Comparison of 2D and 3D measured iodine concentration of gastric adenocarcinoma on spectral-CT and their relationship with perineural invasion

Abstract

Background

To determine the optimal region of interest (ROI) measurement strategy in spectral computed tomography (CT) for the preoperative prediction of perineural invasion (PNI) in gastric adenocarcinoma.

Methods

This retrospective study analyzed 91 gastric adenocarcinoma patients undergoing triple-phase (arterial, venous, and delayed phase; AP/VP/DP) contrast-enhanced spectral CT within two weeks before surgery. Patients were divided into PNI-positive and PNI-negative groups based on pathological findings. Iodine concentration (IC) values were measured using two free-hand ROI approaches: a two-dimensional ROI (2D-ROI) and a three-dimensional volumetric ROI (3D-ROI). Normalized IC (nIC) was also calculated. Consistency and correlation between the two ROI measurements were assessed. Differences in clinicopathological and CT features between the PNI-positive and PNI-negative groups were analyzed. The area under the receiver operating characteristic curve (AUC) was used to evaluate predictive performance. Logistic regression identified independent risk factors for PNI.

Results

The ICs measured by 2D and 3D-ROI showed excellent consistency (ICC = 0.803–0.853) and correlation (r = 0.777–0.797), though 2D-ROI yielded higher values (all P < 0.05). 2D-ICVP, 2D-ICDP, 2D-nICVP, 2D-nICDP and 3D-nICDP, were significantly higher in PNI-positive GC (P < 0.05). 2D-nICDP had the highest predictive AUC (0.761), outperforming other parameters (AUC = 0.622–0.670; P < 0.05). Multivariable analysis confirmed 2D-nICDP as the only independent PNI predictor.

Conclusions

Although the consistency and correlation between 2D and 3D-ROIs were excellent, only 2D-nICDP was an independent predictor for PNI in gastric cancer (GC) and demonstrated superior predictive performance.

Introduction

Gastric cancer (GC) is the fifth most common malignancy worldwide and the fifth leading cause of cancer-related deaths globally [1]. Surgical resection remains the primary curative treatment for patients with GC, frequently accompanied by radiotherapy, chemotherapy, and other comprehensive treatments [2–4]. Despite these interventions, the risk of tumor recurrence within 1 year after surgery exceeds 50% [5, 6].

Perineural invasion (PNI) constitutes a distinct metastatic pathway in GC, characterized by the invasion of tumor cells into neural structures and subsequent dissemination along nerve sheaths [7, 8]. Previous studies have identified PNI as an independent risk factor for recurrence and metastasis [9, 10]. Currently, PNI status can only be diagnosed through postoperative pathology. Consequently, there is an urgent need to investigate noninvasive and effective preoperative methods for assessing the PNI status in GC patients [11–13].

Spectral-computed tomography (CT) is a functional imaging modality that provides a range of quantitative parameters, allowing for the amplification of subtle differences within tissues [14, 15]. A previous study has developed a spectral CT nomogram to predict PNI preoperatively in locally advanced GC with a specificity of 85.3% [16]. A study by Ren et al. [17] demonstrated that combining clinical and spectral-CT parameters effectively evaluates concurrent lymphovascular invasion and PNI status. Notably, previous studies have employed two-dimensional region of interest (2D-ROI) based iodine concentration (IC) measurements and addressed the need for the incorporation of three-dimensional volumetric ROI (3D-ROI) measurements [16–19]. GC is characterized by marked heterogeneity, thus, analyzing the tumor in 3D can provide more comprehensive information than the 2D-ROI method [20]. However, 3D-ROI based IC measurements are time-consuming and labor-intensive, limiting their widespread adoption in clinical practice. Few studies have compared IC values obtained using 2D and 3D-ROI approaches, and even fewer have examined their correlations with PNI in GC.

Therefore, this study aims to evaluate the feasibility of preoperatively predicting PNI using both 2D and 3D measurement approaches with spectral CT in GC, and to determine the optimal ROI methodology for clinical practice.

