T1 mapping and multimodel diffusion-weighted imaging in the assessment of cervical cancer: a preliminary study

Abstract

Objective:

To evaluate the clinical feasibility of T1 mapping and multimodel diffusion-weighted imaging (DWI) for assessing the histological type, grade, and lymphovascular space invasion (LVSI) of cervical cancer.

Methods:

Eighty patients with cervical cancer and 43 patients with a normal cervix underwent T1 mapping and DWI with 11 b-values (0–2000 s/mm²). Monoexponential, biexponential, and kurtosis models were fitted to calculate the apparent diffusion coefficient (ADC), pure molecular diffusion (D), pseudo-diffusion (D*), perfusion fraction (f), mean diffusivity (MD), and mean kurtosis (MK). Native T1 and DWI-derived parameters (ADC_mean, ADC_min, D_mean, D_min, D*, f, MD_mean, MD_min, MK_mean, and MK_max) were compared based on histological type, grade, and LVSI status.

Results:

Native T1 and DWI-derived parameters differed significantly between cervical cancer and normal cervix (all p < 0.05), except D* (p = 0.637). Native T1 and MK_mean varied significantly between squamous cell carcinoma (SCC) and adenocarcinoma (both p < 0.05). ADC_min, D_min, and MD_min were significantly lower while MK_max was significantly higher in the high-grade SCC group than in the low-grade SCC group (all p < 0.05). LVSI-positive SCC had a significantly higher MK_mean than LVSI-negative SCC (p < 0.05).

Conclusion:

Both T1 mapping and multimodel DWI can effectively differentiate cervical cancer from a normal cervix and cervical adenocarcinoma from SCC. Furthermore, multimodel DWI may provide quantitative metrics for non-invasively predicting histological grade and LVSI status in SCC patients.

Advances in knowledge:

Combined use of T1 mapping and multimodel DWI may provide more comprehensive information for non-invasive pre-operative evaluation of cervical cancer.

Introduction

Cervical cancer is the fourth leading cause of cancer-related death among females worldwide. ¹ The common prognostic factors of cervical cancer include tumour stage, histological subtype, grade, and status of lymphovascular space invasion (LVSI). ^2,3 MRI is widely used for the diagnosis and staging of cervical cancer, and the 2018 International Federation of Gynecology and Obstetrics (FIGO) staging system has allowed for incorporation of imaging findings into disease staging. ^4–6 However, conventional morphologic MRI lacks specific parameters reflecting tumour microstructure, making it difficult to assess tumour type, grade, and LVSI.

Diffusion-weighted imaging (DWI) is a well-known functional MRI technique that reflects the random diffusion of water molecules in vivo and has been explored for classifying cervical cancer. ^7,8 Recently, advanced diffusion models, such as intravoxel incoherent motion (IVIM) and diffusion-kurtosis imaging (DKI), have been used to assess cervical cancer. ^9,10 T1 mapping is another non-invasive MRI technique for quantitative analysis of the intrinsic properties of biological tissues; in this technique, the longitudinal relaxation time (T1 value) in each image voxel is measured. T1 mapping has been thoroughly explored in cardiac imaging ^11,12 and has been gradually applied to renal tumour, breast cancer, and lung cancer as a quantitative and contrast-free technique. ^13–16 However, its clinical value in cervical cancer remains unclear. Therefore, the present study aimed to comprehensively investigate the potential of T1 mapping and multimodel DWI in the characterisation of cervical cancer, specifically inidentifying the pathological type, grade, and LVSI status.

Methods and materials

Study population

This prospective single-centre study was approved by the Institutional Review Board of the First Affiliated Hospital of Zhengzhou University, and all patients provided written informed consent prior to examination. Data were collected for 93 consecutive patients with cervical cancer who underwent pelvic MRI examination including T1 mapping and DWI at the First Affiliated Hospital of Zhengzhou University between January 2020 and March 2022. The inclusion criteria were as follows: (i) no pelvic metal artifacts or other contraindications to MRI examination; (ii) no surgery, chemotherapy, or radiotherapy before MRI examination; and (iii) confirmed cervical cancer on the basis of biopsy or surgical pathology after MRI examination. The exclusion criteria were as follows: (i) image quality not meeting the requirements for analysis; (ii) tumour visibility on only one MRI slice; and (iii) rare tumours, such as adenosquamous carcinoma and neuroendocrine carcinoma. A total of 80 patients with cervical cancer were included in this study: 50 cases were confirmed by surgery, and 30 cases were confirmed by biopsy. A pathologist with nine years of experience in the diagnosis of uterine malignancies re-evaluated the slides for each patient without knowing the original diagnosis. Patients were classified into a squamous cervical cancer (SCC) group and an adenocarcinoma group on the basis of histopathological findings. Patients in the SCC group were further classified into a high-grade group (poorly differentiated tumour) and a low-grade group (well/moderately differentiated tumour), as well as an LVSI-positive group and an LVSI-negative group. Owing to the small number of cases in the adenocarcinoma group, their histological grade and LVSI status were not further classified. All cases were staged using the revised 2018 FIGO staging guidelines for cervical carcinoma. ⁵

