Value of MRI and diffusion-weighted imaging in diagnosing normal-sized pelvic lymph nodes metastases in patients with cervical cancer

Objectives:

To investigate the value of conventional MRI and diffusion-weighted imaging (DWI) in diagnosing normal-sized pelvic lymph nodes metastases in patients with cervical cancer.

Methods:

102 patients with cervical cancer who underwent MRI and DWI scan were included. 137 lymph nodes were analyzed, including 44 metastatic lymph nodes (MLNs) and 93 non-metastatic lymph nodes (non-MLNs). The morphology and apparent diffusion coefficient (ADC) value of lymph nodes were measured including short-axis diameter (D_S), long-axis diameter (D_L), ratio of short-to-long-axis diameter (D_R), fatty hilum, asymmetry, ADC_max, ADC_mean and ADC_min. The Mann-Whitney U-test, independent sample t-test and Chi-square test were employed to compare the differences of all criteria between MLNs and non-MLNs. Logistic regression and decision tree were used to develop the combined diagnostic model. ROC analyses were used to evaluate the diagnostic performance.

Results:

The D_S and D_R of MLNs were significantly higher than those of non-MLNs (p < 0.05), the ADC_max, ADC_mean and ADC_min of MLNs were significantly lower than those of non-MLNs (p < 0.05). Presence of fatty hilum and asymmetric lymph nodes between MLNs and non-MLNs were significantly different (p＜0.05). Combined measurement of ADC_min, D_S and D_R had the highest AUC 0.937 with 90.9% sensitivity and 87.1% specificity. The accuracy of decision tree was 88.3%.

Conclusion:

MRI with DWI had potential in diagnosing normal-sized pelvic lymph nodes metastases in patients with cervical cancer. The combined evaluation of D_S, D_R and ADC_min of lymph nodes and decision tree of the combined measure showed better diagnostic performances than sole criteria.

Advances in knowledge:

The short-axis diameter, ratio of short-to-long-axis diameter and ADC_min of lymph nodes have moderate value in the diagnosis of the metastases of the normal-sized lymph nodes for the patient with cervical cancer as the sole indices. The combined evaluation of D_S, D_R and ADC_min is much more valuable in the detection of metastatic lymph nodes.

Introduction

Cervical cancer is the fourth most common females cancer worldwide, and the second most common in low- and middle-income countries.^1,2 International Federation of Gynecology and Obstetrics (FIGO) staging has been consistently identified as the most important staging system worldwide for patients in cervical cancer. According to the latest FIGO staging, the presence of pelvic and/or para-aortic metastatic lymph nodes (MLNs) assigns the cases to Stage IIIC, irrespective of tumor size and extent (with r and p notations). If only pelvic nodes are positive, it is Stage IIIC1; if para-aortic nodes are also involved it is Stage IIIC2.³ According to the National Comprehensive Cancer Network (NCCN) Clinical Practice Guidelines in Oncology, the patients with pelvic or para-aortic MLNs are candidates of EBRT, brachytherapy and chemotherapy instead of surgical operation.⁴ A patient without parametrial or vaginal involvement and distant metastases will be assigned to early T stage and be a candidate for surgical treatment.³ Whereas if the patient like this is found with lymph node metastasis, the treatment will need to be altered completely.³ Thus, the detection of MLNs has a significant influence on the choice of treatment for the patients who do not meet other diagnostic criteria of advanced cervical cancer except lymph nodes metastases. Furthermore lymph node status is one of the most important prognostic factors for patients with uterine cervical carcinoma,^5–7 the patients with MLNs have poorer survival compared with those who do not have MLNs,⁸ thus the accurate evaluation of MLNs is important for selection of the optimal therapeutic strategy for cervical cancer.^9,10

Traditional method for identifying MLNs at MRI is based on the short-axis diameter (D_S) of the lymph nodes, the mainly used cut-off value is 10 mm.^11,12 However, Williams verified that 54.5% of metastatic nodes were less than 10 mm among 504 lymph nodes from patients with gynecologic malignancy via histological analysis.¹³ Song J analyzed normal-sized lymph nodes in the patients with cervical cancer and found that mean D_S of 6.26 mm of MLNs was larger than that of non-MLNs.¹⁴ Klerkx studied lymph nodes metastases in cervical cancer, and the results demonstrated that specificities were improved but sensitivities were decreased when cut-off value of D_S increased from 5 to 10mm.¹⁵ So the cut-off value of 10 mm did not bring a satisfactory evaluation of nodal status, as the reported sensitivities have been varied, ranging from 24 to 73%.^12,16–19 Thus, functional imaging is needed to enhance diagnostic performance.

