Access to this full-text is provided by Frontiers.
Content available from Frontiers in Surgery
This content is subject to copyright.
EDITED BY
Ulrich Ronellenfitsch,
University Hospital Halle (Saale), Germany
REVIEWED BY
Yawei Qian,
Wuhan University, China
Song Su,
Affiliated Hospital of Southwest Medical
University, China
*CORRESPONDENCE
Wenbo Meng
mengwb@lzu.edu.cn
Ping Yue
dryueping@sina. Com
†
These authors have contributed equally to this
work
SPECIALTY SECTION
This article was submitted to Surgical
Oncology, a section of the journal Frontiers in
Surgery
RECEIVED 15 September 2022
ACCEPTED 31 October 2022
PUBLISHED 06 January 2023
CITATION
Ma Y, Lin Y, Lu J, He Y, Shi Q, Liu H, Li J,
Zhang B, Zhang J, Zhang Y, Yue P, Meng W and
Li X (2023) A meta-analysis of based radiomics
for predicting lymph node metastasis in patients
with biliary tract cancers.
Front. Surg. 9:1045295.
doi: 10.3389/fsurg.2022.1045295
COPYRIGHT
© 2023 Ma, Lin, Lu, He, Shi, Liu, Li, Zhang,
Zhang, Zhang, Yue, Meng and Li. This is an
open-access article distributed under the terms
of the Creative Commons Attribution License
(CC BY). The use, distribution or reproduction in
other forums is permitted, provided the original
author(s) and the copyright owner(s) are
credited and that the original publication in this
journal is cited, in accordance with accepted
academic practice. No use, distribution or
reproduction is permitted which does not
comply with these terms.
A meta-analysis of based
radiomics for predicting lymph
node metastasis in patients with
biliary tract cancers
Yuhu Ma1†, Yanyan Lin2†, Jiyuan Lu3†, Yulong He1, Qianling Shi1,
Haoran Liu1, Jianlong Li1, Baoping Zhang1, Jinduo Zhang2,
Yong Zhang2, Ping Yue2*, Wenbo Meng2*and Xun Li2
1
The First School of Clinical Medicine, Lanzhou University, