Abstract
Objectives:
To determine the diagnostic ability of cervical mucosa radiomics signature of sagittal T2WI and T1 contrast-enhanced (CE) imaging in detecting early-stage cervical cancers with negative MRI.
Methods:
Preoperative images of postoperative pathology confirmed early-stage cervical cancer patients and normal cervix patients admitted to our hospital between January 2013 and December 2020 were retrospectively reviewed. Patients with cancer signals on T2WI, T1CE and DWI were deleted. Regions of interests (ROIs) were delineated on cervical mucosa (from cervical canal to cervical dome) with 5 mm width on sagittal T2WI and T1CE. The maximum-relevance and minimumredundancy (mRMR) and least absolute shrinkage and selection operator (LASSO) methods were used for the calculation of radiomics signature scores. Diagnostic performance was assessed and compared between radiomics prediction models (model 1: T1CE; model 2: T2WI; model 3: model one combined with model 2). Differential diagnostic ability of radiomics signature in detecting lymphatic vascular space invasion (LVSI) was further explored.
Results:
Diagnostic performance of model three was higher than model 1 and model 2 both in primary (model 3 0.874, model 1 0.857, model 2 0.816) and validation (model 3 0.853, model 1 0.847, model 2 0.634) cohorts. Model 3 showed statistical diagnostic difference compared with model 2 (primary p = 0.008, validation p = 0.000). However, the diagnostic improvement ability of model 3 showed no statistical difference compared with model 1 (primary p = 0.351, validation p = 0.739). Diagnostic efficiency of model 3 in detecting LVSI was not apparent (AUC 0.64).
Conclusions:
Radiomics analysis of cervical mucosa combining T1CE and T2WI is promising for predicting MRI invisible early-stage cervical cancers, however further ability in detecting LVSI was not apparent.
Advances in knowledge:
Conventional MRI was originally defined as meaningless in very early-stage cervical cancers. However, whether MRI radiomics analysis of cervical mucosa can detecting tiny changes of invisible early stage cervical cancers has not been researched yet.
Introduction
Cervical cancer is the fourth most common malignancy in females worldwide. With the advent of promising screening programs for pre-cancerous lesions, the incidence rate of cervical cancers has declined by 65%, but early-stage cervical cancer diagnosis with a small volume is gradually increasing. 1,2 Especially, adenocarcinoma in early-stage cervical cancers has shown an increase of up to 25%, 3 but cervical cytology screening is less effective and often fails to detect adenocarcinoma. 4 Glandular precancerous lesions (adenocarcinoma in situ) are judged to be harder to sample than squamous precancerous lesions, because they typically develop within the endocervical canal and then often fail to be sampled by cytology. 5 Therefore, preoperative accurate diagnosis of early-stage cervical cancers—particularly endocervical adenocarcinomas—is very difficult, although important to the prognosis.
Magnetic resonance imaging (MRI) is commonly used in identifying and staging gynecological tumors. 6 Although early-stage cervical tumors of >1 cm have a detection rate of ~60%, none of the stage IA cervical cancers were detected in an MRI study because of limited spatial resolution. 7,8 An inconveniently placed endovaginal coil was tentatively applied to improve detection rate, but engendered relatively low acceptability. 1 MRI dynamic contrast-enhanced (CE) analysis has also been attempted, but only provided moderate detection ability (AUC 0.705). 9 However, early detection of MRI-missed cervical cancers can reduce a patient’s need to undergo radical fertility-sparing surgery and increase recurrence-free and survival rates. 10 Therefore, convenient, noninvasive, and accurate image-analysis methods that provide greater applicative potential need to be further explored.
Radiomics analysis extracts a large number ultrastructural quantitative features and could reflect regional microenvironmental tumor heterogeneity so as to improve diagnostic, prognostic, and predictive accuracy. 11,12 Previous MRI radiomics analyses of cervical cancers have detected pathological differences and therapeutic response such as lymph vascular space invasion (LVSI) and chemotherapy sensitivity. 13,14 Radiomics reached higher accuracy and performance than clinical parameters. 15 But their radiomic regions of interest (ROIs) were all based on large, MR-visible cancers. One investigation focused on detecting micro-metastases of lymph nodes earlier than morphological changes. 16 Thus, we wished to ascertain whether micro-canceration of cervical mucosa can also be detected. According to previous pathological research on the origins of early-stage cervical cancers, most HPV-related cancers were located in the basal cell layer of the mucosa, 17 and adenocarcinomas were located in the glandular epithelial cells of the endocervical mucosa. 18 Hence, the ROIs in the current study were delineated by analyzing the tumors invisible to MRI from the cervical canal mucosa to the cervical dome mucosa. Thus, elevated suspicions of early cancers upon MRI may contribute to clinical treatment strategies.