Diffusion-weighted imaging (DWI), as a functional imaging modality, has been increasingly demanded as an additional MRI sequence for cervical cancer nodal staging.^20,21 Some literatures reported that the apparent diffusion coefficient (ADC) values of metastatic nodes were significantly lower than those of benign nodes in the patients with cervical cancers.^7,22–26 For detecting lymph node metastasis, ADC_mean values were reported with sensitivities ranging from 76.4 to 91.3% and specificities ranging from 74.7 to 98%.^7,23–26 Exner M et al reported that use of DWI led to an increase in sensitivity and accuracy of evaluating lymph node metastasis compared with conventional MR criteria like D_S >10 mm, round shape and strong contrast enhancement.²⁷ In the opposite, other studies described no significant difference in ADC_mean values between benign and metastatic nodes.^14,28–31 ADC_mean value reflects the prevalence of tumor cells in a whole lymph node but may not be that accurate for micrometastasis. Thus, all ADC values should be applied for detecting normal-sized nodes metastases. Variety of cancers included in these studies might be another reason that ADC did not contribute to a diagnosis of MLNs.^28–31 Fibrosis and hyperplasia of lymph nodes may also restrict water diffusion.^14,30 So only applying ADC values for detecting pelvic MLNs is still controversial. Combination of morphological criteria and ADC values may facilitate the detection of MLNs. Klerkx and Lin G.et al studied combination of size and ADC values of lymph nodes, whereas their studies included some nodes larger than 10 mm.^15,31 Moreover the study by Lin G included patients with advanced cervical cancer.³¹ To our knowledge, only the study by Song J detected normal-sized MLNs of cervical cancer, and the results showed that combination of size and ADC value of tumor was the best criterion but the sensitivity was low.¹⁴

The purpose of this research was to investigate the value of preoperative conventional MRI combined with DWI scan in diagnosing normal-sized pelvic lymph nodes metastases in patients who were subject to be diagnosed as early stage cervical cancer.

Methods and materials

Patients

The retrospective study was approved by the Institutional Review Board of our institution, and the need for individual consent was waived by the committee. The patients with cervical cancer who underwent surgical operations between December 2014 to July 2018 were included and all of them met the entry criteria. The inclusion criteria were as follows: (1) each patient had never received any treatment for cervical cancer prior to MRI examination; (2) each patient underwent radical hysterectomy and pelvic lymphadenectomy. The time interval between preoperative MRI and surgical operation was less than 2 weeks; (3) the location and number of MLNs from excised lymph nodes were depicted on postoperative histological report of each patient. The exclusion criteria were as follows: (1) patients with other concurrent malignancies; (2) no lymph node was measurable on MRI images; (3) the visible lymph nodes were greater than 10 mm in D_S; (4) the primary tumor was confirmed pathologically as the histological types except squamous cell carcinoma, adenocarcinoma and adeno-squamous carcinoma. In total, 102 patients (age range, 27–70 years) with uterine cervical cancer were enrolled in this study. The characteristics of patients are summarized in Table 1.

Table 1.

The characteristics of patients

Characteristic Number Percentage (%)
Total number of patients	102
Age of patients
Range	27–70
Mean age ± SD	48.5 ± 9.6
Initial FIGOa stage
ⅠB1	67	65.7
ⅠB2	6	5.9
ⅡA1	25	24.5
ⅡA2	4	3.9
Histology
Squamous cell carcinoma	91	89.2
Adenocarcinoma	6	5.9
Adenosquamous carcinoma	5	4.9

SD, Standard deviation.

a

FIGO = The International Federation of Gynecology and Obstetrics (2009).

MRI protocol

All MR examinations were performed using a 3.0 T unit (Magnetom Trio; Siemens Medical Solutions, Germany) with an 8-channel-phased array body coil and respiratory gating technology. Before the examination, subjects were recommended to drink water to fill the bladder at a moderate degree and rest 15 to 30 min. Patients were in the supine position. The MR scan ranged from the superior edge of iliac crest to the lower edge of the pubic symphysis, Table 2 shows the parameters of MRI scanning.

