CT-based Node-RADS classification in the assessment of rectal cancer: diagnostic performance and postoperative prognostic stratification

Abstract

Background

Lymph node metastasis significantly impacts rectal cancer prognosis and treatment planning; yet, current imaging assessment methods lack standardization. This study aimed to evaluate the diagnostic performance and postoperative prognostic implications of the Node Reporting and Data System (Node-RADS) 1.0 categories in the initial regional nodal assessment of rectal cancer.

Methods

This retrospective study included 188 consecutive patients with rectal cancer without preoperative treatment who underwent surgery between 2017 and 2024. Two radiologists evaluated regional lymph nodes according to the Node-RADS 1.0 criteria and assigned each patient the highest Node-RADS category on the basis of computed tomography (CT) images. Survival and recurrence data were retrospectively collected. Interobserver agreement was assessed by weighted kappa statistics. Diagnostic performance was assessed using receiver operating characteristic (ROC) curve analysis. The Kaplan‒Meier method, log-rank test, and Cox regression were used to evaluate overall survival (OS) and recurrence-free survival (RFS). Subgroup analyses were performed according to the adjuvant therapy status.

Results

Node-RADS yielded an area under the curve (AUC) of 0.847, a sensitivity of 71.9% and a specificity of 93.5%. Substantial interobserver agreement was observed in the Node-RADS categories assigned by the two readers (weighted κ = 0.748, 95% CI: 0.692–0.805). On the basis of the optimal cutoff value, patients were divided into low-risk Node-RADS (categories 1–3) and high-risk Node-RADS (categories 4–5) groups. During a median follow-up of 52.1 months, patients with Node-RADS categories 4–5 demonstrated poorer OS and RFS than did those with lower Node-RADS categories (both p < 0.001), and this survival difference remained significant in subgroups, regardless of the postoperative adjuvant therapy status (all p < 0.05). In the multivariable analysis, Node-RADS category 4–5 was significantly associated with both poor OS (hazard ratio [HR] = 2.58, 95% CI: 1.21–5.49, p = 0.014) and poor RFS (HR = 2.99, 95% CI: 1.35–6.61, p = 0.007).

Conclusions

The Node-RADS category during the primary staging of rectal cancer was a significant predictor of postsurgical patient outcomes, but prospective and multicentre validation is needed to confirm its clinical utility. Our findings suggest that CECT-based Node-RADS could serve as a valuable tool for risk stratification and treatment planning for patients with rectal cancer.

Trial registration

This was a retrospective study and did not require clinical trial registration.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12885-026-16480-7.

Background

Colorectal cancer ranks the third most common malignancy globally, with rectal cancer cases comprising one-third of all colorectal cancer cases [1]. Despite advances in treatment strategies, rectal cancer continues to have a poor prognosis, particularly in advanced stages of the disease [2]. Multiple features identifiable on imaging, such as extramural vascular invasion (EMVI), the circumferential resection margin, the mesorectal fascia and the tumour stage, have demonstrated significant prognostic value in rectal cancer patients [3–7]. By leveraging these features, clinicians can stratify patients into risk categories and determine appropriate treatment strategies. However, despite its prognostic significance, regional nodal metastasis status has been largely overlooked in the literature [7, 8].

Given the significant impacts of the nodal status on both treatment planning and survival prediction, accurate assessment of lymph node involvement is essential. However, the accurate assessment of lymph node involvement presents unique challenges in clinical practice. On tumour imaging, lymph node involvement is often identified by an enlarged nodal diameter. However, this approach to identifying affected lymph nodes is often inadequate, as nodal enlargement can also result from inflammatory changes [9], and there is a significant overlap in size between benign and malignant lymph nodes [10]. Recent studies have demonstrated that the integration of additional morphological features, such as border irregularity and signal intensity heterogeneity, significantly improve diagnostic accuracy in nodal staging [11, 12]. Although the European Society of Gastrointestinal and Abdominal Radiology (ESGAR) consensus has established combined size and morphological criteria for accurately staging lymph nodes in rectal cancer patients, the application of these criteria remains limited by the disease- and MRI-specific approaches [13].

To standardize the imaging diagnosis of the lymph node status in all cancers, the Node Reporting and Data System (Node-RADS) was introduced in 2021 [14]. The Node-RADS evaluates lymph nodes on the basis of imaging features, including size, shape, border, and internal texture, and translates these features into a five-tier scoring system (categories 1–5). Owing to its versatile design, which makes it compatible with both MRI and contrast-enhanced CT (CECT), Node-RADS allows the standardized evaluation of lymph node status across multiple anatomical regions. This system provides a clear risk stratification from category 1 (very low) to category 5 (very high), with higher categories indicating an increased likelihood of metastasis. According to a recent meta-analysis, studies in which Node-RADS was used as a standardized lymph node evaluation tool revealed that patients assigned a higher Node-RADS category were more likely to present with malignancies [15]. However, its role in patient outcome assessment has not been fully investigated.

Although magnetic resonance imaging (MRI) remains the gold standard for rectal cancer staging, CECT is more widely available and less expensive and can acquire images faster, making it a highly practical imaging modality in clinical practice, particularly for patients with MRI contraindications or claustrophobia [16, 17]. Furthermore, studies have shown that the diagnostic performance of MRI is comparable to that of CT and enhanced ultrasound in terms of nodal staging (p = 0.16 and p = 0.18, respectively) [18]. Given its standardized approach, the Node-RADS could further improve the diagnostic performance of CT in the nodal assessment in patients with rectal cancer.

Node-RADS has shown promising diagnostic performance in patients with rectal cancer [19, 20]. Our previous study [21] evaluated the diagnostic performance of Node-RADS across multiple imaging modalities for pathological lymph node assessment in rectal cancer. However, whether CT-based Node-RADS can provide postoperative prognostic information remains unclear. Therefore, unlike the prior modality benchmarking study, the present study aimed to investigate the prognostic value of CT-based Node-RADS for postoperative outcome stratification in patients with rectal cancer.

Methods

Study participants and design

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board (IRB), and the requirement for informed consent was waived. All procedures were performed in accordance with the approved guidelines.

We retrospectively collected data from all consecutive patients with stage I–III rectal adenocarcinoma who underwent total mesorectal excision (TME) at our institution between June 2017 and January 2024. All patients who underwent TME during the study period were screened for eligibility. The inclusion criteria were as follows: (a) TME with regional lymphadenectomy, (b) rectal adenocarcinoma confirmed by postoperative pathology, and (c) abdominal contrast-enhanced CT performed within two weeks before surgery. To minimize the potential confounding effects of treatment-related alterations in tumour characteristics, patients who had previously received neoadjuvant therapy were excluded. Additional exclusion criteria included any treatment before surgery, a history of or concurrent other malignancies, poor image quality, and incomplete follow-up data. Clinical stage (I–III) was determined by combining cT and cN categories according to the 8th edition of the AJCC TNM staging system. A flowchart of the study population is provided in Fig. 1. A subset of patients in the present cohort was included in our previously published modality benchmarking study of Node-RADS in rectal cancer [21]. That study focused on comparing the diagnostic performance of Node-RADS among contrast-enhanced CT, T2-weighted MRI, and contrast-enhanced T1-weighted MRI for pathological lymph node assessment. In contrast, the present study specifically evaluates the prognostic value of CT-based Node-RADS for postoperative outcomes. Survival and recurrence analyses reported in the current manuscript were not included in the prior publication.

Fig. 1.

Flowchart of patient selection for the study

At our centre, the indication for neoadjuvant treatment was determined according to institutional practice based on established guideline principles. Patients with high-risk features for local recurrence, such as cT4 disease, cN2 disease, mesorectal fascia involvement, low rectal tumour location, extramural vascular invasion, or enlarged lateral lymph nodes, were generally recommended to undergo preoperative neoadjuvant therapy. However, owing to patient preference, some patients with these preoperative high-risk features proceeded directly to surgery without neoadjuvant treatment and were therefore also included in the present retrospective study.

For each patient, we collected data on the following preoperative clinical factors and postoperative results: demographics (age, sex, and body mass index [BMI]), preoperative serum carcinoembryonic antigen (CEA) level, carbohydrate antigen 19 − 9 (CA19-9) level, number of comorbidities (including hypertension, diabetes, coronary heart disease, stroke, liver cirrhosis, and chronic obstructive pulmonary disease), and radiological features (tumour length, tumour location, clinical tumour stage, clinical lymph node stage, clinical extramural vascular invasion (cEMVI) status, and circumferential resection margin status). The pathological features included the gross tumour type, histological tumour type, pathological grade, pathological tumour stage, pathological lymph node stage, perineural invasion status, EMVI status (pEMVI), and the total number of lymph nodes resected during surgery. The tumour length was defined as the largest diameter measured along the long axis of the bowel.

CT protocol

CECT scans were performed using three different CT scanners. The technical parameters for CT acquisition are provided in Supplementary Table 1. Patients underwent abdominal CT in the supine position within two weeks prior to surgery. All patients had been asked to fast for six hours before the CT examination; no extra bowel preparation was needed. The standard scanning range was from the diaphragm to the pubic symphysis, while an additional chest CT scan was performed in some patients to screen for pulmonary metastases. Following the noncontrast CT scan, the patients were injected with iohexol (350 mg/mL, Tianheng Pharmacy Pharmaceutical Co., Ltd., China), ioversol (350 mg/ml, Jiangsu Hengrui Pharmaceutical Co., Ltd., China), or iodixanol (320 mg/mL, Yangze River Pharmaceutical Co., Ltd., China) contrast agent (1.5–2.0 mL/kg) at a rate of 2.5–3.5 mL/s through the middle of the elbow vein, followed by a 30 mL saline flush. Images were acquired during the arterial (25–35 s) and portal-venous phases (65–70 s) via an empirical timing method. Reconstruction kernels and iterative reconstruction algorithms varied according to vendor implementation; however, all studies were reconstructed using the vendor-recommended soft-tissue/standard kernels and routine clinical iterative reconstruction settings. No ultra-sharp bone kernels or experimental high-level denoising algorithms were used.

Node‑RADS assessment

Two experienced radiologists (reader 1 and reader 2, with at least 4 and 6 years of experience in abdominal imaging, respectively) independently assessed the lymph nodes on CECT images, while being blinded to each other’s evaluations and the patient’s pathological findings. Discrepancies between the two readers were adjudicated by a senior radiologist (reader 3, with > 20 years of experience), who helped the readers reach a consensus; these consensus readings were used for subsequent analyses. No indeterminate reference standard results occurred.

Prior to image analysis, all three readers completed Node-RADS training to familiarize themselves with the evaluation criteria. Each reader then performed practice assessments on 30 rectal cancer cases that were not included in the final study cohort.

Suspicious regional lymph nodes were evaluated on portal venous-phase images. In accordance with the Node-RADS 1.0 criteria, the readers documented the characteristics (location, short-axis diameter, shape, border, texture) and final Node-RADS category of the selected highest-scoring lymph node for each patient. When multiple lymph nodes demonstrated the same highest score, the characteristics of the lymph node with the largest short-axis diameter were retained. Interobserver agreement between the two readers with respect to Node-RADS categorization was assessed. Representative examples of Node-RADS categories are shown in Supplementary Fig. 1. Although lateral pelvic lymph nodes were included in the preoperative imaging evaluation, pathologic confirmation by lateral pelvic lymph node dissection was available only in selected patients with radiologically suspicious nodes, as extended dissection was performed only when clinically indicated.