Data from patients with uterine benign diseases were collected to establish a control group. There were 22, 12, and 9 cases of uterine myoma, endometrial polyps, and endometrial hyperplasia, respectively. No patient in the control group had corresponding clinical symptoms or signs of cervical disease, and cervical pap smears and MRI indicated that the cervix was normal in all cases.

Image acquisition

Magnetic resonance (MR) examinations were performed using a 3T MR scanner (Ingenia CX, Philips Healthcare, Best, the Netherlands) with a 32-channel phased array coil. All patients fasted for at least 6 h before the examination and were scanned in the head-first supine position. Axial T ₂ weighted (T ₂W) turbo spin-echo, DWI, and T1-mapping sequences were performed. Native T1 maps were acquired using the modified Look–Locker inversion recovery (MOLLI) sequence, which was automatically reconstructed by the built-in algorithm on Philips MR console. DWI was obtained in a single-shot echoplanar imaging sequence with diffusion gradient b factors of 0, 10, 20, 50, 100, 200, 400, 800, 1200, 1600, and 2000 s/mm². Diffusion gradients were applied simultaneously along with three orthogonal directions. Table 1 provides a detailed description of the MRI parameters.

Table 1.

MRI sequence parameters

Parameters	T 2WI	DWI	T1 mapping
Imaging technique/orientation	TSE/Axial	EPI/Axial	MOLLI/Axial
TR (ms)	4093	5614	1.02
TE (ms)	100	103	2.2
Field of view (mm2)	200 × 200	240 × 240	380 × 304
Slice thickness (mm)	5	5	5
No. of slices	24	24	12
Acquisition matrix	344 × 306	72 × 67	124 × 123
Voxel size (mm3)	0.58 × 0.65 × 5.00	3.33 × 3.58 × 5.0	2.42 × 2.44 ×5.0
B-values (s/mm2)	NA	0, 10, 20, 50, 100, 200, 400, 800, 1200, 1600, 2000	NA
Flip angle (degrees)	90	90	20
Bandwidth (Hz/pixel)	250.6	1784.8	1083.8
Acquisition time	4 min 30 s	5 min 37 s	2 min 12 s

DWI, diffusion-weighted imaging; EPI, echoplanar imaging; MOLLI, Look–Locker inversion recovery; NA, not applicable; TE, echo time; TR, repetition time; TSE, turbo spin-echo; T ₂WI, T ₂ weighted imaging.

Image analysis and measurements

All data were analysed using a commercially available post-processing workstation (IntelliSpace Portal V10, Philips Healthcare). The DWI data were processed by the application of advanced diffusion analysis on the workstation. The following DWI parameters were derived: apparent diffusion coefficient (ADC) from a monoexponential model (Eq. 1), diffusion coefficient (D), pseudo-diffusion coefficient (D*), perfusion fraction (f) from IVIM (Eq. 2), mean diffusivity (MD), and mean kurtosis (MK) from DKI (Eq. 3).

ADC was calculated by fitting a monoexponential model with two b-values (0 and 800 s/mm²):

, $$\frac{{\mathrm{S}}_{\mathrm{b}}}{{\mathrm{S}}_0}=\mathrm{e}\mathrm{x}\mathrm{p}\left(-\mathrm{b}\bullet \mathrm{D}\right)$$ (1)

where ${\mathrm{S}}_{\mathrm{b}}$ is the signal intensity of diffusion weighting b, and ${\mathrm{S}}_0$ refers to the signal intensity without the diffusion gradient applied.

IVIM was evaluated using the following equation with nine b-values (0, 10, 20, 50, 100, 200, 400, 800, and 1200 s/mm²):

, $$\frac{{\mathrm{S}}_{\mathrm{b}}}{{\mathrm{S}}_0}=\left(1-\mathrm{f}\right)\text{∙exp}\left(-\mathrm{b}\bullet \mathrm{D}\right)+\mathrm{f}\mathrm{e}\mathrm{x}\mathrm{p}\bullet \left(-\mathrm{b}\bullet \left({\mathrm{D}}^{\mathrm{*}}+\mathrm{D}\right)\right)$$ (2)

where D is the diffusion coefficient associated with water diffusion in the extravascular space, D* is the pseudo-diffusion coefficient related to perfusion in the vascular space, and f is the perfusion fraction.