Methods

Patients

This was a retrospective study approved by the institutional review board of our hospital, and the requirement for informed consent was waived. Patients pathologically diagnosed with gastric adenocarcinoma (clinical stage cT1-4aN0/+M0) admitted between February 2024 and September 2024 were enrolled. None had received anti-cancer treatment before surgery. As shown in Fig. 1, the inclusion criteria were as follows: (1) Pathologically confirmed gastric adenocarcinoma and available PNI status; (2) Abdominal triple-phase enhanced spectral CT scanning performed within two weeks before surgery; The exclusion criteria included: (1) Poor CT image quality; (2) Small gastric adenocarcinoma lesion with maximal diameter less than 10 mm. The study ultimately included 91 patients (68 males and 23 females; mean age: 61.75 ± 8.72 years, range: 38 − 80 years), who were classified into PNI-positive and PNI-negative groups according to postoperative results. We also recorded their clinicopathological characteristics, including age, sex, tumor location, size, histodiffferentiation, pathological T stage, N stage, and Lauren’s classification.

Fig. 1.

Flowchart of the inclusion and exclusion criteria. PNI, perineural invasion

Image examination

All patients underwent abdominal triple-phase contrast-enhanced spectral CT scanning (Revolution CT, GE Healthcare) before surgery. Before CT examination, patients were instructed to fast for 8 h [21] and drink 600–1000 mL of warm water to fill the stomach cavity. The contrast agent (Ultravist 370, Bayer Schering Pharma) was administered at a rate of 3.0 mL/s using a pump injector (Urich REF XD 2060-Touch, Ulrich Medical), with a dosage of 1.5 mL/kg. Arterial-phase (AP), venous-phase (VP), and delayed-phase (DP) scans were acquired at 30s, 60s, and 90s after contrast injection, respectively [16]. The scanning parameters were as follows: tube voltage, 80 to 140 kVp fast-switching within a 0.5ms switch time; tube current, 400 mA; width of detector, 80 mm; pitch, 0.992; rotation time, 0.5s; detector collimation, 64 × 0.625 mm; matrix size, 512 × 512; field of view, 450 mm×450 mm; slice thickness, 5 mm.

Image interpretation

Two senior radiologists with 8 and 9 years of experience, blinded to pathologic information, independently reviewed all spectral CT images. Two ROI methods were used in this study to measure iodine concentration (IC). For 2D-ROI analysis, CT images were evaluated using GSI viewer software on a GE AW4.7 workstation. The two radiologists independently selected the largest cross-sectional slice of the tumor and then outlined a free-hand, phase-based ROI along the tumor’s outer margin on IC maps, avoiding vessels. Aortic ICs were obtained by placing a circular ROI within the aorta at the same level. For 3D-ROI analysis, IC maps were digitally transferred to ITK SNAP software (v3.8.0, http://www.itksnap.org). The ROI was manually delineated around the entire tumor margin on each slice of the IC maps, and the IC of 3D-ROI was automatically generated by the software (Fig. 2).The normalized ICs (nICs) for both 2D and 3D-ROIs were acquired by dividing the tumor IC by the aortic IC (nIC = IC _tumor/IC _aorta) [22, 23].

Fig. 2.

Schematic diagram of 3D tumor delineation for IC values. Region of interest (ROI) delineation on the first tumor slice (A), on the tumor’s largest cross-sectional slice (B), on the last tumor slice (C). (D) Coronal view displaying the ROI. (E) Sagittal view displaying the ROI. (F) Fused 3D-ROI reconstruction

Pathological PNI diagnosis

PNI information from 91 patients was collected using surgical specimens. All specimens were derived from primary GCs, sectioned to a thickness of 4 μm, and analyzed for pathological indicators using hematoxylin and eosin (HE) staining. A senior gastrointestinal pathologist with 12 years of experience independently reviewed the slides. PNI-positive was defined as the infiltration of tumor cells into the nerves and/or the endoneurial, perineurial, and epineurial spaces within the neuronal sheath [24, 25].

Statistical analysis

SPSS statistical software (version 27.0, SPSS, IBM) and MedCalc software (version 19.0) were used for the analysis. The Kolmogorov-Smirnov test was used to assess the normality of the data. The quantitative variables were expressed as mean ± standard deviations (SD) or medians and interquartile ranges [M (Q1, Q3)]. Data with normal distributions or nonnormal distributions were compared using the Student’s t-test or Mann-Whitney U test, respectively, and categorical variables were assessed using the chi-squared test. The intraclass correlation coefficient (ICC) value was used to analyze the consistency of ICs measured in 2D and 3D-ROI, and the interobserver agreement for 2D measurements. ICC < 0.4 indicates poor consistency, 0.40–0.59 indicates moderate consistency, 0.60–0.74 indicates good consistency, and 0.75-1.00 indicates excellent consistency. The Spearman correlation coefficient was used to analyze the correlation between the two ROI approaches, with r < 0.4 indicating poor correlation, 0.40–0.59 indicating moderate correlation, 0.60–0.74 indicating good correlation, and 0.75-1.00 indicating excellent correlation. The predictive performance was assessed by the area under the receiver operating characteristic curve (AUC) with 95% confidence intervals (CIs). Multivariable logistic regression analysis was used to identify independent predictors of PNI. A two-sided P value < 0.05 was considered statistically significant.

Results

Patients and histopathology

Among the 91 patients with gastric adenocarcinoma, 36 patients were diagnosed as PNI-positive, and 55 patients were PNI-negative. The patients’ clinicopathological characteristics were shown in Table 1.

Table 1.

Clinicopathological characteristics of patients with gastric adenocarcinomas between PNI-positive and PNI-negative groups

Characteristics	PNI-positive (n = 36)	PNI-negative (n = 55)	t/Z/χ2	P
Gender			0.198	0.657
Male	26	42
Female	10	13
Age (years)	59 (54, 68)	66 (55, 70)	-1.933	0.053
Tumor location			2.037	0.361
Cardia/fundus	18	31
Body	11	19
Antrum	7	5
pT			24.894	< 0.001
1	1	23
2	5	11
3	20	19
4a	10	2
pN			8.093	0.004
Negative	10	32
Positive	26	23
Histodifferentiation			11.149	0.004
Good	24	19
Moderate	12	29
Poor	0	7
Lauren subtype			15.641	< 0.001
Intestinal	2	24
Mixed	18	18
Diffused	16	13
Lymphovascular invasion			16.938	< 0.001
Negative	9	38
Positive	27	17

PNI, perineural invasion; (−), negative; (+), positive; PN, positive node numbers

Comparison of ICs for the 2D and 3D-ROI methods

There were significant positive correlations (all P < 0.001) between 2D-ICAP and 3D-ICAP, 2D-ICVP and 3D-ICVP, 2D-ICDP and 3D-ICDP, with r values of 0.779 (95% CI: 0.680–0.851), 0.777 (95% CI: 0.677–0.849), 0.797 (95% CI: 0.704–0.863), respectively. The consistency analysis results showed excellent consistency between 2D-ICAP and 3D-ICAP, 2D-ICVP and 3D-ICVP, 2D-ICDP and 3D-ICDP, with ICC values of 0.803 (95% CI: 0.716–0.866), 0.853 (95% CI: 0.785–0.901), and 0.844 (95% CI: 0.773–0.894), respectively. However, there were significant statistical differences in average ICs between the 2D-ROI and 3D-ROI methods in the AP, VP and DP (Table 2). An excellent interobserver agreement was achieved between the two readers for the ICC values of ICAP, ICVP, ICDP, nICAP, nICVP, and nICDP, which were found to be 0.933 (0.901–0.955), 0.921 (0.883–0.947), 0.922 (0.883–0.948), 0.920 (0.881–0.946), 0.910 (0.867–0.940), and 0.909 (0.866–0.939), respectively.

Table 2.

The comparison of ICs for all gastric adenocarcinomas in two ROI methods

Parameters	2D-ROI	3D-ROI	Z/t	P
ICAP (mg/mL)	1.50 ± 0.65	1.27 ± 0.55	-2.844	0.004
ICVP (mg/mL)	2.11 ± 0.81	1.78 ± 0.66	-3.708	< 0.001
ICDP (mg/mL)	1.97 ± 0.74	1.69 ± 0.62	2.787	0.006

AP, arterial phase; VP, venous phase; DP, delayed phase; IC, iodine concentration; 2D, two-dimensional; 3D, three-dimensional; ROI, region of interest

The 3D-nICDP, as well as the 2D-ICVP, 2D-ICDP, 2D-nICVP and 2D-nICDP, were significantly higher in the PNI-positive group than that in the PNI-negative group (P < 0.05) (Fig. 3; Table 3). No significant differences were observed in tumor diameter and thickness between the PNI-positive group and PNI-negative group.

Fig. 3.

(A-C) Triple-phase enhanced spectral CT-based iodine maps in a perineural invasion-positive case. (A) Arterial phase iodine map, the IC value was 2.95 (mg/mL). (B) Venous phase iodine map, the IC value was 2.63 (mg/mL). (C) Delayed phase iodine map, the IC value was 2.23 (mg/mL). (D-F) Triple enhanced spectral CT-based iodine maps in a perineural invasion-negative case. (D) Arterial phase iodine map, the IC value was 1.93 (mg/mL). (E) Venous phase iodine map, the IC value was 1.72 (mg/mL). (F) Delayed phase iodine map, the IC value was 1.86 (mg/mL)

Table 3.

Comparison of CT parameters between PNI–positive and PNI–negative groups

Parameters	PNI (−) (n = 55)	PNI (+) (n = 36)	t/χ2 /Z	P
Borrmann classification			5.709	0.058
I	8	2
II	21	8
III	26	26
Diameter (mm)	47.44 ± 21.97	55.47 ± 18.67	1.804	0.075
Thickness (mm)	18.58 ± 6.39	20.62 ± 6.14	1.512	0.134
CT-reported T staging			2.937	0.401
1	5	1
2	9	3
3	32	25
4a	9	7
CT-reported LN status			0.059	0.808
Negative	17	12
Positive	38	24
2D-ICAP (mg/mL)	1.47 (1.05, 1.86)	1.36 (1.08, 1.80)	-0.141	0.849
3D-ICAP (mg/mL)	1.19 (0.93, 1.51)	1.14 (0.90, 1.14)	-0.410	0.682
2D-ICVP (mg/mL)	1.90 (1.58, 2.33)	2.19 (1.77, 2.69)	1.964	0.049
3D-ICVP (mg/mL)	1.61 (1.42, 1.99)	1.68 (1.43, 2.22)	0.767	0.443
2D-ICDP (mg/mL)	1.82 ± 0.77	2.18 ± 0.64	2.237	0.028
3D-ICDP (mg/mL)	1.52 (1.29, 1.82)	1.74 (1.41, 1.91)	1.826	0.068
2D-nICAP	0.15 (0.11, 0.18)	0.15 (0.12, 0.19)	0.525	0.600
3D-nICAP	0.12 (0.09, 0.15)	0.13 (0.10, 0.17)	0.875	0.382
2D-nICVP	0.42 (0.35, 0.53)	0.48 (0.37, 0.60)	2.242	0.025
3D-nICVP	0.36 (0.31, 0.44)	0.40 (0.32, 0.51)	1.182	0.237
2D-nICDP	0.49 ± 0.17	0.62 ± 0.13	3.911	< 0.001
3D-nICDP	0.44 ± 0.14	0.51 ± 0.13	2.319	0.023

AP, arterial phase; VP, venous phase; DP, delayed phase; IC, iodine concentration; nIC, normalized iodine concentration; PNI, perineural invasion; (−), negative; (+), positive; 2D, two-dimensional; 3D, three-dimensional

ROC-curve analysis and the optimal ROI strategy

The diagnostic efficacy of each spectral-CT parameter in predicting PNI in gastric adenocarcinomas were shown in Table 4; Fig. 4. The AUC of 2D-nICDP was 0.761, significantly higher than those of 2D-ICVP, 2D-ICDP and 2D-nICVP (AUC = 0.622, 0.670, 0.639), 3D-nICDP (AUC = 0.663). Additionally, multivariable logistic regression analysis revealed that only 2D-nICDP was an independent predictor of PNI (Table 5).