LVSI is an important pathologic evaluation factor following surgery to treat cervical cancers. In stage IA patients, lymphadenectomy principally depends upon the presence of LVSI, 19 and it is also associated with diminished 5-year disease-free survival in stage IA and IB1 cervical cancers. 20,21 Patients with LVSI also undergo significantly more radiotherapy and chemotherapy than patients without LVSI. 22
Therefore, we aimed herein 1 to establish and validate radiomics models of cervical mucosa based on T2WI and T1CE for the prediction of MRI-invisible early-stage cervical cancers, and 2 to further explore the detection ability of a radiomics signature for LVSI in MRI-invisible cervical cancers.
Methods and materials
Patients
This retrospective study was approved by the Institutional Review Board of our hospital, and the requirement for informed consent was waived. The preoperative MR images of 161 cervical cancer patients (45.43 ± 9.62 years of age, from January 2013 to December 2020) and 115 normal cervix patients (44.87 ± 11.15 years of age, from January 2017 to December 2020) who had been previously admitted to our hospital were retrospectively reviewed (Figure 1). The MRI-invisible cervical cancers were all corroborated by postoperative pathology, and normal cervices were confirmed by thin-prep cytology tests in triplicate or by postoperative pathology (due to uterine leiomyomas or ovarian lesions). No obvious malignant cells or inflammatory reactive cells were included in the normal cervical group. We randomly allocated the subjects into primary and validation cohorts at a ratio of 7:3. The primary cohort was composed of 195 patients with 120 early stage MRI-invisible cancers (FIGO stage IA, 73; IB, 47), and the validation cohort encompassed 81 patients with 41 early stage MRI-invisible cancers (FIGO stage IA, 23; IB, 18).
Figure 1.
Flowchart of the inclusion, exclusion, and grouping criteria for cervical cancer and normal cervix patients.
We used the following inclusion criteria (1): patients who were pathologically confirmed with normal cervices or with cervical cancer that included carcinoma in situ; (2) patients without visible cancer signals upon T2WI, T1CE, or DWI; (3) the availability of clinical characteristics; and (4) the presence of qualified MR images. Patients were excluded for the following reasons : (1) having pathologically confirmed precancerous lesions ; (2) having an atrophic and thin (less than 5 mm) cervical mucosa; (3) having undergone adjuvant chemotherapy or radiotherapy performed prior to MRI; and (4) possessing endocervical endometrial cancers.
MRI acquisition
MRI examinations were performed using v. 3.0 T (MAGNETOM TrioTim, Siemens) and 1.5 T MRI scanners (MAGNETOM Aera, Siemens). Previous radiomics analysis executed by combining data from 3T and 1.5T images exhibited a high level of stability. 23,24 For 3T, we acquired sagittal T2WI with a TR/TE of 4520/125 ms, a field of view of 250 × 250 mm2, and section thickness of 3 mm; sagittal T1CE images were acquired with a TR/TE of 650/9.3 ms, field of view of 279 × 279 mm2, and section thickness of 3.5 mm. For 1.5T, we acquired sagittal T2WI with a TR/TE of 4430/90 ms, field of view of 230 × 230 mm2, and section thickness of 3.5 mm; sagittal T1CE images were acquired with a TR/TE of 420/21 ms, field of view of 230 × 230 mm2, and a section thickness of 3.5 mm. An automated injector system (Stellant MR Injection System, Medrad, Germany) was used (GE HealthCare, 0.5 mmol ml−1), and we calculated dosage (ml) = weight (kg) ×0.2 ml/kg, with a flow rate of 3 ml s−1.
Image segmentation
We used ITK-SNAP (v.3.6.0; www.itksnap.org) for manual 3D segmentation of MRIs. ROIs were manually segmented on each sagittal section by a radiologist with seven years of pelvic MRI-diagnosis experience and a senior radiologist with 20 years of pelvic MRI-diagnosis experience to validate each processed segmentation. They were both blinded to pathology results. ROIs were delineated on cervical mucosa of the same width (5 mm, which could be unified by ITK-SNAP) on T2WI and T1CE images from the cervical canal to the cervical dome. The reliability between ROIs across scanners and across observers was calculated using the interclass correlation coefficient (ICC), and features with an ICC higher than 0.8 were reserved.
Extraction of radiomics features
Feature extraction was performed using AK software (Analysis Kit, GE Healthcare). A total of 402 features were separately extracted including first-order features (histogram features based on the gradient), second-order features (grayscale co-occurrence matrix, gray-level run-length matrix, and gray-level zone-length matrix), and size- and shape-related features.