Surgery and pathological assessment

All patients underwent oncologic abdominoperineal excision or low anterior resection with regional lymphadenectomy, encompassing the mesorectum and the superior rectal artery. Because this retrospective study spanned the transition period between AJCC editions, all cases were uniformly reviewed and retrospectively staged according to the 8th edition of the AJCC TNM staging system [22] to ensure consistency in the study analysis.

In terms of pathological lymph node staging, the patients were classified as pN+ (pN1-2) if they had histologically confirmed regional nodal involvement [23]. Patients without regional nodal involvement were classified as pN- (pN0). Histopathological assessment was performed by staff pathologists as part of routine clinical care without specific knowledge of Node-RADS categorization.

Follow-up

Patients were followed up from the date of surgery until death or the last follow-up (October 2024). Routine postoperative surveillance included dynamic contrast-enhanced CT or MRI and serological tumour marker evaluation every 1–3 months during the first two years and every 3–6 months thereafter. Disease recurrence was diagnosed based on serial radiological evidence showing progressive enlargement of new or existing lesions on images from two consecutive imaging studies (CT, MRI, or positron emission tomography/computed tomography [PET/CT]), which were performed at least 4 weeks apart, or histological confirmation through biopsy or surgical resection of suspected lesions.

During follow-up, information on postoperative treatment status was also collected, including whether patients received adjuvant chemotherapy and/or radiotherapy. Because this was a retrospective study, postoperative treatment was not fully standardized. The main adjuvant chemotherapy regimens included mFOLFOX6, capecitabine monotherapy, and CAPOX/XELOX. In selected patients, raltitrexed plus oxaliplatin was used as an alternative regimen based on clinical considerations. Bevacizumab was administered only in a small number of individualized cases and was not routinely used as part of standard postoperative adjuvant therapy. In addition, a small proportion of patients received postoperative radiotherapy according to their clinical and pathological risk profiles.

Additional follow-up was conducted via telephone between September and October 2024 to update the survival status of all living patients. Overall survival (OS) was defined as the time from surgery to death, whereas recurrence-free survival (RFS) was defined as the time from surgery to the first indication of tumour recurrence or metastasis.

Statistical analysis

Statistical analyses and data visualization were used to compare clinical parameters and Node-RADS categories between the pathological lymph node status groups (pN- vs. pN+). Group differences were assessed with the Mann‒Whitney U test for continuous variables and Pearson’s chi‒square test for categorical variables. The diagnostic performance of the Node-RADS was evaluated using receiver operating characteristic (ROC) curve analysis. The optimal cutoff point for the Node-RADS categories was defined according to the Youden index. Interobserver agreement for the Node-RADS category was evaluated via Cohen’s weighted kappa coefficient. The strength of the agreement was interpreted as follows: κ values of 0.00-0.20 indicated no agreement; 0.21–0.39, weak agreement; 0.40–0.59, minimal agreement; 0.60–0.79, moderate agreement; 0.80–0.90, strong agreement; and > 0.90, almost perfect agreement [24].

To identify clinically meaningful risk groups and establish a practical risk stratification system, we evaluated the following three Node-RADS category grouping schemes: (1) a three-category scheme (NR 1–2 vs. NR 3 vs. NR 4–5); (2) a dichotomization scheme comparing NR 1–3 vs. NR 4–5; and (3) another dichotomization scheme comparing NR 1–2 vs. NR 3–5. The optimal grouping scheme was selected based on the comprehensive evaluation as follows: (1) statistical significance of the survival differences between groups via Kaplan-Meier analysis with a log-rank test; (2) clinical practicality of the grouping scheme; and (3) consistency with the diagnostic performance analysis. Survival analyses were performed via the Kaplan‒Meier method; between-group differences in survival were evaluated with the log-rank test, with p values adjusted with the Bonferroni correction method for multiple comparisons. Univariable and multivariable Cox regression analyses were performed to identify preoperative factors associated with OS and RFS. Variables with p < 0.1 in the univariable analysis were included in the multivariable model. For these analyses, variables were dichotomized based on previously established clinical thresholds as follows: the BMIs were categorized into a normal group (18.5–24.9 kg/m²) versus abnormal groups (< 18.5 or ≥ 25 kg/m²) based on the WHO classification criteria; the CEA and CA19-9 levels were dichotomized based on their laboratory-defined normal upper limits (CEA: 5 ng/mL; CA19-9: 37 U/mL), which are widely accepted clinical thresholds for distinguishing normal and elevated levels [25]. Based on previous studies and guidelines [26–28], tumour-related variables were dichotomized as follows: tumour length (≤ 5 cm vs. >5 cm), tumour location (low vs. mid-high rectum, with the low rectum defined as ≤ 5 cm from the anal verge), and clinical T stage (early disease [cT1-2] vs. locally advanced disease [cT3-4]). The cEMVI and cMRF statuses were each classified as positive versus negative based on the radiological reports. The Node-RADS categories were dichotomized (1–3 vs. 4–5) based on our preliminary analysis of the three grouping schemes. Age was included as a continuous variable in the Cox regression models, given its established status as a prognostic factor and the absence of a clear biological cutoff for dichotomization.

To compare the prognostic utility of Node-RADS classification with the standard clinical node (cN) stage, we employed Cox proportional hazards regression models for OS and RFS. The discriminative ability of each model was assessed using Harrell’s concordance index (C-index) with its 95% confidence interval. Furthermore, likelihood ratio (LR) tests were performed to evaluate the incremental prognostic value of Node-RADS classification beyond that provided by cN staging, achieved by comparing the fit of a baseline model containing only cN with an extended model that included both cN and Node-RADS. This was achieved by comparing the fit of a baseline model containing only cN with an extended model that included both cN and Node-RADS.

All the statistical analyses were conducted in IBM SPSS Statistics Software (version 26.0; IBM), MedCalc (version 15.2.2; MedCalc Software bvba), or R software (version 4.0.3). Two-tailed p < 0.05 was considered to indicate statistical significance.