Table 2.

The parameters of MRI scanning

plane	TR(ms)/ TE(ms)	Slice Thickness	Gap (mm)	Fov (mm)	Number of Excitations	b-values (s mm−2)
Axial TSE T1W	514/11	5 mm	2	512 × 640	2
Axial FS TSE T2W	3000/106	5 mm	2	294 × 448	2
Sagittal TSE T2W	3800/116	4 mm	0.8	396 × 448	2
Coronal FS TSE T1W	590/11	5 mm	1.25	544 × 640	2
DWI(SS-EPI)	3500/93	4 mm	0	256 × 320	2	0, 1000

Fov, Field of view; FS, Fat saturation; T1W, T1-weighted; T2W, T2-weighted; TE, Echo time; TR, Time of repetition; TSE, Turbo spin-echo.

Image analysis

The pelvic lymph nodes were grouped into six regions: bilateral obturator chains, external iliac and internal iliac on conventional axial T₂W and DWI sequences by an experienced radiologist as previously described.⁷ The MRI images were analyzed by one radiologist (Y.Y.Y. with 5 years experience in pelvic MR imaging) and one fellow (S.Q.L. with 2 years experience in pelvic MR imaging), respectively, according to previous issued papers,^29,32 both of them were unaware of the histological findings. The first radiologist chose all visible lymph nodes in each region on axial T₂W images, sagittal T₂W images and coronal T₁W images, then recorded location and all morphological indices. The D_S and long-axis diameter (D_L) of selected nodes were measured in the three planes and the mean values of D_S were calculated. The final D_L of a node was the largest diameter selected in the three planes. The ratio of short to long-axis diameter (D_R = D_S/D_L) of the nodes were calculated with the final D_L and the mean D_S (Figure 1). Moreover, asymmetric lymph nodes (included number or size) and presence of fatty hilum (classified as present and absent) and were analyzed (Figure 2).³³ Accordingly, the hyperintense nodes in each region on DW images (b = 1000 s mm⁻²）were identified and correlated with T₂W images, then the ADC values of these nodes were measured. Measurement of the ADC values of lymph node: ADC maps that were exported by the DICOM protocol were imported into a prototype software programme (Omni Kinetics, GE Healthcare). The regions of interest (ROI) were manually drawn on all axial sections by referring to T₂W and DWI (b = 0 s mm⁻²). ROI was carefully placed around the center of a node on each section, encompassed as much area of the node as possible and avoided including the margin and adjacent vessels. The ADC_max value, ADC_mean value and ADC_min value were calculated by the software. The fellow (S.Q.L.) performed the measurement again according to the above measuring method to obtain all the imaging data. The exact same lymph nodes selected by two readers, respectively, were included in the study. The corresponding results of two readers were used for the evaluation of interobserver agreement.

Figure 1.

(a-c) Images of a 43-year-old cervical cancer patient, a left external iliac lymph node (arrow) is observed on MRI images. (a) Transverse T2-weighted image of the node (arrow). The D_S and D_L of the node are 8.73mm and 11.20mm. (b) Coronal T1-weighted image of the corresponding node (arrow). The D_S and D_Lof the node are 8.96mm and 11.30mm. (c) Sagittal T2-weighted image of the same node (arrow). The D_S and D_L of the node are 9.12mm and 12.20mm. The mean D_S of the node is 8.94mm, final D_L is 12.20mm, the D_R is 0.73. D_S, short-axis diameter; D_L, long-axis diameter；D_R, ratio of short to long-axis diameter..

Figure 2.

(a) Transverse T2-weighted image of right obturator lymph node (arrow) shows absent fatty hilum，the D_S is 9.07mm. This node was histologically positive. (b) Transverse T2-weighted image of right external iliac lymph node (arrow) shows the presence of fatty hilum. This node was histologically negative. D_S, short-axis diameter.