Feature selection and signature construction
Cervical mucosal features (n = 402) were extracted from T2WI and T1CE images. After implementation of maximum-relevance and minimum-redundancy (mRMR) to eliminate the redundant and irrelevant features, 11 T2WI features and 17 T1CE features were retained. We then used the least absolute shrinkage and selection operator (LASSO) regression-analysis method to choose the optimized subset of features. After the total number of features was determined, the most predictive subset of features was chosen and the corresponding coefficients were evaluated. We then calculated radiomics score (Rad-score) by summing the selected features, weighted by their coefficients.(supplementary material)
Statistical analysis
We conducted statistical analysis using R (https://www.r-project.org), MedCalc (v.15.0), and SPSS (PASW Statistics v.18.0 SPSS). The independent-sample Student’s t test was used to compare continuous variables (age and Rad-score).
The diagnostic performance of the Rad-score was assessed using receiver operating characteristic (ROC) curves. The diagnostic accuracy, area under the ROC curve (AUC), sensitivity, specificity, positive-predictive value (PPV), and negative-predictive value (NPV) were also calculated. The comparison of diagnostic performance (AUC) between model 3 and models 1 and 2 was conducted with the DeLong test.
The performances of the three models were validated in the validation cohort using their ROC curves. The diagnostic accuracy, AUC, sensitivity, specificity, PPV, and NPV of the models were also calculated. We then compared differences in diagnostic performance (AUC) with the DeLong test.
The diagnostic ability of the Rad-score in detecting LVSI among early-stage cervical cancers was then further analyzed. A P-value of < 0.05 was considered a significant difference.
Results
Patient characteristics
A total of 276 patients (45.20 ± 10.27 years of age) with 161 early-stage cervical cancers and 115 patients with normal cervices were enrolled and allotted to the primary cohort (n = 195) or the validation cohort (n = 81). The preoperative characteristics of the cervical cancer patients are listed in Table 1. There was no significant difference in age between the primary and validation cohorts (p = 0.95, p = 0.46). However, Rad-scores (T1CE, T2WI, and combined) were all higher in the cancer group than in the normal cervix group (Table 1). The radiomics analysis process is presented in Figure 2.
Table 1.
Characteristics of Patients in the Primary and Validation Cohorts
| Characteristic | Primary Cohort (n = 195) | P-value | Validation Cohort (n = 81) | P-value | ||
|---|---|---|---|---|---|---|
| Cancer (n = 120) | Normal (n = 75) | Cancer (n = 41) | Normal (n = 40) | |||
| Age (Mean ± SD) | 44.8 ± 9.1 | 44.7 ± 10.1 | 0.95 | 47.1 ± 10.9 | 45.1 ± 13.0 | 0.46 |
| FIGO stage (n%) | ||||||
| IA | 73 (60.8) | 23 (56.1) | ||||
| IB | 47 (39.2) | 18 (43.9) | ||||
| Histology (n%) | ||||||
| Squamous cell carcinoma | 88 (73.3) | 30 (73.2) | ||||
| Adenocarcinoma | 30 (25.0) | 10 (24.4) | ||||
| Others | 2 (1.7) | 1 (2.4) | ||||
| LVSI (n%) | ||||||
| Positive | 39 (32.5) | 13 (29.3) | ||||
| Negative | 75 (62.5) | 28 (68.3) | ||||
| Lymph node (n%) | ||||||
| Positive | 9 (7.5) | 1 (2.4) | ||||
| Negative | 105 (87.5) | 40 (97.6) | ||||
| Tumor Rad-score (T1CE) | 1.45 ± 1.40 | −0.85 ± 3.25 | 0.00* | 1.57 ± 1.70 | −0.50 ± 1.72 | 0.00* |
| Tumor Rad-score (T2WI) | 1.07 ± 1.00 | −0.15 ± 0.99 | 0.00* | 1.10 ± 1.20 | 0.61 ± 1.04 | 0.06 |
| Tumor Rad-score (Combine) | 1.52 ± 1.37 | −0.89 ± 2.74 | 0.00* | 1.67 ± 1.05 | −0.12 ± 1.56 | 0.00* |
FIGO: Federation International of Gynecology and Obstetrics; LVSI: Lymph vascular space invasion
P-values were derived from independent-sample T-test (age, tumor Rad-score)
T1CE: T1 contrast-enhanced imaging
T2WI: T2 -weighted imaging
* p < 0.05
Figure 2.
The radiomics-analysis process. First, regions of interest (ROIs) of cervical mucosa were manually segmented in 3D; and 402 radiomics features were separately extracted from T1CE and T2WI modalities. The optimized subset of features was then selected and Rad-scores were calculated. Finally, the diagnostic performance (ROC) of the three models in the primary and validation cohorts were compared (model 1, T1CE; model 2, T2WI; model 3, combined).
Diagnostic performance of model 1 (T1CE) in detecting MRI-invisible cervical cancers
We chose the 17 most predictive features of T1CE. The Rad-score diagnostic AUC of model 1 was 0.857 (range: 0.804–0.911), with an accuracy of 0.778, sensitivity of 75.9%, specificity of 80.8%, PPV of 86.3%, and an NPV of 67.8% (Table 2).
Table 2.
Diagnostic Performance comparison in the Primary and Validation Cohorts
| Cohorts | AUC | Accuracy | Sensitivity | Specificity | PPV | NPV | |
|---|---|---|---|---|---|---|---|
| Primary | T1CE | 0.857 | 0.778 | 0.759 | 0.808 | 0.863 | 0.678 |
| T2WI | 0.816 | 0.772 | 0.793 | 0.740 | 0.829 | 0.692 | |
| Combined | 0.874 | 0.799 | 0.776 | 0.836 | 0.882 | 0.701 | |
| Validation | T1CE | 0.847 | 0.813 | 0.825 | 0.800 | 0.805 | 0.821 |
| T2WI | 0.634 | 0.575 | 0.800 | 0.350 | 0.552 | 0.636 | |
| Combined | 0.853 | 0.800 | 0.900 | 0.700 | 0.750 | 0.875 |
AUC: the area under the curve; NPV: positive-predictive value;PPV: positive-predictive value.
T1CE: T1 contrast-enhanced imaging
T2WI: T2 -weighted imaging
Diagnostic performance of model 2 (T2WI) in detecting MRI-invisible cervical cancers
The 11 most predictive features of T2WI were chosen. The Rad-score diagnostic AUC of model 2 was 0.816 (range: 0.752–0.881), with an accuracy of 0.772, sensitivity of 79.3%, specificity of 74.0%, PPV of 82.9%, and NPV of 69.2% (Table 2).
Diagnostic performance of model 3 (combined) in detecting MRI-invisible cervical cancers
Radiomics features of T1CE and T2WI were combined and the 16 most predictive features were selected. The Rad-score diagnostic AUC of model 3 was 0.874 (range: 0.824–0.924), with an accuracy of 0.799, sensitivity of 77.6%, specificity of 83.6%, PPV of 88.2%, and NPV of 70.1% (Table 2).
Validation of models
Model one achieved a diagnostic AUC of 0.847 (range: 0.780–0.934), with an accuracy of 0.813, sensitivity of 82.5%, specificity of 80.0%, PPV of 80.5%, and NPV of 82.1%.
Model two achieved a diagnostic AUC of 0.634 (range: 0.510–0.757), with an accuracy of 0.575, sensitivity of 80.0%, specificity of 35.0%, PPV of 55.2%, and NPV of 63.6%.
Model three achieved a diagnostic AUC of 0.853 (range: 0.768–0.938), with an accuracy of 0.800, sensitivity of 90.0%, specificity of 70.0%, PPV of 75.0%, and NPV of 87.5% (Table 2).
Comparison of diagnostic performance among the three models
The diagnostic performance (AUC) of model three was higher than either model one or 2, both in the primary (model 3, 0.874; model 1, 0.857; model 2, 0.816) and validation cohorts (model 3, 0.853; model 1, 0.847; model 2, 0.634). Although model three showed a statistical difference compared with model 2 (primary p = 0.008, validation p = 0.000), the improvement ability of model three showed no difference compared with model 1 (primary p = 0.351, validation p = 0.739) (Figure 3).
Figure 3.
Comparison of diagnostic performance in the primary and validation cohorts.
Examples of predictive ability comparisons of Rad-score on MRI-invisible cervical cancers are depicted in Figure 4. Rad-scores (T1CE, T2WI, and combined) were all higher in the cervical cancer group than the normal cervix group.
Figure 4.
Examples of predictive-ability comparisons of radiomics signatures on MRI-invisible cervical cancers. Rad-scores (T1CE, T2WI, and combined) were all higher in the cervical cancer group than in the normal cervix group.
Diagnostic performance of Rad-scores in detecting LVSI
Although Rad-score (combined) was higher in LVSI (+) patients than in LVSI (−) patients of the primary cancer group (2.01 ± 1.53 vs 1.33 ± 1.24, respectively, p = 0.01), this tendency was not observed for the validation group (p = 0.98). Rad-score (T1CE, T2WI) also did not show statistical significance in either the primary or validation group (Table 3).
Table 3.
Comparison of cervical mucosa radiomics signature in detecting LVSI in early stage cervical cancers
| Characteristic | Primary Cohort (n = 114) | P-value | Validation Cohort (n = 41) | P-value | ||
|---|---|---|---|---|---|---|
| LVSI
(+) (n = 39) |
LVSI
(-) (n = 75) |
LVSI
(+) (n = 13) |
LVSI
(-) (n = 28) |
|||
| Tumor Rad-score (T1CE) | 1.71 ± 1.70 | 1.36 ± 1.23 | 0.23 | 1.80 ± 1.58 | 1.56 ± 1.80 | 0.69 |
| Tumor Rad-score (T2WI) | 1.34 ± 0.94 | 0.96 ± 1.03 | 0.06 | 1.16 ± 0.78 | 1.10 ± 1.39 | 0.88 |
| Tumor Rad-score (Combine) | 2.01 ± 1.53 | 1.33 ± 1.24 | 0.01* | 1.71 ± 0.87 | 1.72 ± 1.14 | 0.98 |
LVSI: Lymph vascular space invasion
P-values were derived from independent-sample T test
T1CE: T1 contrast-enhanced imaging
T2WI: T2 -weighted imaging
*p<0.05
The diagnostic capabilities with respect to Rad-score AUCs in detecting LSVI were 0.553 for T1CE, 0.622 for T2WI, and 0.639 for the combined model.
Discussion and conclusions
Early detecting of cervical cancers can influence treatment strategy and improve prognosis. Our current investigation focused on using radiomics analysis to detect MRI-invisible early-stage cancers of the cervical mucosa, and we observed that the combination of T1CE and T2WI was optimal for predicting invisible early-stage cervical cancers, with an AUC of 0.874. However, any further ability to use the radiomics signature in detecting LVSI was not apparent (AUC, 0.64).
In earlier published studies, MRI was perceived as being inconsequential in detecting very early-stage cervical cancers, 25 and, more importantly, the sensitivity rate with cytological smears of early stage cervical cancers was only ~60–70%. 26 Detection was especially difficult in endocervical adenocarcinoma due to its deep location and wide variety of complicated proliferations 27 ; thus, early cervical abnormalities could be easily missed and untreated. Only one analysis focused on detecting MRI-invisible early-stage cervical cancers; these authors discovered that the Ktrans of the DCE MRI parameters exhibited the highest diagnostic AUC of 0.705, with a sensitivity of 67.6% and specificity of 68% in detecting MRI-invisible, residual cervical cancers after conization. 9 The combined diagnostic ability of T2WI and T1CE imaging (AUC 0.874), sensitivity, (77.6%) and specificity (83.6%) in our study was discernibly higher than in the aforementioned study.
Our occurrence rate for adenocarcinoma was high with a rate of 25%, and this coincided with the increasing adenocarcinoma trend observed for early-stage cervical cancers. 28 In a study focused that on early-stage cervical cancer (<5 mm of depth invasion), the LVSI rate was 26.9% (the squamous-cell carcinoma rate was 32.1% and the adenocarcinoma was 15.7%) and the lymph-node metastasis rate was 3.5% 19 ; the slightly elevated occurrence rate in our study (32.3%) was thus comparable. The detection of LVSI was significant even in very early-stage cervical cancers. In stage IA patients, lymphadenectomy primarily depended upon the presence of LVSI, 19 and LVSI was also associated with reduced 5-year disease-free survival and additional chemoradiotherapy after surgery. 22 However, a previous analysis revealed that the detection ability of LVSI using a radiomics nomogram for all surgery patients was moderate (primary cohort, 0.754; validation cohort, 0.727), 14 while the diagnostic ability in our evaluation of LVSI among very early-stage cancers was relatively lower (AUC, 0.64). Therefore, further research focusing on LVSI in very early-stage cancers is of clinical significance and sorely needed.
Radiomics analysis of MRI has been previously used to study early-stage cervical cancers—including its various subtypes, 29 parametrial invasion, 30 lymph-node metastasis, 31 and survival. 32 However, their ROIs were all based on visible cancers. The detection ability of a fixed region prior to morphological change has not yet been undertaken. Previous pathological analyses determined that most HPV-related cancers are located in the basal cell layer of the mucosa, 17 and that adenocarcinomas are localized to the glandular epithelial cells of the endocervical mucosa. 18 The ROIs we used were delineated so as to assess MRI-invisible tumors from the cervical canal mucosa to the cervical dome mucosa. We therefore hypothesized that radiomics analysis would allow us to detect miniscule changes, and our results showed favorable efficiency when combining T2WI and T1CE in detecting MRI-invisible early-stage cervical cancers. The diagnostic performance of T1CE was greater than that of T2WI, and on the other hand, hemodynamic changes can be observed earlier than inflammatory or edematous changes in very early cervical cancers.
