛ (51.2) ˢᵖ	2 (8.7) (4.5)
Positive	20 (32.3) (48.8)	42 (67.7) ᵖᵖᵛ (95.5) ˢˢ						Analysis 3 vs. Analysis 4	0.002
ccLS accuracy 1			85.9	0.870	0.720	0.857 (0.038)	< 0.001	Analysis 3 vs. ccLS accuracy 1	0.675
Negative	33 (89.2) ⁿᵖᵛ (80.5) ˢᵖ	4 (10.8) (9.1)						Analysis 3 vs. ccLS accuracy 2	< 0.001
Positive (ccLS ≥ 4)	8 (16.7) (19.5)	40 (83.3) ᵖᵖᵛ (90.9) ˢˢ
ccLS accuracy 2			67.1	0.754	0.418	0.659 (0.039)	< 0.001	Analysis 4 vs. ccLS accuracy 1	0.002
Negative	14 (93.3) ⁿᵖᵛ (34.1) ˢᵖ	1 (6.7) (2.3)						Analysis 4 vs. ccLS accuracy 2	0.020
Positive (ccLS ≥ 3)	27 (38.6) (65.9)	43 (61.4) ᵖᵖᵛ (97.7) ˢˢ						ccLS accuracy 1 vs. ccLS accuracy 2	< 0.001
Mass diameter > 4 cm
Analysis 1			67.3	0.750	0.352	0.650 (0.057)	0.009	Analysis 1 vs. Analysis 2	0.038
Negative	10 (76.9) ⁿᵖᵛ (40.0) ˢᵖ	3 (23.1) (10.0)						Analysis 1 vs. Analysis 3	0.999
Positive	15 (35.7) (60.0)	27 (64.3) ᵖᵖᵛ (90.0) ˢˢ						Analysis 1 vs. Analysis 4	0.289
Analysis 2			76.4	0.794	0.522	0.757 (0.059)	< 0.001	Analysis 1 vs. ccLS accuracy 1	0.038
Negative	17 (77.3) ⁿᵖᵛ (68.0) ˢᵖ	5 (22.7) (16.7)						Analysis 1 vs. ccLS accuracy 2	0.289
Positive	8 (24.2) (32.0)	25 (75.8) ᵖᵖᵛ (83.3) ˢˢ
Analysis 3			67.3	0.750	0.352	0.650 (0.057)	0.009	Analysis 2 vs. Analysis 3	0.038
Negative	10 (76.9) ⁿᵖᵛ (40.0) ˢᵖ	3 (23.1) (10.0)						Analysis 2 vs. Analysis 4	0.010
Positive	15 (35.7) (60.0)	27 (64.3) ᵖᵖᵛ (90.0) ˢˢ						Analysis 2 vs. ccLS accuracy 1	0.999
Analysis 4			63.6	0.744	0.309	0.603 (0.047)	0.027	Analysis 2 vs. ccLS accuracy 2	0.010
Negative	6 (85.7) ⁿᵖᵛ (24.0) ˢᵖ	1 (14.3) (3.3)
Positive	19 (39.6) (76.0)	29 (60.4) ᵖᵖᵛ (96.7) ˢˢ						Analysis 3 vs. Analysis 4	0.289
ccLS accuracy 1			76.4	0.794	0.522	0.757 (0.059)	< 0.001	Analysis 3 vs. ccLS accuracy 1	0.038
Negative	17 (77.3) ⁿᵖᵛ (68.0) ˢᵖ	5 (22.7) (16.7)						Analysis 3 vs. ccLS accuracy 2	0.289
Positive (ccLS ≥ 4)	8 (24.2) (32.0)	25 (75.8) ᵖᵖᵛ (83.3) ˢˢ
ccLS accuracy 2			63.6	0.744	0.309	0.603 (0.047)	0.027	Analysis 4 vs. ccLS accuracy 1	0.010
Negative	6 (85.7) ⁿᵖᵛ (24.0) ˢᵖ	1 (14.3) (3.3)						Analysis 4 vs. ccLS accuracy 2	0.999
Positive (ccLS ≥ 3)	19 (39.6) (76.0)	29 (60.4) ᵖᵖᵛ (96.7) ˢˢ						ccLS accuracy 1 vs. ccLS accuracy 2	0.010

ROC curve Analysis (Youden index J - Delong); AUC: Area Under the Curve; SE: Standard Error; ˢˢ Sensitivity; ˢᵖ Specificity; ⁿᵖᵛ Negative Predictive Value; ᵖᵖᵛ Positive Predictive Value; vs.: versus