Histopathological evaluation

The patients underwent radical hysterectomy and pelvic lymphadenectomy after MRI examination. The excised lymph nodes were labeled as the follow groups: bilateral obturator chains, external iliac and internal iliac. For a region, the resected nodes were cut into slices of thickness 2–3 mm. The resected lymph nodes from the same region were observed on a section to compare the size (D_S and D_L) and shape of these nodes (Figure 3). The pathologist documented the characteristic of all lymph nodes, and the lymph nodes of the non-metastatic region were selected in the non-metastatic group. The characteristic of the lymph nodes in the metastatic region was discussed by the pathologist and two imaging readers according to the size and shape of the nodes on the MR images and pathological slices, and the lymph nodes that were undoubtedly matched between the MR images and histopathological findings were eventually included in the metastatic group.

Figure 3.

(a,b) Images of a 28-year-old cervical cancer patient, an external iliac lymph node (arrow) was observed on MRI images, the D_S and D_L are 8.86mm and 9.74mm.(c) A pathologic section with resected lymph nodes from the same region, the D_S and D_L of the round node(arrow) are 9mm and 10mm,the size of this node is close to the node observed on MRI images. (d) Histopathological examination (hematoxylin and eosin, magnification, ×200) shows the node is metastatic positive. D_S, short-axis diameter; D_L, long-axis diameter.

Statistical analysis

Statistical analysis was performed by using SPSS v.17 (Statistical Package for Social Sciences software package for Windows, SPSS Chicago, IL, USA). First, Kolmogorov⁃Smirnov test was used to test whether all criteria followed a normal distribution. Intraclass correlation coefficient (ICC) was calculated to assess the interobserver reliability of the size-based and ADC-based criteria measurements. The evaluation criterion of reliability is as follows: excellent, 1 ≥ ICC ≥ 0.75; fair to good, 0.75 > ICC ≥ 0.40; and poor, ICC < 0.40.³⁴ Mann-Whitney U-test was employed to analyze the differences in D_S and D_L between MLNs and non-MLNs. Independent samples t-test was performed to compare the differences in D_R, ADC_max, ADC_mean and ADC_min values between the MLNs and non-MLNs. Chi-square test was employed for categorical variables. Probability was calculated regarding the combined measure of size-based and ADC-based criteria of lymph nodes using binary logistic regression. The area under the curve (AUC), cut-off value, sensitivity, and specificity for distinguishing MLNs from non-MLNs were determined by receiver operating characteristic (ROC) curve analysis. P-values less than 0.05 were considered as statistically significance. To integrate the logistic regression results in clinical practice a decision tree was proposed by referring to previous paper.³¹ The number of true-positive (TP), false-positive (FP), false-negative (FN), true-negative (TN) lymph nodes and accuracy were calculated.

Results

Clinicopathological result

Of the 102 patients, 33 (32.3%) patients had pelvic lymph node metastasis. 137 lymph nodes (D_S range from 3.30 to 9.75 mm) were obtained through the above lymph nodes selecting processing, including 44 MLNs(Figure 4) and 93 non-MLNs. For patients with MLNs, 56 nodes were initially identified on MRI then 12 nodes were excluded after referring to histological results. The concordance rate of MLNs between MRI findings and histologic results was 78.6%.The prevalence of pelvic lymph node metastasis was as follows: right obturator (n = 10); left obturator (n = 18); right external iliac (n = 9); left external iliac (n = 5); right internal iliac (n = 1); left internal iliac (n = 1).

Figure 4.

(a,b) Images of a 51-year-old cervical cancer patient diagnosed with left obturator and internal iliac lymph nodes metastases. (a) The nodes are slight hyperintensity signal at axial fat suppressed T2-weighted TSE sequence; the D_S of left obturator node（yellow arrow） is 6.31mm and the D_S of internal iliac node（green arrow） is 9.24mm. (b)The lymph nodes show hyperintensity on DWI; the maximum, mean and minimum ADC values of the left obturator lymph node（yellow arrow） are 0.893×10^-3mm²/s,0.642×10^-3mm²/s, 0.545×10^-3mm²/s; and the ADC values of the internal iliac node（green arrow） are 0.947×10^-3mm²/s,0.773×10^-3mm²/s,0.608×10^-3mm²/s, respectively. (c,d) Images of a 28-year-old cervical cancer patient diagnosed with left obturator lymph node metastasis. (c) The node（yellow arrow） is isointensity signal at axial fat suppressed T2-weighted TSE sequence; the D_S is 6.23mm. (d) The same lymph node（yellow arrow） shows hyperintensity on DWI; the maximum, mean and minimum ADC values of the lymph node are 1.135×10^-3mm²/s,0.957×10^-3mm²/s,0.773×10^-3mm²/s,respectively. TSE, turbo spin-echo; D_S,short-axis diameter; ADC, apparent diffusion coefficient.