Table 4.

Comparison of the predictive performance of the spectral-CT parameter

Parameters	AUC (95% CI)	Youden index	Cut-off value	Sensitivity (%)	Specificity (%)	Z
2D-ICVP	0.622 (0.514–0.722)	0.258	1.940	69.44	56.36	2.005
2D-ICDP	0.670 (0.563–0.765)	0.352	1.670	86.11	49.09	2.962
2D-nICVP	0.639 (0.532–0.737)	0.313	0.439	69.44	61.82	2.311
2D-nICDP	0.761 (0.661–0.845)	0.460	0.532	80.56	65.45	5.139
3D-nICDP	0.663 (0.556–0.758)	0.463	0.321	66.67	65.45	2.742

VP, venous phase; DP, delayed phase; IC, iodine concentration; nIC, normalized iodine concentration; 2D, two-dimensional; 3D, three-dimensional; AUC, area under the receiver operating characteristic curve; CI, confidence interval

Fig. 4.

ROC analyses of IC values for the prediction of perineural invasion. The 2D-nICDP parameter yielded the highest area under the receiver operating characteristic curve of 0.761 (95% CI, 0.661–0.845)

Table 5.

Multivariate logistic regression analysis for predicting PNI

Variables	Odds ratio	Wald value	95% CI	P-value
2D-ICVP	0.917	0.007	0.112–7.479	0.935
2D-nICVP	0.976	0.349	0.902–1.057	0.555
2D-ICDP	0.952	0.002	0.129–7.012	0.961
2D-nICDP	1.091	4.934	1.010–1.178	0.026
3D-nICDP	0.985	0.262	0.929–1.044	0.609

VP, venous phase; DP, delayed phase; IC, iodine concentration; nIC, normalized iodine concentration; 2D, two-dimensional; 3D, three-dimensional; CI, confidence interval

Discussion

In this retrospective study, we investigated the correlations between the ICs of 2D-ROI and 3D-ROI approaches, and evaluated the predictive performances of ICs and nICs for PNI in GC. Excellent consistency and correlation, but significant differences in average values were identified between ICs obtained with 2D and 3D methods. 2D-ICVP, 2D-ICDP, 2D-nICVP, and 2D-nICDP, as well as 3D-nICDP were associated with PNI positivity, with 2D-nICDP demonstrating the highest performance with an AUC of 0.761 and serving as an independent predictor of PNI.

Studies have shown that 3D-ROI segmentation reflects more comprehensive tumor information and tumor heterogeneity [26–28]. However, whether 3D is truly more valuable than 2D remains unclear, and there is little research on the comparison between 2D-ROI ICs and 3D-ROI ICs for GC. Our findings indicated that the consistency and correlation of ICs between the two ROI approaches were excellent, demonstrating that the 2D-ROI method not only reduces post-processing time and improves feasibility for radiologists in a busy clinical setting but also ensures the reliability of the results. However, the average ICs delineated in 2D-ROI were all higher than those in 3D-ROI in the AP/VP/DP. The reasons may be that the 3D-ROI reflects the IC of the entire GC lesion, which can include more tumor features, such as small cystic changes or areas of poor blood supply at the tumor edge, etc. In contrast, the 2D-ROI selected the layer with the largest tumor solid area, which can reduce the impact of these factors. Although 3D-ROI is based on the entire gastric adenocarcinoma lesion, making it relatively more objective, the operation is laborious, time-consuming and operator-dependent, and its measurements can be highly variable depending on tumor heterogeneity and segmentation boundaries, which limits its clinical application in the real world.