Analysis 1: ccLS = 3 & nephrometry grade = 3 → accepted as ccRCC; Analysis 2: ccLS = 4 & nephrometry grade = 1 → designated as non-ccRCC; Analysis 3: combination of Analysis 1 and 2; Analysis 4: ccLS = 3 & nephrometry grade 1 assigned as non-ccRCC

The level of interobserver reliability between radiologists for the ccLS on a 1–5 scale was substantial (κ = 0.70). The presence of microscopic fat showed the highest agreement among major features, each of which exhibited substantial concordance, while the ancillary features demonstrated moderate consistency. Of note, the radiologists reached almost perfect agreement for the overall nephrometry score rating (κ = 0.927), with the highest reliability noted for tumor diameter, whereas exo/endophytic status yielded the least consistent evaluations. Comprehensive data on interobserver agreement for both the ccLS and nephrometry score assessments are provided in Table 5.

Table 5.

Interobserver consistency for both the clear cell likelihood score (ccLS) and the RENAL nephrometry score

Scoring systems	κ or ICC (95% CI)	p
Clear cell likelihood score
Signal intensity on T2WI	0.676 (0.572, 0.780) a	< 0.001
CME (category)	0.671 (0.563, 0.779) a	< 0.001
CME (%)	0.787 (0.703, 0.848) c	< 0.001
Microscopic fat	0.784 (0.680, 0.888) b	< 0.001
SEI	0.486 (0.055, 0.917) b	0.003
ADER	0.597 (0.276, 0.918) b	< 0.001
Marked restriction on DWI	0.468 (0.315, 0.621) b	< 0.001
Homogeneity	0.553 (0.375, 0.731) b	< 0.001
ccLS (overall)	0.700 (0.610, 0.790) a	< 0.001
RENAL nephrometry score
Radius	0.903 (0.838, 0.968) a	< 0.001
Exo/endophytic status	0.419 (0.296, 0.542) a	< 0.001
Nearness	0.715 (0.586, 0.844) a	< 0.001
Anterior/posterior/neither	0.583 (0.473, 0.693) a	< 0.001
Location	0.607 (0.499, 0.715) a	< 0.001
Nephrometry score (overall)	0.927 (0.898, 0.948) c	< 0.001

κ: kappa coefficient; ICC: Intraclass correlation coefficient; CI: confidence interval; CME: corticomedullary phase enhancement; SEI: segmental enhancement inversion; ADER: arterial-to-delayed enhancement ratio

Interobserver agreements were assessed using ^a Weighted kappa, ^b Cohen’s kappa, and ^c the Intraclass correlation coefficient (two-way random, absolute agreement)

Discussion

The present study systematically evaluated the combined radiological approach for both predicting and excluding ccRCC; the results revealed that its main contribution lies in optimizing the diagnostic confidence of the ccLS system by reducing false-positive outcomes when excluding ccRCC, especially in lesions ≤ 4 cm and when classified as ccLS 3, rather than in improving ccRCC detection. Thus, the integration of the RENAL nephrometry score into the ccLS framework may aid in better risk stratification of equivocal lesions and have the potential to guide clinical decision-making in the context of active surveillance.

In line with previous studies [29], the current study also displayed high sensitivity, PPV, and accuracy, yet with moderate specificity, using a ccLS threshold of ≥ 4 to predict ccRCC in the entire study cohort. In contrast, a cut-off of ccLS ≤ 2 yielded higher NPV, but at the expense of lower accuracy. Despite these encouraging outcomes, 43% of lesions assigned to ccLS 4 were eventually proven to be non-ccRCC, with oncocytoma and chrRCC being the most common histological subtypes, which potentially explains the relatively low specificity at this threshold. Moreover, a relatively higher proportion of ccLS 3 category, considered as indeterminate and mainly comprising non-ccRCC lesions, in the present study [26.4% vs. 13–20% in prior reports [23, 30] may have contributed to the reduced accuracy observed at this cut-off. Considering these drawbacks, any supplementary diagnostic tool that enables better stratification of these equivocal lesions or reduces false-positive results would enhance the diagnostic performance of the ccLS algorithm.