Size-based criteria result

The ICC (95% CI) of inter observer reproducibility for measuring the D_S, D_L, D_R, ADC_max, ADC_mean and ADC_min values were listed in Table 3. The mean D_S (± standard deviation) was 7.03 ± 1.32 mm for MLNs and 5.27 ± 1.20 mm for non-MLNs. The D_S of MLNs was significantly greater than that of non-MLNs (p < 0.001). The differences in D_L between MLNs and non-MLNs was not significant (p > 0.05). The D_R of MLNs and non-MLNs were 0.74 ± 0.11 and 0.61 ± 0.11, respectively. The D_R of MLNs was significantly greater than that of non-MLNs (p < 0.001). (Figure 5). The differences in presence of fatty hilum and asymmetric lymph nodes between MLNs and non-MLNs were significant (p < 0.05) (Table 4).

Table 3.

The ICC (95% CI) of inter observer reproducibility from two radiologists

	Intraclass correlation coefficient	95% confidence interval
Short-axis diameter	0.889	0.829–0.932
Long-axis diameter	0.861	0.782–0.913
Short- to long-axis diameter	0.887	0.821–0.929
ADCmax value	0.921	0.873–0.952
ADCmean value	0.910	0.852–0.941
ADCmin value	0.942	0.908–0.961

95%CI, 95% confidence intervals; ADC, Apparent diffusion coefficient; ICC, Intraclass correlation coefficient.

Figure 5.

(a)The box plot compares the short-axis diameter between MLNs and non-MLNs. (b) The box plot compares the ratio of short to long-axis diameter between MLNs and non-MLNs. (c) The box plots compare the values of ADC_max, ADC_mean and ADC_minbetween MLNs and non-MLNs. MLN, metastatic lymph node; ADC, apparent diffusion coefficient.

Table 4.

The presence of fatty hilum and asymmetric lymph nodes observed on MRI

	Metastatic	Non-metastatic	p
Presence of fatty hilum			0.022
Present	14(44)	49(93)
Absent	30(44)	44(93)
Asymmetric lymph nodes			0.031
Symmetric	15(44)	50(93)
Asymmetric	29(44)	43(93)

ADC-based criteria result

The difference of ADC-based criteria between MLNs and non-MLNs: the ADC_max value, ADC_mean value, ADC_min value of MLNs were (1.119 ± 0.145) × 10⁻³ mm² s⁻¹, (0.917 ± 0.155) × 10⁻³ mm² s⁻¹ and (0.753 ± 0.161) × 10⁻³ mm² s⁻¹ respectively; and those of non-MLNs were (1.315 ± 0.146) × 10⁻³ mm² s⁻¹, (1.132 ± 0.141) × 10⁻³ mm² s⁻¹ and (0.967 ± 0.127) × 10⁻³ mm² s^−1, respectively. The ADC_max value, ADC_mean value and ADC_min value of MLNs were significantly lower than those of non-MLNs (p＜0.001) (Figure 5).

Combined measure result

D_S, D_R and ADC_min of lymph node were independent factors for detecting MLNs by using Logistic regression analysis. The prediction model was built as Logit (P1) = -4.66 + (−0.01 × ADC_min) + (0.62 × D_S) + (11.27 × D_R). The formula of the probability was as follow: P1 = EXP [-4.66–0.01 × ADC_min+0.62×D_S+11.27×D_R)/(1 + EXP (-4.66–0.01 × ADC_min + 0.62 × D_S + 11.27 × D_R)].

ROC curve analysis result

Table 5 summarized the AUC and the optimal cut-off of morphological criteria (D_S, D_R, fatty hilum and asymmetry), ADC-based criteria (ADC_max value, ADC_mean value, ADC_min value) of lymph nodes and combined measure (from D_S, D_R and ADC_min) for distinguishing MLNs from non-MLNs. Using 6.12 mm as the cut-off value of short-axis diameter, the sensitivity and specificity for selection of MLNs from non-MLNs were 75.0 and 78.4%, respectively. The sensitivity and specificity were 71.0 and 88.6% by using 0.910 × 10⁻³ mm² s⁻¹ as ADC_min cut-off value. The sensitivity and specificity were 90.9 and 87.1% of combined measure P1 in Logistic regression analysis for detecting MLNs. Figure 6 described ROC of morphological criteria, ADCs values of lymph nodes and combined measure P1 for detecting MLNs.