DKI was evaluated using the following equation with eight b-values (0, 100, 200, 400, 800, 1200, 1600, and 2000 s/mm²):

, $$\frac{{\mathrm{S}}_{\mathrm{b}}}{{\mathrm{S}}_0}=\mathrm{e}\mathrm{x}\mathrm{p}\left(-\mathrm{b}\bullet {\mathrm{D}}_{\mathrm{a}\mathrm{p}\mathrm{p}}+\frac{1}{6}{\mathrm{b}}^2{\bullet \mathrm{D}}_{\mathrm{a}\mathrm{p}\mathrm{p}}{\bullet \mathrm{K}}_{\mathrm{a}\mathrm{p}\mathrm{p}}\right)$$ (3)

where ${\mathrm{D}}_{\mathrm{a}\mathrm{p}\mathrm{p}}$ is the corrected ADC derived from the non-Gaussian model, and ${\mathrm{K}}_{\mathrm{a}\mathrm{p}\mathrm{p}}$ is a unitless parameter of the apparent kurtosis coefficient. MD and MK are the averages of ${\mathrm{D}}_{\mathrm{a}\mathrm{p}\mathrm{p}}$ and ${\mathrm{K}}_{\mathrm{a}\mathrm{p}\mathrm{p}}$ among three distributed directions, respectively.

All image data were analysed independently by two radiologists with more than 5 years of experience in uterine MR diagnosis without knowledge of the clinical/pathological information. The average of the measurements by the two radiologists was used for subsequent statistical analysis. The polygonal regions of interest (ROIs) were drawn along the inside of the tumour margin on the DWI original maps (b = 1200 s/mm²) with reference to corresponding T ₂W images. The software copies all ROIs to the ADC, D, D*, f, MD, MK,and native T1 map to calculate the average and minimum or maximum values. ¹⁷ In the control group, three to four consecutive slices with the largest cervical area were selected. ROIs included all cervical stroma and excluded the mucosa. The parameters were recorded as follows: mean native T1 (native T1), mean ADC (ADC_mean), minimum ADC (ADC_min), mean D (D_mean), minimum D (D_min), mean D* (D*), mean f (f), mean MD (MD_mean), minimum MD (MD_min), mean MK (MK_mean), and maximum MK (MK_max).

Statistical analysis

Statistical analyses were performed using SPSS v. 22.0 (IBM Corp., Armonk/NY) and MedCalc v. 19.0 (MedCalc Software, Mariakerke, Belgium). Intraclass correlation coefficients (ICCs) were used to evaluate the agreement of quantitative measures between two observers. ICC values greater than 0.8, 0.6–0.8, 0.4–0.6, and less than 0.4 were considered indicative of excellent, good, moderate, and poor agreement, respectively. The Shapiro–Wilk test was adopted to verify whether the measures conformed to a normal distribution. Continuous variables with a normal distribution were compared between groups using the independent samples t-test, whereas variables with a non-normal distribution were compared using the Mann–Whitney U test. Receiver operating characteristic curves were used to evaluate the diagnostic efficacy of each parameter, and the DeLong test was used to compare differences in the efficacy of each parameter. Binary logistic regression and receiver operating characteristic curve analyses were used to evaluate the diagnostic efficacy of the combination of the two parameters. Differences with p values < 0.05 were considered statistically significant.

Results

Study population

Figure 1 shows a flowchart of the study population selection. A total of 80 patients (mean age: 55.9 ± 11.9 years; range, 28–77 years) with histologically diagnosed cervical cancer were recruited into the patient group of this study: 38.8% (31/80) were pre-menopausal females and 61.2% (49/80) were post-menopausal females. 43 patients (mean age: 50.2 ± 10.4 years; range, 23–74 years) with a normal cervix were included in the control group. Among the control group, 44.2% (19/43) were pre-menopausal females and 55.8% (24/43) were post-menopausal females. The distribution of participant age (p = 0.218) and menstrual status (p = 0.558) did not differ between the patient group and the control group. Table 2 shows the clinical characteristics of the patient group.

Figure 1.

Flowchart of subject inclusion and exclusion.

Table 2.