Results

Patient characteristics

In this retrospective study, 216 consecutive patients with rectal adenocarcinoma who underwent preoperative contrast-enhanced CT between June 2017 and January 2024 were initially identified. Following the application of the eligibility criteria, 188 patients were included in the final analysis (median age, 67 [IQR, 59–72] years; 116 males). Among the 188 patients, 64 patients were classified as pN+ (pN1-2) and 124 as pN- (pN0) based on histological examination of regional lymph nodes. The pN+ group (median age, 66 [IQR, 61–72] years; 41 males) and the pN- group (median age, 67 [IQR, 59–73] years; 75 males) were compared. Patients in the pN+ group presented significantly higher levels of serum CEA and CA19-9, more advanced T stages, higher rates of perineural and lymphovascular invasion, poorer histological grades, and more frequent postoperative treatment (all p < 0.05). No significant differences were observed between the groups with respect to the other demographic and clinicopathological characteristics (Table 1).

Table 1.

Characteristics of patients with pN- and pN+ stage

Parameters	ALL (n = 188)	pN- (n = 124)	pN+ (n = 64)	p value
Age, years*	67 [59; 72]	67 [59; 73]	66 [61; 72]	0.630
Sex, n (%)				0.632
Male	116 (61.7%)	75 (60.5%)	41 (64.1%)
Female	72 (38.3%)	49 (39.5%)	23 (35.9%)
BMI (kg/m²)				0.888
18.5–24.9	128 (68.1%)	84 (67.7%)	44 (68.8%)
≥ 25.0 & <18.5	60 (31.9%)	40 (32.3%)	20 (31.3%)
CEA (ng/ml)				0.022
< 5	118 (62.8%)	85 (68.5%)	33 (51.6%)
≥ 5	70 (37.2%)	39 (31.5%)	31 (48.4%)
CA19-9 (U/ml)				0.019
< 37	174 (92.6%)	119 (96.0%)	55 (85.9%)
≥ 37	14 (7.4%)	5 (4.0%)	9 (14.1%)
Tumour length (cm)				0.245
< 5	139 (73.9%)	95 (76.6%)	44 (68.8%)
≥ 5	49 (26.1%)	29 (23.4%)	20 (31.3%)
Number of comorbidities				0.658
0	81 (43.1%)	52 (41.9%)	29 (45.3%)
≥ 1	107 (56.9%)	72 (58.1%)	35 (54.7%)
Clinical tumour stage, n (%)				< 0.001
cT1-2	74 (39.4%)	62 (50.0%)	12 (18.8%)
cT3-4	114 (60.6%)	62 (50.0%)	52 (81.2%)
Clinical node stage, n (%)				< 0.001
cN0	116 (61.7%)	99 (79.8%)	17 (26.6%)
cN1-2	72 (38.3%)	25 (20.2%)	47 (73.4%)
Clinical stage⋇, n (%)				< 0.001
I	65 (34.6%)	58 (46.8%)	7 (10.9%)
II	51 (27.1%)	41 (33.1%)	10 (15.6%)
III	72 (38.3%)	25 (20.2%)	47 (73.4%)
Pathological tumour stage, n (%)				< 0.001
pT1-2	71 (37.8%)	63 (50.8%)	8 (12.5%)
pT3-4	117 (62.2%)	61 (49.2%)	56 (87.5%)
Histological grade, n (%)				0.001
Grade Ⅰ/Ⅱ	163 (86.7%)	115 (92.7%)	48 (75.0%)
Grade Ⅲ/Ⅳ	25 (13.3%)	9 (7.3%)	16 (25.0%)
Perineural invasion, n (%)				0.004
Present	60 (31.9%)	30 (24.2%)	30 (46.9%)
Absent	68 (36.2%)	47 (37.9%)	21 (32.8%)
Unknown	60 (31.9%)	47 (37.9%)	13 (20.3%)
Lymphovascular invasion, n (%)				0.003
Present	35 (18.6%)	15 (12.1%)	20 (31.3%)
Absent	86 (45.7%)	58 (46.8%)	28 (43.8%)
Unknown	67 (35.6%)	51 (41.1%)	16 (25.0%)
Number of dissected lymph nodes*	13 [10; 16]	13 [9; 16]	13 [11, 17]	0.191
Postoperative hospital stay, days*	12 [9.0, 15.0]	12 [9.0, 15.0]	12 [9.0, 14.0]	0.885
Postoperative treatment				< 0.001
Present	88 (46.8%)	46 (37.1%)	42 (65.6%)
Absent	100 (53.2%)	78 (62.9%)	22 (34.4%)

CEA Carcinoembryonic antigen, CA19-9 Carbohydrate antigen 19 − 9, Comorbidities included hypertension, diabetes, coronary heart disease, stroke, liver cirrhosis, and chronic obstructive pulmonary disease

*Data are the medians with IQRs [25%, 75%]

⋇All patients were clinically M0. Clinical stage (I–III) was assigned according to the 8th edition of the AJCC TNM staging system based on the combination of cT and cN categories

Bold values indicate statistical significance (P < 0.05)

Diagnostic performance of Node-RADS

The Node-RADS categories were distributed as follows in the entire cohort (n = 188): category 1 (n = 31, 16.5%), category 2 (n = 44, 23.4%), category 3 (n = 59, 31.4%), category 4 (n = 31, 16.5%), and category 5 (n = 23, 12.2%). The distribution of Node-RADS categories differed significantly between the pN- and pN+ groups (p < 0.001). Among the pN- patients, Node-RADS categories 1–3 were predominant (93.6%, 116/124), whereas in the pN+ patients, Node-RADS categories 4–5 were more frequent (71.9%, 46/64) (Table 2; additional details regarding the Node-RADS assessment are presented in Supplementary Table 2).

Table 2.