Another intriguing finding was that our two cases of large, MRI-invisible adenocarcinomas were negative upon T2WI, T1CE, and DWI, but that pathology was confirmed at ~4 cm. Although we know of no study that has addressed such a discrepancy, it may be attributed to the characteristics of adenocarcinomas (which possess a lower DWI signal than squamous-cell carcinoma) or menopausal state (non-menopausal patients show lower DWI signals). 33,34 Thus, the accurate pathological specificity caused by imaging disparities needs to be further explored.
There were several limitations to our study. First, MR images were collected jointly using 3.0 T and 1.5 T scanners. However, radiomics features still exhibited a high level of stability, as previously verified by researchers. We will continue to collect MRI-invisible cancer cases and further analyze them on a single fixed scanner. Second, we did not compare the detection ability between squamous-cell carcinomas and adenocarcinomas. Clinical and pathological detection was also particularly difficult in early endocervical adenocarcinomas due to their deep location and wide variety of complicated proliferations. Third, other functional sequences, such as DWI, were not included, as the relatively low resolution of DWI did not allow accurate delineation of the cervical border.
In conclusion, radiomics analysis of cervical mucosa that combines T1CE and T2WI is promising in the prediction of MRI-invisible early stage cervical cancers. However, improving the detection ability for LVSI requires further study.
Our original data can be accessed at https://pan.baidu.com/s/1WCqe0d5G7w6OU5YC16lRrQ upon request from the corresponding author, who will supply the password.
Supplementary Material
Footnotes
The authors Qiming Hu and Jinming Shi contributed equally to the work.
Contributor Information
Qiming Hu, Email: huqiming101@163.com, Department of Obstetrics & Gynecology, the First Affiliated Hospital of Nanjing Medical University, Nanjing, China .
Jinming Shi, Email: jinmingjs@126.com, Department of Radiology, the First Affiliated Hospital of Nanjing Medical University, Nanjing, China .
Aining Zhang, Email: ningzc123@126.com, Department of Radiology, the First Affiliated Hospital of Nanjing Medical University, Nanjing, China .
Shaofeng Duan, Email: duanfeng12@126.com, GE Healthcare, Precision Health Institution, Shanghai, China .
Jiacheng Song, Email: yysjc1990@163.com, Department of Radiology, the First Affiliated Hospital of Nanjing Medical University, Nanjing, China .
Ting Chen, Email: chentingwzc@163.com, Department of Radiology, the First Affiliated Hospital of Nanjing Medical University, Nanjing, China .
REFERENCES
- 1. Downey K, Attygalle AD, Morgan VA, Giles SL, MacDonald A, et al. . Comparison of optimised endovaginal vs external array coil t2-weighted and diffusion-weighted imaging techniques for detecting suspected early stage (ia/ib1) uterine cervical cancer . Eur Radiol 2016. ; 26: 941 – 50 . doi: 10.1007/s00330-015-3899-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Park J-Y, Lee J-W, Park BK, Lee Y-Y, Choi CH, et al. . Postoperative outcomes of mr-invisible stage ib1 cervical cancer . Am J Obstet Gynecol 2014. ; 211: 168 . S0002-9378(14)00174-4 . doi: 10.1016/j.ajog.2014.02.032 [DOI] [PubMed] [Google Scholar]
- 3. van der Horst J, Siebers AG, Bulten J, Massuger LF, de Kok IM . Increasing incidence of invasive and in situ cervical adenocarcinoma in the netherlands during 2004-2013 . Cancer Med 2017. ; 6: 416 – 23 . doi: 10.1002/cam4.971 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Castanon A, Landy R, Sasieni PD . Is cervical screening preventing adenocarcinoma and adenosquamous carcinoma of the cervix? Int J Cancer 2016. ; 139: 1040 – 45 . doi: 10.1002/ijc.30152 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Sasieni P, Castanon A, Cuzick J . Screening and adenocarcinoma of the cervix . Int J Cancer 2009. ; 125: 525 – 29 . doi: 10.1002/ijc.24410 [DOI] [PubMed] [Google Scholar]
- 6. Moro F, Bonanno GM, Gui B, Scambia G, Testa AC . Imaging modalities in fertility preservation in patients with gynecologic cancers . Int J Gynecol Cancer 2021. ; 31: 323 – 31 . doi: 10.1136/ijgc-2020-002109 [DOI] [PubMed] [Google Scholar]
- 7. Itoh K, Shiozawa T, Ohira S, Shiohara S, Konishi I . Correlation between mri and histopathologic findings in stage i cervical carcinomas: influence of stromal desmoplastic reaction . Int J Gynecol Cancer 2006. ; 16: 610 – 14 . doi: 10.1111/j.1525-1438.2006.00383.x [DOI] [PubMed] [Google Scholar]
- 8. Hriack H, Gatsonis C, Coakley F, Snyder B, Reinhold C, et al. . Early invasive cervical cancer: ct and mr imaging in preoperative evaluation . Radiology 2007. ; 245: 491 – 98 . [DOI] [PubMed] [Google Scholar]
- 9. Huang JW, Song JC, Chen T, Yang M, Ma ZL . Making the invisible visible: improving detectability of mri-invisible residual cervical cancer after conisation by dce-mri . Clin Radiol 2019. ; 74: 166 . S0009-9260(18)30582-8 . doi: 10.1016/j.crad.2018.10.013 [DOI] [PubMed] [Google Scholar]
- 10. Roh HJ, Go EB, Kim KB, Lee JH, Lee SH . The diagnostic accuracy and postoperative outcomes of cervical cancer patients for mr-invisible or mr-visible diagnosis of combined t2- and diffusion-weighted 3t mri using the external phased-array receiver . Anticancer Res 2019; 39: 6945–56. 10.21873/anticanres.13916 [DOI] [PubMed] [Google Scholar]
- 11. Jin X, Ai Y, Zhang J, Zhu H, Jin J, et al. . Noninvasive prediction of lymph node status for patients with early-stage cervical cancer based on radiomics features from ultrasound images . Eur Radiol 2020. ; 30: 4117 – 24 . doi: 10.1007/s00330-020-06692-1 [DOI] [PubMed] [Google Scholar]
- 12. Caruso D, Polici M, Zerunian M, Pucciarelli F, Guido G, et al. . Radiomics in oncology, part 1: technical principles and gastrointestinal application in ct and mri . Cancers (Basel) 2021; 13(11): 2522. 10.3390/cancers13112522 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Liu Y, Zhang Y, Cheng R, Liu S, Qu F, et al. . Radiomics analysis of apparent diffusion coefficient in cervical cancer: a preliminary study on histological grade evaluation . J Magn Reson Imaging 2019. ; 49: 280 – 90 . doi: 10.1002/jmri.26192 [DOI] [PubMed] [Google Scholar]
- 14. Li Z, Li H, Wang S, Dong D, Yin F, et al. . MR-based radiomics nomogram of cervical cancer in prediction of the lymph-vascular space invasion preoperatively . J Magn Reson Imaging 2019. ; 49: 1420 – 26 . doi: 10.1002/jmri.26531 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Caruso D, Polici M, Zerunian M, Pucciarelli F, Guido G, et al. . Radiomics in oncology, part 2: thoracic, genito-urinary, breast, neurological, hematologic and musculoskeletal applications . Cancers (Basel) 2021; 13(11): 2681. 10.3390/cancers13112681 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Song J, Hu Q, Ma Z, Zhao M, Chen T, et al. . Feasibility of t2wi-mri-based radiomics nomogram for predicting normal-sized pelvic lymph node metastasis in cervical cancer patients . Eur Radiol 2021. ; 31: 6938 – 48 . doi: 10.1007/s00330-021-07735-x [DOI] [PubMed] [Google Scholar]
- 17. . Herfs M, Soong TR, Delvenne P, Crum CP. Deciphering the multifactorial susceptibility of mucosal junction cells to HPV infection and related carcinogenesis . Viruses 2017. ; 9: 85 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. . Mockler D, Escobar-Hoyos LF, Akalin A, Romeiser J, Laurie Shroyer A, Shroyer KR. Keratin 17 is a prognostic biomarker in endocervical glandular neoplasia . Am J Clin Pathol 2017. ; 148: 264 – 273 [DOI] [PubMed] [Google Scholar]
- 19. Wenzel HHB, Van Kol KGG, Nijman HW, Lemmens V, Aa MA, et al. . ( n.d .). Cervical cancer with ≤5 mm depth of invasion and >7 mm horizontal spread . [DOI] [PubMed]
- 20. . Margolis B, Cagle-Colon K, Chen L, Tergas AI, Boyd L, Wright JD. Prognostic significance of lymphovascular space invasion for stage IA1 and IA2 cervical cancer . Int J Gynecol Cancer 2020. ; 30: 735 – 743 [DOI] [PubMed] [Google Scholar]