Summary of patient characteristics

Characteristics	Number of patients
Total number of patients	80
Age (years), mean ± SD	52.9 ± 11.9
Pathological type
SCC	65
Adenocarcinoma	15
Histological grade
Low grade (well/moderately differentiated)	34
High grade (poorly differentiated)	27
Unknown	19
LVSI
Positive	26
Negative	24
Unknown	30
FIGO stage
IB	20
IIA	27
IIB	8
IIIA	3
IIIB	1
IIIC	21

FIGO, International Federation of Gynecology and Obstetrics; LVSI, lymphovascular space invasion; SCC, squamous cell carcinoma; SD, standard deviation.

Interobserver agreement

The ICCs of measured parameters of either the cervical cancer or normal cervix tissue were larger than 0.66, thus indicating good or excellent agreement between the two observers. Therefore, the average measurements by the two observers were utilised in the subsequent analysis. Table 3 summarises the ICC results.

Table 3.

Interobserver agreement for each MR parameter of cervical cancer and normal cervix

MR parameter	Intraclass correlation coefficient (ICC, 95% confidence interval)
	Cervical cancer	Normal cervix
Native T1 (ms)	0.91 (0.83–0.95)	0.94 (0.91–0.97)
ADCmean (×10−3 mm2/s)	0.88 (0.80–0.94)	0.92 (0.86–0.96)
ADCmin (×10−3 mm2/s)	0.84 (0.77–0.90)	0.90 (0.85–0.93)
Dmean (×10−3 mm2/s)	0.86 (0.76–0.92)	0.89 (0.75–0.95)
Dmin (×10−3 mm2/s)	0.82 (0.67–0.92)	0.86 (0.76–0.93)
D* (×10−3 mm2/s)	0.68 (0.54–0.79)	0.66 (0.50–0.78)
f	0.79 (0.56–0.88)	0.77 (0.64–0.83)
MKmean	0.83 (0.67–0.91)	0.82 (0.65–0.91)
MKmax	0.81 (0.66–0.89)	0.83 (0.68–0.91)
MDmean (×10−3 mm2/s)	0.88 (0.74–0.95)	0.87 (0.75–0.94)
MDmin (×10−3 mm2/s)	0.85 (0.74–0.92)	0.81 (0.67–0.89)

ADC, apparent diffusion coefficient; D*, pseudodiffusion; D, pure molecular diffusion; f, perfusion fraction; MD, mean diffusivity; MK, mean kurtosis; MR, magnetic resonance

Comparison of the quantitative parameters between the tumour and normal cervix

The ADC_mean, ADC_min, D_mean, D_min, f, MD_mean, and MD_min were significantly lower and the native T1, MK_mean, and MK_max were significantly higher in the patient group than in the control group (all p < 0.05). The difference in D* between the two groups was not significant (p = 0.637). Table 4 shows the MR quantitative parameters of the patient and control groups.

Table 4.