Distribution of Node-RADS categories

Node-RADS Category	pN- n = 124 (%)	pN+ n = 64 (%)	Total n = 188 (%)
1	28 (22.6%)	3 (4.7%)	31 (16.5%)
2	39 (31.5%)	5 (7.8%)	44 (23.4%)
3	49 (39.5%)	10 (15.6%)	59 (31.4%)
4	5 (4.0%)	26 (40.6%)	31 (16.5%)
5	3 (2.4%)	20 (31.3%)	23 (12.2%)

As shown in Fig. 2, the Node-RADS category performed better than the short-axis diameter in predicting nodal metastasis, with areas under the curve (AUCs) of 0.847 (95% CI: 0.788–0.896) and 0.782 (95% CI: 0.717–0.839), respectively (DeLong test, p = 0.012). According to the results of ROC curve analysis, the optimal cutoff value of the Node-RADS was 3.5 (mathematically calculated), corresponding to a clinical cutoff of 3 points for the Node-RADS category. By dividing patients according to this cutoff value into low-grade (Node-RADS 1–3) and high-grade (Node-RADS 4–5) groups, this new classification yielded a sensitivity of 71.9% and specificity of 93.5% in predicting nodal metastasis. Assessment of the interobserver agreement for the highest Node-RADS categories between the two readers indicated substantial interreader reliability (weighted κ = 0.748, 95% CI: 0.692–0.805).

Fig. 2.

ROC curves of the Node-RADS category and short-axis diameter in predicting lymph node metastasis. The blue solid line represents the Node-RADS category (AUC = 0.847), and the red dashed line represents the short-axis diameter (AUC = 0.783)

Survival analyses and prognostic value of Node-RADS

The next phase of our analysis focused on evaluating long-term outcomes (OS and RFS) and their associations with Node-RADS categories. The median follow-up duration was 52.1 months (interquartile range, 29.3–68.9 months). During this period, 33 patients (17.6%) died from cancer or other causes, and 29 patients (15.4%) experienced local recurrence or distant metastasis. The 3-year RFS and OS rates were 78.7% and 75.6%, respectively, whereas the corresponding 5-year rates were 67.0% and 63.7%, respectively. Patients with pN+ status exhibited poorer survival outcomes than did those with pN- status (median OS: 39.5 vs. 49.7 months, p = 0.002; median RFS: 33.2 vs. 49.1 months, p < 0.001).

Survival analyses were conducted for multiple Node-RADS category grouping schemes, as illustrated in Fig. 3 and summarized in Table 3. Figures 3a and b show the result of the three-category analysis and indicates showed significant survival differences between groups NR 1–2 and NR 4–5 in terms of both OS (p = 0.001) and RFS (p = 0.004) and between NR 3 and NR 4–5 in terms of RFS (p < 0.001), whereas no significant survival differences were identified between the other pairs of groups and outcomes (p > 0.05). On the basis of these findings, we evaluated two dichotomization schemes. Kaplan‒Meier analysis for both grouping methods revealed that both schemes yielded significant prognostic value for OS and RFS (Fig. 3c-f, all p < 0.05). The dichotomization of NR 1–3 versus NR 4–5 was selected as the optimal grouping scheme because (1) this grouping demonstrated significant prognostic stratification for both OS and RFS (p < 0.05), and its findings aligned with those of the three-category analysis that NR 4–5 consistently showed significant survival differences from the other categories, while NR 1–2 and NR 3 showed similar survival patterns; (2) a dichotomous classification system is more practical for clinical decision-making than a three-tier system; (3) this cutoff corresponds well with the diagnostic threshold identified by our ROC curve analysis for detecting pathologically positive nodes; and (4) initial analysis showed no significant differences in OS (p = 0.907) or RFS (p = 0.526) between patients with Node-RADS category 4 (n = 31) and Node-RADS category 5 (n = 23) (Supplementary Fig. 2). Therefore, we established a two-tier risk stratification system, classifying patients with Node-RADS categories 1–3 as a low-risk group and patients with categories 4–5 as a high-risk group for subsequent analyses. It should be noted that the lack of a significant survival difference between Node-RADS Category 4 and Category 5 should be interpreted with caution, as the relatively small sample size in these high-risk subgroups (combined n = 54) may have limited the statistical power to detect a true difference. Patients with high Node-RADS categories (4–5) had significantly worse outcomes than those with low scores (1–3), with 5-year OS rates of 59.1% versus 83.2% and 5-year RFS rates of 56.9% versus 91.1%, respectively (both p < 0.001). These differences were reflected in the median survival times, with shorter OS (39.4 vs. 49.4 months) and RFS (35.5 vs. 48.4 months) in the high Node-RADS group. These analyses demonstrated that Node-RADS categories could effectively stratify patients into low-risk (categories 1–3) and high-risk (categories 4–5) groups, with the high-risk group showing significantly worse survival outcomes in terms of both OS (p < 0.001) and RFS (p < 0.001).

Fig. 3.

Survival analysis stratified by Node-RADS categories. (a, b) Overall survival (OS) (a) and recurrence-free survival (RFS) (b) curves for Node-RADS categories 1–2, 3, and 4–5; p values are adjusted for pairwise comparisons with the Bonferroni correction. (c, d) Overall survival (c) and recurrence-free survival (d). curves for Node-RADS categories 1–3 versus 4–5. (e, f) Overall survival (e) and recurrence-free survival (f) curves for Node-RADS categories 1–2 versus 3–5. p values were determined with the log-rank test. The numbers of patients at risk are shown below each graph

Table 3.