As the RENAL nephrometry score not only encompasses anatomical features but also reflects histological information, considering that ccRCC was found to have higher values compared with pRCC and benign lesions [12, 13], incorporating this score into the ccLS system may therefore improve diagnostic metrics. Notably, among the individual RENAL components, the exo/endophytic status showed the strongest, though only nominally significant, association with ccRCC status, with ccRCC lesions disproportionately assigned to the intermediate exophytic category, while the nearness component showed a non-significant trend. This exo/endophytic finding did not survive correction for multiple comparisons and should therefore be regarded as hypothesis-generating rather than confirmatory. Prior work has independently shown that individual RENAL components, rather than the composite score alone, can carry predictive value for malignant and high-grade pathology [12, 15]; our exploratory finding is compatible with this broader premise but cannot itself confirm it, and exo/endophytic status in particular may merit further investigation as a discrete anatomical correlate of ccRCC in adequately powered studies. The composite score’s overall discriminative value in our cohort, meanwhile, likely reflects cumulative anatomical complexity rather than any single descriptor. In the present study, consistent with previous reports, high-complexity lesions were more frequently detected in patients with ccRCC, while low-complexity lesions were more prevalently observed among those with pRCC. However, this difference was not statistically significant, probably due to the relatively small sample size of pRCC. Nonetheless, the combined approach significantly strengthened the specificity, PPV, and accuracy, while maintaining a high NPV, at a threshold of ccLS ≤ 2 for excluding ccRCC, when ccLS3 lesions with low complexity were reclassified as non-ccRCC, allowing appropriate stratification of five pRCC, one chrRCC, and one fpAML. On the other hand, a similar assumption for ccLS 4 lesions at a ccLS cut-off of 4 resulted in the accurate categorization of three chrRCC, one fpAML, and one other RCC, while three ccRCC were misclassified as non-ccRCC, leading to a modest, non-significant improvement in accuracy. Furthermore, diagnostic accuracy declined when ccLS 3 lesions with high complexity were designated as ccRCC, due to an increased number of false-positive outcomes.

Two previous studies, including five common histological types of SRMs, demonstrated that the ccLS algorithm exhibited comparable diagnostic performance in both cT1a and cT1b stages for diagnosing ccRCC [17, 24]. Conversely, in the present study, improved diagnostic accuracy was observed in lesions ≤ 4 cm for both predicting and excluding ccRCC. This discrepancy might partially reflect the difference in enrolled histological types, as apart from those identified by the ccLS system, other RCCs were also assessed in our study. Likewise, the potential role of the combined approach as a rule-out criterion could not be demonstrated in lesions ≥ 4 cm, as no ccLS 3 or 4 lesions in this size group exhibited low complexity. In addition, the number of false-positive results increased in both size subgroups when ccLS 3 lesions with high complexity were considered as ccRCC, given that nearly half of the lesions with high complexity were diagnosed as non-ccRCC. Taken together, these results suggest that the contribution of the combined approach appears to lie in exclusion rather than in prediction of ccRCC, particularly in ccLS 3 lesions ≤ 4 cm.

From a clinical standpoint, our findings have the greatest potential relevance for the ccLS 3 category specifically, as this is the group in which the management pathway is currently least defined. In the developers’ series, ccLS 3 lesions comprised a near-even mixture of ccRCC and benign histology, such that definitive treatment without pathological confirmation would result in an unacceptably high rate of treating such tumors; accordingly, confirmatory renal mass biopsy has been advised specifically for ccLS 3 lesions by the developers of the ccLS algorithm, whereas ccLS 1–2 lesions rarely require biopsy given the negligible likelihood of ccRCC, and ccLS 4–5 lesions are generally treated definitively without biopsy [8]. In our own cohort, ccLS 3 lesions were even more skewed toward non-ccRCC histology, including benign or indolent lesions. This makes ccLS 3 the precise point in the diagnostic pathway at which an adjunctive stratification tool could have the greatest clinical impact. Our finding that low-complexity ccLS 3 lesions could be reclassified as non-ccRCC with a statistically significant improvement in diagnostic performance (7 of 8 lesions correctly reclassified; 1 false negative) suggests that the RENAL nephrometry score may help identify a subset of ccLS 3 lesions in which biopsy could be deferred in favor of active surveillance with imaging follow-up. Because RENAL nephrometry scoring is already performed independently of ccLS as part of routine preoperative planning [10, 11], this additional stratification would require no extra imaging. This proposed application should, however, be interpreted with caution given the small subgroup size and the residual, non-zero false-negative rate (12.5%), and requires prospective validation before it can be used to inform biopsy decisions for ccLS 3 lesions in practice. Nephrometry-based reclassification could nonetheless be evaluated, alongside established clinical factors such as age, comorbidity, and patient preference, as a candidate additional parameter in future prospective studies defining active surveillance eligibility criteria for indeterminate renal masses.