Our results indicate that ICs in VP and DP can effectively predict the PNI status of GC. Previous studies [23] have shown that the results of nIC obtained by removing individual circulatory differences between patients are more reliable than IC values, in view of this, this study included nIC for analysis. The results indicated statistically significant differences in nICVP and nICDP between the PNI positive group and the PNI negative group, suggesting the introduction of nIC can mitigate the IC drift resulting from inter-patient circulation differences, thereby more reliably reflecting the IC in GC. However, there was no statistical significance in the ICs and nICs in AP, which may be related to the enhancement mode of GC. GC is a tumor with abundant interstitial fibers [29], characterized by continuous enhancement in the venous and delayed phases [30, 31]. This study found that 3D-nICDP was related to PNI, but it was excluded in the multivariable regression analysis. Only the 2D-nICDP was an independent predictor of PNI positivity in GC, indicating that 3D-nICDP cannot independently predict PNI. This is possibly because the 2D-ROI, placed on the single largest and most prominently enhanced slice, may selectively sample the most biologically aggressive and vascularly rich core of the tumor, which could be more directly associated with PNI. In contrast, the 3D-ROI incorporates the entire tumor heterogeneity, including hypovascular or necrotic areas, which might dilute the specific signal from the invasive front. Therefore, its role in identifying PNI is limited, and the application of 3D-ROI in GC still needs further exploration through large-scale studies.

This study has several limitations. First, the 2D-ROI measurement method, which is based on the largest cross-section of the tumor, may introduce sampling bias as it fails to integrate the entire tumor volume compared to the 3D-ROI method. Nevertheless, this study confirms that the two ROI methods exhibit excellent consistency and correlation in IC measurements. Furthermore, the 2D-ROI approach significantly reduces post-processing time and better aligns with the practical demands of busy clinical workflows. Second, while we discussed potential reasons, the lack of correlation between 3D-ROI ICs and PNI has not yet been sufficiently explained. As our study primarily aims to develop a time-efficient and clinically practical ROI delineation method with simplified operation, future research should focus on validating the predictive value demonstrated by the 2D-nICDP and determining whether it genuinely reflects biological reality. Such validation is essential for establishing a solid theoretical foundation for using IC as a surrogate biomarker for PNI. Additionally, the sample size was relatively small, which might affect the statistical power. In future studies, we plan to expand the sample size and incorporate more comprehensive pathological evaluations to obtain more robust evidence.

Conclusions

In summary, excellent correlations were identified between the ICs measured by 2D and 3D-ROI approaches. IC values from 2D-ROI were superior to those of 3D-ROI for the preoperative prediction of PNI, serving as an easy-to-use method in spectral CT analysis for GC. 2D-nICDP is the most useful parameter.

Abbreviations

AP

Arterial phase

AUC

Area under the receiver operating characteristic curve

CI

Confidence interval

CT

Computed tomography

DP

Delayed phase

GC

Gastric cancer

IC

Iodine concentration

ICC

Intraclass correlation coefficient

nIC

Normalized iodine concentration

PNI

Perineural invasion

ROI

Region of interest

SD

Standard deviations

VP

Venous phase

Author contributions

Conceptualization: [Tianxia Bei], [Jinrong Qu]; Methodology: [Jing Li], [Xiaoqiang Yao]; Formal analysis and investigation: [Tianxia Bei], [Jing Li], [Xiaoqiang Yao], [Xuejun Chen], [Yue Wu]; Writing - original draft preparation: All authors; Writing - review and editing: [Jingrong Qu], [Xuejun Chen]; Funding acquisition: [Jing Li]; Resources: [Jing Li], [Jinrong Qu]; Supervision: [Jinrong Qu]. All authors read and approved the final manuscript.

Funding

This study was supported by National Natural Science Foundation of China (82202146); Henan Provincial Medical Science and Technology Project (SBGJ202402030); Science and Technology Development Foundation of Henan Province (242102311173); Special Funding of Henan Health Science and Technology Innovation Talent Project (YXKC2021054); Henan Province Central Plains Talent Program (Nurturing talent Series) (RS0011); Special funding of the Henan Health Science and Technology Innovation Talent Project (YXKC2020011).

Data availability

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the institutional review board of the Affiliated Cancer Hospital of Zhengzhou University & Henan Cancer Hospital (Zhengzhou, 450008, China) in accordance with the Declaration of Helsinki. All methods were carried out in accordance with relevant guidelines and regulations. The requirement for written informed consent was waived.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.