Survival analysis of different Node-RADS category grouping schemes

Analysis Type	Category Groups	Comparison Groups	OS p value	RFS p value
Three-category Analysis (Scheme 1)	NR 1–2 vs. NR 3 vs. NR 4–5	NR 1–2 vs. NR 3	p = 0.503†	p = 1.0†
		NR 3 vs. NR 4–5	p = 0.081†	p < 0.001†
		NR 1–2 vs. NR 4–5	p = 0.001†	p = 0.004†
Two-category Analyses	Scheme 2	NR 1–3 vs. 4–5	p < 0.001	p < 0.001
	Scheme 3	NR 1–2 vs. 3–5	p = 0.007	p = 0.020

^† p values were adjusted with the Bonferroni correction method

Bold values indicate statistical significance (P < 0.05)

To further validate the prognostic value of this dichotomization, we performed additional subgroup analyses. When grouped by the postoperative treatment status, the Node-RADS risk groups (1–3 vs. 4–5) maintained strong prognostic value among untreated patients for both OS and RFS (both p < 0.001, Fig. 4a and b). In the treated cohort, the survival differences between the Node-RADS categories remained significant for OS (p = 0.017) and RFS (p = 0.032, Fig. 4c and d). We further performed an exploratory subgroup analysis in patients with Node-RADS category 4–5 to compare survival according to postoperative adjuvant chemotherapy status. Although patients who received adjuvant chemotherapy showed numerically better RFS and OS, the differences were not statistically significant (log-rank p = 0.96 and p = 0.24, respectively). Given the limited sample size of this subgroup (n = 54), these findings should be interpreted with caution and are presented in Supplementary Fig. 3.

Fig. 4.

Survival curves by Node-RADS (1–3 vs. 4–5) for non-adjuvant and adjuvant groups. Overall survival (a) and recurrence-free survival (b) curves for Group A (no adjuvant therapy). Overall survival (c) and recurrence-free survival (d) curves for Group B (adjuvant therapy). P values were calculated with the log-rank test. The numbers of patients at risk are shown below each graph

For survival analyses, we evaluated the prognostic significance of demographic features (age and sex), clinical parameters (BMI, CEA, and CA19-9), tumour characteristics (length, location, and cT stage), and imaging findings (cEMVI, cMRF, and Node-RADS category). The results of univariable and multivariable Cox regression analyses for OS and RFS are shown in Table 4 and illustrated in Fig. 5. In the univariable analysis, age, CA19-9 level, cT stage, cEMVI status, and Node-RADS category were significantly associated with OS, whereas CEA level, comorbidity, clinical tumour stage, clinical EMVI status, and Node-RADS category were significantly associated with RFS. Variables with p < 0.1 in the univariable Cox regression analysis were considered candidates for the multivariable analysis. Both age (HR = 1.06, 95% CI: 1.02–1.11, p = 0.003) and Node-RADS category 4–5 (HR = 2.58, 95% CI: 1.21–5.49, p = 0.014) remained significant predictors of a worse OS in the multivariable analysis, while for RFS, Node-RADS category 4–5 was significantly associated with poor outcomes (HR = 2.99, 95% CI: 1.35–6.61, p = 0.007). These findings demonstrated that the Node-RADS category, along with the age for OS, was an independent preoperative prognostic factor, highlighting its potential value in primary risk stratification.

Table 4.

Univariable and multivariable analyses of factors affecting overall survival and recurrence-free survival

Variables	OS Univariable Analysis				OS Multivariable Analysis				RFS Univariable Analysis				RFS Multivariable Analysis
	HR	95% CI		p value	HR	95% CI		p value	HR	95% CI		p value	HR	95% CI		p value
Age (years)	1.06	1.02	1.1	0.006	1.06	1.02	1.11	0.003	1.03	0.99	1.07	0.12
Sex (male)	1.58	0.73	3.4	0.245					1.62	0.72	3.66	0.246
BMI (≥25 & <18.5 kg/m²)	0.95	0.45	2	0.899					0.81	0.36	1.83	0.61
CEA level (≥5 ng/mL)	1.6	0.81	3.16	0.179					2.55	1.22	5.34	0.013	1.87	0.88	3.98	0.106
CA19-9 level (≥37 U/mL)	2.36	0.91	6.11	0.078	2.04	0.72	5.73	0.179	2.38	0.83	6.84	0.107
Comorbidities (≥1)	0.9	0.45	1.79	0.767					0.52	0.25	1.09	0.083	0.55	0.26	1.17	0.123
Location (Distal)	0.98	0.44	2.18	0.963					0.92	0.37	2.26	0.855
Clinical Tumour Stage (3-4)	2.11	0.95	4.68	0.066	1.31	0.53	3.24	0.566	4.41	1.53	12.67	0.006	2.78	0.91	8.47	0.073
Clinical EMVI Status (+)	1.98	0.92	4.26	0.082	1.73	0.71	4.23	0.229	2.3	1.05	5.05	0.038	1.11	0.48	2.57	0.806
Clinical MRF Status (+)	1.51	0.68	3.35	0.317					1.92	0.85	4.33	0.118
Tumour Size (≥ 5 cm)	0.73	0.32	1.69	0.468					0.9	0.39	2.11	0.812
Node-RADS Category (4-5)	3.22	1.62	6.38	0.001	2.58	1.21	5.49	0.014	4.63	2.18	9.81	<0.001	2.99	1.35	6.61	0.007

Fig. 5.

Forest plots representing univariable and multivariable Cox regression analyses for overall survival (OS, upper panel) and recurrence-free survival (RFS, lower panel)

To directly compare the prognostic performance of standard cN stage and Node-RADS, Harrell’s C-index was calculated separately for cN and Node-RADS. For OS, the C-index was 0.569 (95% CI: 0.478–0.659) for cN and 0.630 (95% CI: 0.541–0.719) for Node-RADS. For RFS, the corresponding C-index values were 0.587 (95% CI: 0.494–0.681) for cN and 0.659 (95% CI: 0.567–0.751) for Node-RADS, indicating better prognostic discrimination of Node-RADS than cN. The LR test further demonstrated that the addition of Node-RADS significantly improved the model’s fit for both OS (p = 0.003) and RFS (p < 0.001), confirming its prognostic contribution. The comparative results are presented in Table 5.