Native T1- and DWI-derived parameters in cervical cancer and normal cervix

Parameters	Tissue		p	Tumour type		p	SCC grade		p	SCC LVSI		p
	Tumour (n = 80)	Normal (n = 43)		SCC (n = 65)	AC (n = 15)		High-grade (n = 19)	Low-grade (n = 30)		Positive (n = 20)	Negative (n = 23)
Native T1 (ms)	1450.5 ± 101.5	1265.2 ± 159.5	0.004a	1433.4 ± 99.5	1524.7 ± 75.0	0.001a	1414.9 ± 113.7	1425.3 ± 90.8	0.726a	1418.4 ± 90.5	1404.0 ± 101.8	0.635a
ADCmean (×10−3 mm2/s)	0.84 ± 0.17	1.70 ± 0.30	<0.001b	0.84 ± 0.18	0.81 ± 0.12	0.653b	0.79 ± 0.16	0.87 ± 0.20	0.137b	0.81 ± 0.11	0.89 ± 0.25	0.098b
ADCmin (×10−3 mm2/s)	0.52 ± 0.14	1.08 ± 0.38	<0.001a	0.53 ± 0.15	0.47 ± 0.08	0.194a	0.45 ± 0.12	0.56 ± 0.16	0.014a	0.55 ± 0.12	0.51 ± 0.20	0.494a
Dmean (×10−3 mm2/s)	0.70 ± 0.11	1.33 ± 0.32	<0.001b	0.71 ± 0.11	0.68 ± 0.10	0.378b	0.67 ± 0.14	0.73 ± 0.09	0.062b	0.71 ± 0.11	0.72 ± 0.14	0.479b
Dmin (×10−3 mm2/s)	0.42 ± 0.12	0.92 ± 0.30	<0.001b	0.43 ± 0.12	0.37 ± 0.10	0.227b	0.37 ± 0.12	0.46 ± 0.12	0.004b	0.46 ± 0.09	0.41 ± 0.15	0.595b
D* (×10−3 mm2/s)	43.59 ± 18.50	49.14 ± 19.53	0.637a	45.44 ± 18.17	35.56 ± 18.38	0.062a	49.81 ± 19.27	44.35 ± 17.81	0.316a	49.75 ± 19.67	45.60 ± 18.77	0.719a
f	0.17 ± 0.08	0.28 ± 0.06	<0.001b	0.16 ± 0.05	0.20 ± 0.14	0.114b	0.14 ± 0.03	0.16 ± 0.04	0.087b	0.14 ± 0.03	0.17 ± 0.05	0.063b
MKmean	0.79 ± 0.20	0.40 ± 0.13	<0.001a	0.75 ± 0.17	0.99 ± 0.23	<0.001b	0.76 ± 0.19	0.72 ± 0.16	0.367b	0.79 ± 0.13	0.66 ± 0.20	0.017a
MKmax	1.50 ± 0.35	0.82 ± 0.30	<0.001b	1.45 ± 0.31	1.68 ± 0.45	0.051b	1.59 ± 0.34	1.38 ± 0.27	0.004b	1.38 ± 0.27	1.46 ± 0.34	0.960b
MDmean (×10−3 mm2/s)	1.03 ± 0.17	1.80 ± 0.31	<0.001b	1.04 ± 0.17	1.02 ± 0.16	0.844b	0.98 ± 0.19	1.06 ± 0.16	0.151b	1.02 ± 0.12	1.05 ± 0.22	0.487b
MDmin (×10−3 mm2/s)	0.54 ± 0.13	1.15 ± 0.31	<0.001a	0.55 ± 0.14	0.50 ± 0.10	0.204a	0.47 ± 0.10	0.60 ± 0.14	0.001a	0.56 ± 0.14	0.55 ± 0.17	0.761a

AC, adenocarcinoma; ADC, apparent diffusion coefficient; D, pure molecular diffusion; D*, pseudo-diffusion; DWI, diffusion-weighted imaging; LVSI, lymphovascular space invasion, MD, mean diffusivity; MK, mean kurtosis; SCC, squamous cell carcinoma; f, perfusion fraction.

The bold typeface in the table indicates statistically significant comparisons.

^aComparisons were performed via an independent samples t-test.

^bComparisons were performed via the Mann–Whitney U test.

Comparison of quantitative parameters between different histological characteristics of cervical cancer

Table 4 shows the detailed results. The native T1 and MK_mean were significantly lower in the SCC group than in the adenocarcinoma group (both p < 0.05); no other DWI-derived parameters differed significantly between the SCC and adenocarcinoma groups (all p > 0.05). ADC_min, D_min, and MD_min were significantly lower in the high-grade SCC group than in the low-grade SCC group, whereas MK_max was significantly higher in the high-grade SCC group than in the low-grade SCC group (all p < 0.05). Neither native T1 nor other DWI-derived parameters differed significantly between the two SCC groups (all p > 0.05). Within the SCC group, the LVSI-positive subgroup had a significantly higher MK_mean than the LVSI-negative subgroup (p < 0.05); no other MR parameters differed significantly between the two groups (all p > 0.05). Figures 2 and 3 show the representative images of normal cervix and cervical adenocarcinoma, respectively.

Figure 2.

A 39-year-old female with normal cervix. (a) The axial T ₂ weighted image and (b) DWI original map (b = 1200 s/mm²) show the normal cervical stroma and mucosa. (c) The native T1 map shows a relatively moderate T1 relaxation time; (d) ADC map; (e) D map; (f) D* map. (g) f map. (h) MK map. (i) MD map. ADC, apparent diffusion coefficient; D*, pseudo-diffusion; D, pure molecular diffusion; DWI, diffusion-weighted imaging; MD, mean diffusivity; MK, mean kurtosis.

Figure 3.

A 43-year-old female with G3, LVSI positive cervical adenocarcinoma (arrows). (a) The axial T ₂ weighted image shows a well-defined hyperintense mass confined to the cervix. (b) The DWI original map (b = 1200 s/mm²) shows a hyperintense mass with marked restricted diffusion. (c) Native T1 map shows a relatively increased T1 relaxation time. (d) ADC map. (e) D map. (f) D* map. (g) f map. (h) MK map. (i) MD map. ADC, apparent diffusion coefficient; D*, pseudo-diffusion; D, pure molecular diffusion; DWI, diffusion-weighted imaging; LVSI, lymphovascular space invasion; MD, mean diffusivity; MK, mean kurtosis.