Table 5.

The diagnostic performances of morphological criteria and ADC-based criteria for detecting metastatic lymph nodes

	AUC	Cut-off value  a	Sensitivity	Specificity
Short-axis diameter	0.837	6.12	75.0%	78.4%
Short to long-axis diameter	0.810	0.69	77.3%	75.3%
Presence of fatty hilum	0.588		65.9%	51.6%
Asymmetric lymph nodes	0.565		61.4%	51.6%
ADCmax value	0.822	1.215	73.1%	77.2%
ADCmean value	0.840	1.066	72.0%	81.8%
ADCmin value	0.845	0.910	71.0%	88.6%
Combined measureb	0.937	0.322	90.9%	87.1%

ADC, Apparent diffusion coefficient.

a

Short-axis diameter of lymph nodes is in unit of mm, ADC_max value, ADC_mean value, ADC_min value of lymph nodes are in units of 10⁻³ mm² s⁻¹.

b

Probability (0.322) calculated from D_S, D_R and ADC_min of lymph node was used for diagnosis.

Figure 6.

The receiver operating characteristic (ROC) curve for size-based criteria and ADC values of lymph nodes and combined measure P1 to differentiate MLNs from non-MLNs.The AUC of combined measure P1 is larger thanother indices indicate combined measure P1 is superior to other indexes. MLN,metastatic lymph node; ADC, apparent diffusion coefficient, AUC, area under the curve.

Decision tree result

The cut-off values of indices incorporated in the combined measure were employed to propose the decision tree(Figure 7). The TP, TN, FP and FN were 38, 83, 10 and 6, respectively. The accuracy was 88.3%.

Figure 7.

The decision tree according to logistic equation in cervical cancer patients. First, no visible node on T2W or DWI was score 0. Second, ADC_min was employed to evaluate if the patients have MLNs, ADC_min value of a node less than 0.910 ×10^-3mm²/s was score 2, a node that ADC_minvalue was more than 0.910 ×10^-3mm²/s was score 1. Then D_Sand D_R were employed, the node that D_S ≥6.12mm or D_R≥0.69 was score 3, otherwise score 2. Finally, if the node that D_S ≥6.12mm and D_R ≥0.69 was score 4. No lymph node metastasis when score was 0,1 or 2, and a node was considered as malignant when score was 3 or 4. ADC,apparent diffusion coefficient. D_S, short-axis diameter. D_R,ratio of short to long-axis diameter.

Discussion

The accurate detection of MLNs is especially important for patients who were subject to be diagnosed as early stage cervical cancer according to FIGO 2009. Because the status of lymph nodes will result in two entirely different treatments of these patients. Some studies reported that surgical lymph node assessment was irreplaceable in diagnosing MLNs, but the risk of immediate and delayed complications would increase.^22,26 For these reasons, a non-invasive and accurate method of detecting lymph node metastasis like MRI is desirable. More care should be taken to evaluate the lymph nodes less than 1cm given that normal-sized MLNs were subject to be omitted. A lymph node with the D_S less than 10 mm was identified traditionally as benign frequently using the size of lymph node as the sole criterion for estimating LN metastasis. However, Benedetti’s study demonstrated that the D_S of more than 80% metastatic nodes was less than 10 mm in 225 cervical cancer patients who underwent radical hysterectomy and pelvic lymphadenectomy.³⁵ In addition, Song J et al also reported approximately 42.4% nodes were metastases in MRI recognizable lymph nodes with the D_S between 5 and 10 mm.¹⁴ The likelihood of missing diagnose would increase if using the criterion 10 mm for the assessment of lymph nodes status. Thus, the study focused on the patients who were diagnosed as early stage cervical cancer with normal-sized pelvic lymph nodes. Not only tumor cells but also some benign abnormalities can also limit water diffusion, the separate analysis of ADC values is not the best way to detect MLNs. In contrast to previous papers this study used combination of morphologic and ADC criteria for detecting MLNs and achieved a good performance. To apply the results in the clinical practice a decision tree was proposed in this study and obtained a good accuracy.