Table 5.

Comparative prognostic performance of clinical N stage (cN) and Node-RADS in predicting overall survival and recurrence-free survival

Outcome	Measure	clinical N stage	Node‑RADS
Overall survival (OS)	C index (95% CI)	0.569 (0.478–0.659)	0.630 (0.541–0.719)
	Likelihood ratio† χ² (p)	-	8.600 (p = 0.003)
Recurrence-free survival (RFS)	C index (95% CI)	0.587 (0.494–0.681)	0.659 (0.567–0.751)
	Likelihood ratio† χ² (p)	-	11.604 (p < 0.001)

†The likelihood ratio (LR) χ² values represent the comparison between the baseline Cox proportional hazards model containing clinical N (cN) stage alone and the nested extended model containing both cN stage and Node-RADS. The statistically significant LR tests indicate the incremental prognostic value of Node-RADS beyond cN stage

Discussion

Although nodal staging is crucial for treatment planning and prognostication in rectal cancer, standardized reporting of lymph node features on CT has been lacking. The Node-RADS was recently proposed to address this need, but its diagnostic performance and prognostic value remain to be validated. This study demonstrated that the Node-RADS showed excellent performance in predicting nodal metastasis, with an AUC of 0.847 (95% CI: 0.788–0.896). Furthermore, the Node-RADS categories independently predicted survival outcomes, with high-risk categories (4–5) associated with worse OS (HR = 2.58, 95% CI: 1.21–5.49) and RFS (HR = 2.99, 95% CI: 1.35–6.61). These findings establish CT-derived Node-RADS as a reliable tool for both nodal staging and risk stratification in rectal cancer.

Accurate lymph node staging informs the selection of appropriate neoadjuvant treatment regimen and other key decision-making [29]. In the National Comprehensive Cancer Network (NCCN) guidelines, patients with node-positive disease, regardless of T stage, are categorized as having locally advanced rectal cancer (clinical stage III-IV) and are recommended to undergo neoadjuvant therapy prior to surgery [26]. However, distinguishing node-positive disease from node-negative disease preoperatively is challenging. A recent UK multicentre audit revealed substantial variability in lymph node assessment methods, identifying the use of different combinations of criteria, including ESGAR guidelines, chemical shifts, node signal intensity, and morphological features among different radiologists [30]. To address this variability and provide a more universally applicable framework for both CT and MRI, the Node-RADS was proposed [14]. In rectal cancer specifically, the Node-RADS has been validated for nodal staging, demonstrating superior diagnostic performance to the ESGAR classification and greater accuracy than the size criteria alone [19, 20]. Our recent modality benchmarking study further confirmed the diagnostic utility of Node-RADS across CECT, T2-weighted MRI, and contrast-enhanced T1-weighted MRI for primary lymph node assessment in rectal cancer [21] (with partial patient overlap with the present study). Our study confirms this advantage, showing that the CT-based Node-RADS achieved higher diagnostic accuracy than the size criteria did (AUC 0.847 vs. 0.782, p = 0.012), which highlights that the Node-RADS approach of integrating morphological features with size measurements improves nodal staging accuracy [31, 32]. This superior performance can be attributed to the incorporation of multiple imaging features, which have been shown to be more reliable indicators of metastatic involvement than the size alone [12, 16]. This enhanced diagnostic capability could potentially reduce both over- and undertreatment in clinical practice, as there is a broad size overlap between histologically benign (2–10 mm) and malignant (3–15 mm) lymph nodes [11]. These results demonstrate the robust diagnostic accuracy of Node-RADS in assessing the nodal stage in rectal cancer. Future studies should focus on further exploring its integration into routine clinical workflows to optimize staging accuracy and treatment decision-making.

In our cohort of patients who underwent a TME without neoadjuvant therapy, the Node-RADS categories derived from preoperative CT demonstrated significant prognostic value. High-risk Node-RADS categories (4–5) were associated with significantly worse OS and RFS than low-risk categories were (all p < 0.001). This finding is particularly relevant given that lymph node metastasis is a critical prognostic indicator in rectal cancer, with the Norwegian Rectal Cancer Project showing increased local recurrence rates from 8% in node-negative to 16–30% in node-positive T3 rectal cancer patients undergoing a TME alone [33]. The strong prognostic performance of the Node-RADS in our study likely stems from its comprehensive evaluation system that incorporates both traditional size criteria and specific morphological features, with the latter demonstrating independent prognostic value in previous studies. For example, extranodal extension, which manifests as the “irregular border” feature in the Node-RADS, has been independently associated with a poor 3-year disease-free survival (HR, 3.00; p = 0.020) [34]. The robust performance of the Node-RADS in our study aligns with the findings in other malignancies, such as renal cell carcinoma, wherein high Node-RADS categories similarly predicted poorer progression-free survival and OS (both p < 0.001) [35]. These results suggest that the Node-RADS could serve as an effective preoperative risk stratification tool, potentially facilitating patient selection for multimodal therapy.

Adjuvant therapy has long been recognized as a pivotal treatment strategy for improving survival outcomes, particularly in patients with locally advanced rectal cancer [36]. In our study, we further explored the interplay between adjuvant therapy and Node-RADS risk categories concerning their prognostic impact. Our subgroup analysis, stratified by adjuvant therapy status, revealed that patients classified into the high-risk Node-RADS categories (4–5) exhibited significantly worse survival compared to those in the low-risk group. This disparity persisted irrespective of whether patients received adjuvant therapy (all p < 0.05). These findings strongly suggest that the presence of high-risk Node-RADS features may be indicative of an inherently aggressive tumor biology. Such tumors appear to be associated with poor outcomes even with the administration of adjuvant therapy, thereby underscoring the critical need for more intensified treatment strategies in this specific subgroup of patients. Indeed, recent trials have demonstrated encouraging results with intensified neoadjuvant regimens for locally advanced rectal cancer, achieving an average pathological complete response rate (ypT0N0M0) of 17.5% [37]. While further subgroup analyses within our cohort indicated similar outcomes between Node-RADS categories 4 and 5, suggesting comparable risk levels, these observations should be interpreted with caution given our limited sample size. Although our study has effectively demonstrated the prognostic value of Node-RADS in rectal cancer patients undergoing primary surgery, additional extensive research is warranted to fully investigate the role of Node-RADS categories in the prognostic evaluation of rectal cancer in the context of neoadjuvant therapy.