- 21. Balaya V, Guani B, Magaud L, Bonsang-Kitzis H, Ngô C, et al. . Validation of the 2018 figo classification for cervical cancer: lymphovascular space invasion should be considered in ib1 stage . Cancers (Basel) 2020; 12(12): E3554. 10.3390/cancers12123554 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Weyl A, Illac C, Lusque A, Leray H, Vaysse C, et al. . Prognostic value of lymphovascular space invasion in early-stage cervical cancer . Int J Gynecol Cancer 2020. ; 30: 1493 – 99 . doi: 10.1136/ijgc-2020-001274 [DOI] [PubMed] [Google Scholar]
- 23. Fasmer KE, Hodneland E, Dybvik JA, Wagner-Larsen K, Trovik J, et al. . Whole-volume tumor mri radiomics for prognostic modeling in endometrial cancer . J Magn Reson Imaging 2021. ; 53: 928 – 37 . doi: 10.1002/jmri.27444 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Zhang X, Tian Q, Wang L, Liu Y, Li B, et al. . Radiomics strategy for molecular subtype stratification of lower-grade glioma: detecting idh and tp53 mutations based on multimodal mri . J Magn Reson Imaging 2018. ; 48: 916 – 26 . doi: 10.1002/jmri.25960 [DOI] [PubMed] [Google Scholar]
- 25. Li P, Chen L . ( n.d .). Comparison between laparoscopic and abdominal radical hysterectomy for stage ib1 and tumor size . [DOI] [PMC free article] [PubMed]
- 26. de Bie RP, Vergers-Spooren HC, Massuger LFAG, Siebers AG, Salet-van der Pol MRJ, et al. . Patients with cervical cancer: why did screening not prevent these cases? Am J Obstet Gynecol 2011. ; 205: 64 . doi: 10.1016/j.ajog.2011.02.046 [DOI] [PubMed] [Google Scholar]
- 27. Wheeler DT, Kurman RJ . The relationship of glands to thick-wall blood vessels as a marker of invasion in endocervical adenocarcinoma . Int J Gynecol Pathol 2005. ; 24: 125 – 30 . doi: 10.1097/01.pgp.0000152025.45106.6d [DOI] [PubMed] [Google Scholar]
- 28. Gadducci A, Guerrieri ME, Cosio S . Adenocarcinoma of the uterine cervix: pathologic features, treatment options, clinical outcome and prognostic variables . Crit Rev Oncol Hematol 2019. ; 135: 103 – 14 : S1040-8428(18)30417-7 . doi: 10.1016/j.critrevonc.2019.01.006 [DOI] [PubMed] [Google Scholar]
- 29. Wang W, Jiao Y, Zhang L, Fu C, Zhu X, et al. . Multiparametric mri-based radiomics analysis: differentiation of subtypes of cervical cancer in the early stage . Acta Radiol 2021; 2021: 2841851211014188. 10.1177/02841851211014188 [DOI] [PubMed] [Google Scholar]
- 30. Wang T, Gao T, Guo H, Wang Y, Zhou X, et al. . Preoperative prediction of parametrial invasion in early-stage cervical cancer with mri-based radiomics nomogram . Eur Radiol 2020. ; 30: 3585 – 93 . doi: 10.1007/s00330-019-06655-1 [DOI] [PubMed] [Google Scholar]
- 31. Xiao M, Ma F, Li Y, Li Y, Li M, et al. . Multiparametric mri-based radiomics nomogram for predicting lymph node metastasis in early-stage cervical cancer . J Magn Reson Imaging 2020. ; 52: 885 – 96 . doi: 10.1002/jmri.27101 [DOI] [PubMed] [Google Scholar]
- 32. Fang J, Zhang B, Wang S, Jin Y, Wang F, et al. . Association of mri-derived radiomic biomarker with disease-free survival in patients with early-stage cervical cancer . Theranostics 2020. ; 10: 2284 – 92 . doi: 10.7150/thno.37429 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Karunya RJ, Tharani P, John S, Kumar RM, Das S . Role of functional magnetic resonance imaging derived parameters as imaging biomarkers and correlation with clinicopathological features in carcinoma of uterine cervix . J Clin Diagn Res 2017. ; 11: XC06 - XC11 . doi: 10.7860/JCDR/2017/29165.10426 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Dong Y, Dong RT, Zhang XM, Song QL, Yu T, et al. . Influence of menstrual status and pathological type on the apparent diffusion coefficient in cervical cancer: a primary study . Acta Radiol 2021. ; 62: 430 – 36 . doi: 10.1177/0284185120926897 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.