Receiver operating characteristic analysis

Table 5 lists the area under the curve (AUC), threshold, sensitivity, and specificity of each parameter for discriminating the four categories (cervical cancer /normal cervix, tumour type, grade, and LVSI status). For distinguishing between cervical cancer and the normal cervix, D_mean achieved the highest AUC (0.994), followed by ADC_mean (0.991), MD_mean (0.990), MD_min (0.966), MK_mean (0.956), D_min (0.942), f (0.938), MK_max (0.934), ADC_min (0.911), and native T1 (0.838). The DeLong test revealed that the AUC for native T1 was significantly lower than those for other DWI parameters (all p < 0.05), except those for ADC_min and f, which did not differ significantly (p = 0.216, p = 0.056). For distinguishing between SCC and adenocarcinoma, the AUCs for MK_mean and native T1 were 0.800 and 0.783, respectively, and the difference was not statistically significant (p = 0.832). Combining MK_mean and native T1 improved the AUC to 0.852, which was significantly higher than that for MK_mean (p = 0.032). For distinguishing between high-grade SCC and low-grade SCC, the AUCs for MD_min, MK_max, D_min, and ADC_min were 0.787, 0.746, 0.746, and 0.729, respectively. There were no significant differences in AUCs among the different parameters (all p > 0.05). For distinguishing between LVSI-positive SCC and LVSI-negative SCC, only the AUC for MK_mean was statistically significant (AUC = 0.724, p = 0.004) (Figure 4).

Table 5.

Diagnostic performance of native T1- and DWI-derived parameters

Category	Threshold	AUC (95% CI)	p-value	Sensitivity	Specificity
Tumour vs normal
Native T1 (ms)	1295.4	0.838 (0.761–0.898)	<0.001	96.2%	58.1%
ADCmean (×10−3 mm2/s)	1.11	0.991 (0.954–1.000)	<0.001	96.2%	100.0%
ADCmin (×10−3 mm2/s)	0.69	0.911 (0.846–0.955)	<0.001	93.7%	86.0%
Dmean (×10−3 mm2/s)	0.90	0994 (0.960–1.000)	<0.001	96.2%	95.3%
Dmin (×10−3 mm2/s)	0.61	0.942 (0.884–0.976)	<0.001	98.7%	90.7%
f	0.22	0.938 (0.879–0.973)	<0.001	96.2%	90.7%
MKmean	0.55	0.956 (0.903–0.985)	<0.001	92.5%	95.3%
MKmax	1.05	0.934 (0.874–0.971)	<0.001	98.7%	79.1%
MDmean (×10−3 mm2/s)	1.24	0.990 (0.953–1.000)	<0.001	90.0%	100.0%
MDmin (×10−3 mm2/s)	0.76	0.966 (0.916–0.990)	<0.001	98.7%	90.7%
SCC vs AC
Native T1 (ms)	1481.5	0.783 (0.677–0.867)	<0.001	80.0%	69.2%
MKmean	0.91	0.800 (0.696–0.881)	<0.001	73.3%	89.2%
Native T1+MKmean	NA	0.852 (0.755–0.922)	<0.001	73.3%	92.3%
SCC grade: high vs low
ADCmin (×10−3 mm2/s)	0.51	0.729 (0.583–0.846)	0.002	78.9%	66.7%
Dmin (×10−3 mm2/s)	0.48	0.746 (0.602–0.860)	0.001	94.7%	50.0%
MKmax	1.52	0.746 (0.602–0.860)	0.001	68.4%	86.7%
MDmin (×10−3 mm2/s)	0.6	0.787 (0.646–0.891)	<0.001	100.0%	53.3%
SCC LVSI (＋) vs (－)
MKmean	0.65	0.728 (0.571–0.852)	0.004	95.0%	47.8%

AC, adenocarcinoma; ADC, apparent diffusion coefficient; AUC, area under the curve; CI, confidence interval; D, pure molecular diffusion; D*, pseudo-diffusion; DWI, diffusion-weighted imaging; LVSI, lymphovascular space invasion; MD, mean diffusivity; MK, mean kurtosis; SCC, squamous cell carcinoma; f, perfusion fraction.

Figure 4.

Receiver operating characteristic curves of native T1, ADC_mean, ADC_min, D_mean, D_min, f, MD_mean, MD_min, MK_mean, and MK_max for distinguishing between (a, b) cervical cancer and normal cervix, (c) SCC and adenocarcinoma, (d) high- and low-grade SCC, and (e) LVSI-positive and LVSI-negative SCC. ADC, apparent diffusion coefficient; D*, pseudo-diffusion; D, pure molecular diffusion; DWI, diffusion-weighted imaging; LVSI, lymphovascular space invasion; MD, mean diffusivity; MK, mean kurtosis SCC, squamous cell carcinoma.

Discussion

In this study, we conducted an initial investigation of the feasibility of T1 mapping and DWI by using monoexponential, biexponential, and DKI models for the non-invasive pre-operative evaluation of cervical cancer. Our findings indicated that T1 mapping and multimodel DWI can be used to distinguish cervical cancer from the normal cervix and SCC from adenocarcinoma. In addition, multimodel DWI may be useful for determining the pathological grade and LVSI status in patients with SCC.

T1 mapping for quantitative evaluation of cervical cancer

Quantitative MR imaging has a high sensitivity to tissue components and provides in vivo tissue characterisation. Native T1 is an intrinsic and fundamental tissue property, reflecting a composite signal of intracellular and extracellular composition and underlying pathophysiological processes. ¹¹ Currently, the principal techniques for obtaining the native T1 for bodily tissues include variable flip angle (VFA), Look–Locker, and MOLLI pulse sequences. The VFA sequence is quickly acquired in a single breath-hold but is known to be susceptible to B1 inhomogeneity of the MR system. Although acquisition times are relatively longer when using Look–Locker or MOLLI sequences, they are more stable, less affected by B1 magnetic field inhomogeneity, and more reproducible than VFA sequences. Tirkes et al ¹⁸ reported that the MOLLI sequence provided superior precision to VFA during phantom testing. In this study, we used a MOLLI-based T1 mapping sequence to determine the native T1 of cervical cancer and normal cervical tissue and observed that the native T1 was significantly higher in cervical cancer than in the normal cervix. The increased native T1 in cervical cancer may be due to the high cell density, increased protein, polypeptides, and other macromolecules, as well as micronecrosis within the tumour. ¹⁴

Cervical adenocarcinoma is a common subtype of cervical cancer second only to SCC in incidence and is less sensitive to chemoradiotherapy than SCC, with an increased likelihood of metastasis and poorer prognosis. Therefore, obtaining more information about the differences in tumour heterogeneity between SCC and adenocarcinoma may provide complementary information for biopsy, which facilitates individualised therapeutic management. This study demonstrated that the native T1 of adenocarcinoma was significantly higher than that of SCC, which may be related to the differences in histological structure between the two tumours. Different from SCC, cervical adenocarcinoma originates from endocervical cells with abundant glandular structures and strong secretion function, which may lead to increased water content and macromolecular concentrations in tumour tissue and thus increased native T1. Our results were consistent with previous studies. For example, Chang et al ¹⁹ stated that native T1 can be used to distinguish between different types of thymic epithelial tumours and Li et al ¹⁶ found significant differences in native T1 among different types of lung cancer. In addition, the present study found no significant difference in native T1 between high- and low-grade SCC or between LVSI-positive and LVSI-negative SCC. These results suggest that either different SCC grades or different LVSI statuses affected the native T1 to a similar extent.

Multimodel DWI for quantitative evaluation of cervical cancer

Our analyses revealed significant differences between cervical cancer and the normal cervix for all parameters except D*, which was consistent with previous studies. ¹⁰ Compared with normal cervical tissue, tumour tissue exhibits higher cellularity, resulting in significant restriction of water molecule diffusion and thus significantly lower diffusion parameters. The parameter MK (range: 0–1) has been shown to be associated with the complexity of various diseases. ^20,21 In the present study, significantly increased MK_mean and MK_max values were found in cervical cancer, which may be due to the increased complexity caused by a higher degree of heterogeneous cellularity, tortuous vascular hyperplasia, and intravoxel microscopic necrotic foci.