The thickness of axial image in the study was 5mm. Some previous studies measured lymph nodes only in axial planes with the similar thickness and the results resembled those of this study.^7,26 However, the nodes larger than 10mm were also included in those studies.^7,26 Song J measured normal-sized lymph nodes by using 3.5mm axial T2W sequence and reported the D_S of MLNs was larger than thatof non-MLNs, but the performance of D_S was not evaluated.¹⁴ A 3.0-TMR unit provided a high spatial resolution in this study but the thickness of conventional sequences was not the optimal for normal-sized lymph nodes. To measurethe size of lymph nodes more accurately, the nodes were observed in axial, sagittal and coronal planes in this study.

Using 6.12 mm as the cut-off value of D_S for distinguishing MLNs from non-MLNs with normal size, the sensitivity and specificity were 75.0 and 78.4%, respectively. Similarly, Liu Y et al reported that the sensitivity was 52% when 10 mm was used as a cut-off value, but it raised to 76% using D_S of 7.75 mm as the cut-off value; however, the specificity was decreased compared with cut-off value 10 mm.²⁶ The sensitivity of this study was close to that of Liu Y’s study, which meant that nearly 25% MLNs were false negative according to 6.12 mm of D_S.²⁶ The specificity of this study reached a relatively high level but sensitivity was low, thus D_S was difficult to be the sole diagnostic index to satisfy clinical expectations. The cut-off value of D_R was 0.69 for differentiating MLNs from non-MLNs, which meant lymph nodes were more likely metastases if their forms tend to be round. The results of D_S and D_R obtained from the present study suggested the morphological criteria are valuable in some cases, but not conclusive indices for identifying MLNs. The differences in presence of fatty hilum and asymmetry between MLNs and non-MLNs were significant but diagnostic performances were pretty low.

Since the lymph nodes included in this study were subcentimeter, the thickness 4mm of DWI was not the best for the measurement of ADC values. For normal-sized lymph nodes, some study reported that the differences of ADC values between MLNs and non-MLNs were not significant.^14,29,30 Maybe there was a probability of partial volume effects when measuring ADC values of subcentimeter nodes. Thus, ROIs were manually drawn on all sections of a node to decrease the influence of other tissues. Some previous study measured ADC values on all slices of subcentimeter nodes, then obtained the similar results to this study.^36,37

The ADC_max, ADC_mean, and ADC_min values of MLNs were all significantly lower than those of non-MLNs. The results were in accordance with results of several former correlational researches, which reported that DWI had higher diagnostic accuracy than conventional MRI for discriminating MLNs from non-MLNs.^26,27,38 Restriction in the diffusion of water molecules is directly proportional to the degree of cellularity of the tissue. Compared to benign tissue cells, the malignant tumor cells have the characteristics of rapid growth patterns with an increase in amount and density, which result in disordered arrangement of the intracellular structure and the decreased extracellular spaces. The intracellular spaces reduce due to higher nucleus/cytoplasm ratio of malignant tumor cells.^7,39 Ultimately, ADC values of MLNs decreased as a result of the restrictions of both intracellular and extracellular water molecules movement which were caused by the above histopathological changes. The results of ADC-based criteria in the study were in keeping with the above researches. The ADC_mean had the sensitivity 72.0% and the specificity 81.8% for distinguishing MLNs from non-MLNs, the sensitivity and specificity of ADC_min was 71.0 and 88.6%, respectively. Compare with D_S and D_R, the sensitivities of ADC_mean and ADC_min values were similar to D_S and the specificity of ADC_min was the highest in all sole criteria. The diagnostic performances of ADC-based criteria were better in comparison with morphologic-based criteria, and especially ADC_min value was the best sole discriminator, which was similar to some early literatures.^7,26 Choi EK.et al and Liu Y. et al also obtained results showing that ADC_min was the best sole index with the highest AUC for evaluating the status of lymph nodes.^22,26 These similar results might be able to be interpreted that some MLNs were partially replaced by tumor cells rather than entirely occupied by tumor cells, leading to heterogeneous ADC values areas. The character of a lymph node is contingent on tumor cells infiltration regardless the number of tumor cells and the extent of malignant spread. Thus, it is critical to detect MLN which had only focal infiltration with malignant tumor cells. The ADC_min can be an important criterion to facilitate the selection of the MLNs timely. Consistent with the above viewpoints, the ROC curve demonstrated that ADC_min had the highest AUC and specificity in differentiating MLNs from non-MLNs than other sole criteria.