Like other RADSs, the Node-RADS requires continuous optimization. From a technical perspective, multiple investigations have demonstrated the utility of deep learning models for investigating clinical endpoints, particularly regarding prognostic assessment and survival outcomes [38]. While the current interobserver agreement of Node-RADS is acceptable, emerging deep learning technologies may further improve its reproducibility, similar to the enhanced performance demonstrated in Prostate Imaging Reporting and Data System (PI-RADS) assessment [39]. Such advancements could improve the standardization of lymph node evaluation and clinical applicability.

Because patients who received neoadjuvant therapy were excluded, the present findings should be interpreted in the context of a treatment-naive surgical cohort. This design allowed us to evaluate the association between baseline preoperative Node-RADS assessment, pathological nodal status, and prognosis without the confounding effects of therapy-related changes. However, the applicability of Node-RADS in the modern total neoadjuvant therapy era remains uncertain. In post-treatment restaging, treatment-induced fibrosis, nodal shrinkage, and other morphological alterations may substantially affect the imaging features incorporated into Node-RADS and thereby influence its diagnostic performance. Therefore, the role of Node-RADS in the post-neoadjuvant setting requires separate validation in dedicated prospective or external cohorts.

An additional issue is the presence of tumour deposits (TDs). TDs are clinically relevant in rectal cancer, are linked to worse oncologic outcomes, and are incorporated into the current nodal staging framework [40]; therefore, their inclusion in the pathological positive group was appropriate for this prognostic analysis. At the same time, TDs should be reported separately in routine rectal cancer assessment. Because TDs and metastatic lymph nodes are often difficult to differentiate on CT, especially when presenting as irregular or indistinct perirectal nodules, their presence may complicate lesion-level interpretation of Node-RADS features. Although this limitation is unlikely to substantially change the main prognostic conclusions of the present study, it highlights an important practical point: radiologists using Node-RADS and clinicians interpreting these findings should remain aware of the possible presence and significance of TDs. Future studies with lesion-level radiologic-pathologic correlation are needed to determine whether TDs should be incorporated into Node-RADS or evaluated as a separate category.

This study has several limitations that warrant discussion. First, its retrospective single-centre design introduces inherent selection bias and confounding factors, and therefore its findings need to be externally validated through prospective multicentre studies. Second, the absence of a statistically significant survival difference in subgroups should be interpreted with caution, as the relatively small sample size may have limited the statistical power to detect a true difference. Third, our decision to perform patient-level rather than node-level analyses might have affected the accuracy of the evaluations, as direct radiologic‒pathologic associations for individual lymph nodes were not obtained in this study. In addition, the interobserver consistency of node-level Node-RADS features, including the potential influence of different reconstruction kernels and reconstruction algorithms, warrants further investigation. Additionally, the inclusion of patients with tumour deposits in the pN+ group, which may not be reliably distinguished from lymph nodes on CT imaging, could have influence the accuracy of the Node-RADS assessment. Nevertheless, since current treatment decisions are primarily based on the overall N stage rather than the status of individual lymph nodes, our patient-level analysis approach remains clinically relevant.

Conclusions

In conclusion, CECT-based Node-RADS v1.0 demonstrated robust diagnostic performance and prognostic value in primary rectal cancer staging. Specifically, higher categories (Node-RADS 4–5) indicated worse survival outcomes, implying that patients with Node-RADS category ≥ 4 may benefit from more intensive surveillance and treatment strategies. While these results are encouraging, prospective validation in larger cohorts would be beneficial for confirming the clinical utility of this system, particularly in evaluating the potential role of the Node-RADS in guiding treatment decisions.

Abbreviations

Node-RADS

Node Reporting and Data System

CECT

Contrast-enhanced computed tomography

ROC

Receiver-operating characteristic curve

OS

Overall survival

RFS

Recurrence-free survival

AUC

Area under the curve

HR

Hazard ratio

EMVI

Extramural vascular invasion

ESGAR

European Society of Gastrointestinal and Abdominal Radiology

MRI

Magnetic resonance imaging

TME

Total mesorectal excision

BMI

Body mass index

CEA

Carcinoembryonic antigen

CA19-9

Carbohydrate antigen 19-9

Authors’ contributions

Lr. Z, Dq. W, L. J and Zj. Z conceived and designed the study. Fg. Z and X. T acquired the data. Ht. Z, Lr. Z and Dq. W implemented quality control of data. L. J and Zj. Z did the image analysis. L. J and Fg. Z performed the statistical analysis. Zj. Z and X. T made figures and tables. L. J and Zj. Z prepared the first draft of the manuscript. Ht. Z, Lr. Z and Dq. W reviewed the manuscript. Lr. Z and Dq. W had direct access to and verified the data. All authors read and approved the final manuscript.

Funding

This study was supported by National Natural Science Foundation of China (Grant No. 82272066); Key Project of Jiangsu Provincial Health Commission (grant number K2019024); Natural Science Foundation of Jiangsu Province (grant number BK20191223); and Scientific research project of the Jinshan Talent Training Program for high-level leading Talents (grant number YLJ202111).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the ethical review board of the Affiliated Hospital of Jiangsu University with waiver of informed consent, approval number SWYXLL20190225-3. This study was conducted in accordance with the Declaration of Helsinki. All methods were performed in accordance with the relevant guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.