Previous studies have evaluated the potential for different DWI models to differentiate between cervical SCC and adenocarcinoma, but their results were mixed. ^9,21–23 Winfield et al ⁹ showed that IVIM-derived f and D* and DKI-derived MK exhibited significant differences in the differentiation of SCC and adenocarcinoma. Moreover, Wang et al ²¹ and Zhang et al ²² found that ADC and MD can both be used for the typing of cervical cancer, whereas Meng et al ²³ found that both MD and MK showed discriminative value in this aspect. In the present study, only MK_mean was effective in discriminating between SCC and adenocarcinoma, which was also quite inconsistent with the results of previous studies. ^9,21–23 There are three possible reasons for the mixed results. The first is the choice of b values, including the number of b values and the maximum b value. ^24,25 The second is the reproducibility of the different model parameters. Previous studies have shown that the reproducibility of IVIM parameters (especially f and D*) is significantly lower than that of ADC derived from the monoexponential model. ^26,27 The third is the different research subjects. The present study included a low proportion of adenocarcinoma cases, whereas differences in the distribution of cases between SCC and adenocarcinoma were relatively small in other studies, ^9,21–23 which may have affected the present results. Therefore, large-sample-size, multicentre studies are still necessary to confirm the potential of different DWI parameters for cervical cancer typing.

Owing to the limited number of adenocarcinoma cases, we only analysed histological grades and LVSI status for SCC. Our study revealed that ADC_min, D_min, and MD_min were significantly lower while MK_max was significantly higher in high-grade SCC than in low-grade SCC. The increased tumour cellularity and higher nuclear-to-cytoplasmic ratios in high-grade SCC, which are not conducive to the diffusion of water molecules, may result in decreased D_min, MD_min, and ADC_min, whereas increased nuclear polymorphism and cellular heterogeneity may lead to increased MK_max. However, we observed no significant differences in the mean values of multimodel DWI parameters in terms of cervical cancer grades, similar to findings reported by Wang et al. ²¹ The measurement of mean values of diffusion parameters may be subject to underestimation of tumour cellularity, especially when the tumour contains more areas of micronecrosis, whereas the minimum values of ADC, D, and MD correspond to the lowest tumour diffusion region, which may be the most actively proliferating area. ²⁸ Similarly, the maximum value of MK indicates the maximum heterogeneity of tumours, which is also an important feature of high-grade tumours. Ghosh et al ²⁹ reported that ADC_min was superior to ADC_mean in distinguishing MYCN-amplified from non-amplified neuroblastoma. Arslan et al ³⁰ found that ADC_min, D_min, MD_min, and MK_mean could be used to differentiate benign and malignant musculoskeletal tumours. Xiao et al ³¹ found that MK_max derived from DKI is the strongest independent factor for the prediction of the Ki-67 proliferation status of sinonasal malignancies. Additionally, our study indicated that MK_mean was significantly higher in LVSI-positive SCC than in LVSI-negative SCC, thus suggesting that LVSI is more likely to occur in SCC with more complex histology. Moreover, tumour tissue containing capillaries with a high degree of degeneration, necrosis, and occlusion is prone to metastasis, which may also result in an increased MK_mean.

Comparison between T1 mapping and multi-model DWI

In this study, receiver operating characteristic analysis revealed that native T1 and MK_mean had similar diagnostic efficacy, but only with respect to the pathological typing of cervical cancer. Multimodel DWI was superior to T1 mapping in the differentiation of cervical cancer from the normal cervix and in the prediction of high-grade SCC and LVSI-positive SCC. Given that multimodel DWI can reflect the pathological features with respect to tumour cellularity, heterogeneity, and vascularity, it may provide incremental diagnostic value beyond native T1. However, the T1 mapping approach of this study is already commercially available and T1 maps can be generated online automatically without the complex mathematical model. The acquisition time of T1 mapping is also acceptable. Our results achieved a significant information gain for distinguishing between SCC and adenocarcinoma by combining native T1 and MK_mean. Therefore, further work may focus on the possibility of T1 mapping integrated into multiparametric approaches or radiomics analyses.

This study has some limitations. First, owing to the relatively low incidence of adenocarcinoma, the sample size of patients with adenocarcinoma was small. Second, rare histological subtypes such as adenosquamous carcinoma and neuroendocrine carcinomas of the cervix were excluded from the study, which may have resulted in selection bias. Third, we used 11 b values for the DWI scans, but we did not evaluate the variability of different b value combinations for the multimodel DWI. Further studies are required to identify the optimal b values that can provide the most useful information for pre-operative evaluation of cervical cancer.

Conclusions

In summary, the current findings indicate that both T1 mapping and multimodel DWI can be used to discriminate cervical cancer from normal cervical tissue and cervical adenocarcinoma from SCC. However, multimodel DWI may also be useful for predicting histological grade and LVSI status in cases of SCC, providing more comprehensive information for non-invasive pre-operative evaluation of cervical cancer.