The results of this study confirmed the previous findings about morphological and ADC criteria. We further investigated combined morphological criteria and ADC values of normal-sized lymph nodes as a new measurement. Logistic regression analysis showed that D_S, D_R and ADC_min of lymph node were independent factors for differentiating MLNs from non-MLNs, and the prediction model combing D_S, D_R with the ADC_min of lymph nodes obtained the highest AUC of 0.937 for the diagnosis of MLNs. Because the combined prediction model included both morphological and functional factors, the combination of the D_S, D_R and ADC_min could increase the sensitivity to 90.9%, and the specificity maintained 87.1% close to the specificity of 88.6% obtained by using ADC_min alone. Kang S.et al also investigated combined evaluation of tumor size with lymph node size in the detection of lymph node metastasis obtained the highest AUC and sensitivity 83.3% and the specificity 55.8%, but the specificity of combined measure had significant decrease comparing with the sole criteria in the study.⁴⁰ The specificity of combined measure in this study did not decrease significantly, presumably because the weight of ADC_min was high according to logistic analysis. Another study from Lin G.et al discussed the diagnostic performance of the combined measure, which was composed of the size of lymph nodes and relative ADC values between tumors and nodes, and the results revealed that combined measure outperformed the sole criteria with the sensitivity 83.3% and specificity 99.0% on a region basis.³¹ The specificity of combined measure maintained a high level in the study from Lin G.et al, which was in accord with the finding from this study.³¹ The decision tree we proposed simplified the logistic regression results as a flowchart and demonstrated a good accuracy 88.3%. Combination of ADC-based and morphological criteria in this study exhibited a better performance for discriminating MLNs from non-MLNs than all sole criteria.

There were several limitations in our study. First, our study was a retrospective study with a relatively small patient population in a single center, and the cases were in relative early T stage. Therefore, more researches with larger group of cases should be performed to verify the results of the present study. The second limitation is matching the lymph nodes observed on MRI with histological results. To overcome this limitation, the lymph nodes identified on MRI were correlated with histological results by radiologists and pathologists in consensus. Only nodes which characteristics observed on MRI were the same as those on pathological sections were included. Third, the slice 5mm thickness of conventional MRI and 4mm of DWI are suboptimal for measurements of lymph nodes diameters below 1 cm. Thus,the results of this study need further validation. A follow up study with 3D T2W sequences and thinner sliced DWI should be performed. To improve the accuracy of the measurements of diameters of lymph nodes the D_S and D_L were measured in the three planes. The ADC values of lymph nodes were measured on all sections to minimize the influence of other tissues. Finally, although the presence of para-aortic lymph node metastasis is an important prognostic factor, this was not analyzed in our study because of the limitation of scanning range. A further study including the presence of para-aortic lymph node metastasis will enhance our results.

Conclusion

The D_S, D_R and ADC values of lymph nodes were moderately valuable for detection of normal-sized MLNs as sole indices. Combined the D_S, D_R with ADC_min of lymph nodes showed the best diagnostic performance among all criteria. The decision tree of combined measure had a good accuracy of the detection of lymph nodes metastases. MRI with DWI had potential in diagnosing normal-sized pelvic lymph nodes metastases in patients with cervical cancer.

Funding

Climbing fund of the National Cancer Center of China NCC201806B011 Support program of the Youth Science and Technology Innovation Talents of Shenyang City RC180269. The funding sources had no such involvement (in study design; in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication) in the study.

Contributors

Qingling Song: Conceptualization, Methodology, Formal Analysis, Investigation, Writing – Original Draft. Yanyan Yu: Formal Analysis, Investigation, Data Curation. Xiaomiao Zhang: Formal Analysis, Investigation. Yanmei Zhu: Pathology analysis, Resources. Yahong Luo: Writing – Review & Editing. Tao Yu: Writing – Review & Editing. Jiayi Sun: Writing – Editing. Fan Liu: Investigation, Data Curation. Yue Dong: Conceptualization, Methodology, Writing – Review & Editing, Supervision, Project Administration Funding Acquisition