In the current study, the overall nephrometry score attained superior interrater agreement compared to any of the major and ancillary features of the ccLS system, exhibiting moderate to substantial consistency. Comparable outcomes have also been outlined in recent studies evaluating the interobserver reliability of these algorithms in different cohorts [22, 24, 31], suggesting that the combined approach may provide repeatable performance. This discrepancy likely reflects a structural difference between the two systems: the RENAL score is derived from discrete anatomical measurements, whereas ccLS integrates multiple qualitative and semi-quantitative imaging features requiring more subjective, Likert-based judgment. This difference in reproducibility should be considered when combining the two systems in clinical practice, as the added value of nephrometry-based reclassification is unlikely to be substantially degraded by interobserver variability, in contrast to ccLS itself.

The findings should be interpreted in light of several limitations, primarily related to the retrospective design. First, all enrolled cases were verified histopathologically to ensure diagnostic certainty; however, this design inherently excludes small renal masses managed with active surveillance without biopsy or resection, which represent a substantial proportion of incidentally detected lesions in clinical practice. This may have led to underrepresentation of indolent or low-complexity lesions and could limit the generalizability of our findings to the broader population of SRMs currently managed non-operatively, in addition to introducing potential sampling bias given the relatively low frequency of benign lesions in our cohort. Second, several non-ccRCC histological subgroups were small, e.g., fpAML and oncocytoma, and histological-subtype- and size-stratified subgroup analyses should therefore be regarded as exploratory and hypothesis-generating rather than confirmatory. Third, due to the long study period, different MRI systems were employed. Although all examinations covered the fundamental sequences that enable comprehensive assessment via the nephrometry score and the updated ccLS algorithm, further studies are necessary to elucidate whether, and to what extent, variations in sequence settings influence the application of the combined approach. Fourth, although the diagnostic-accuracy improvement associated with the combined approach reached statistical significance, its absolute magnitude was modest, and diagnostic-accuracy metrics alone cannot establish clinical significance; demonstrating genuine clinical benefit will require prospective studies that directly measure clinical outcomes and decisions, such as biopsies avoided or the downstream course of missed ccRCC cases, rather than diagnostic-accuracy metrics in isolation. Finally, the data were derived from a single center, which inevitably limited the sample size and constrained the generalizability of the outcomes, warranting future validation in larger, multi-institutional cohorts.

In conclusion, this preliminary, single-center study suggests that the combined radiological approach may aid in excluding ccRCC among ccLS 3 lesions, particularly in lesions ≤ 4 cm, which could provide improved delineation of SRMs for active surveillance and thus facilitate individualized clinical management. As this study represents the first attempt to incorporate the RENAL nephrometry score into the ccLS algorithm, its findings should be regarded as hypothesis-generating; additional prospective, multi-institutional research is needed to externally validate these results and establish the clinical utility of this proposed integration strategy before it can inform routine clinical decision-making.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ADER

Arterial-to-delayed enhancement ratio

Chr

Chromophobe

ccLS

Clear cell likelihood score

ccRCC

Clear cell renal cell carcinoma

CMP

Corticomedullary phase

fpAML

Fat-poor angiomyolipoma

ICC

Intraclass Correlation Coefficient

MCC

Matthews Correlation Coefficient

mpMRI

Multiparametric magnetic resonance imaging

pRCC

Papillary renal cell carcinoma

PACS

Picture archiving and communication system

PPV

Positive predictive values

NPV

Negative predictive values

RCC

Renal cell carcinoma

SEI

Segmental enhancement inversion

SRMs

Solid renal masses

SD

Standard deviation

Author contributions

İ.H. wrote the manuscript; İ.H. performed the data analysis; İ.H. and A.A.T. performed the image evaluation; B.S.K. performed the pathological evaluation; İ.H. and E.G. prepared all tables and figures; B.S.K. and E.G. performed data collection and supervision; E.G. performed editing. All authors reviewed the manuscript.

Funding

None.

Data availability

All study data are presented in this paper.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the principles of the Declaration of Helsinki. The study received approval from the Institutional Review Board of Ege University Faculty of Medicine (decision number: 25T-8/53). The Ethics Committee of Ege University Faculty of Medicine waived the requirement for written informed consent due to the study’s retrospective nature.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.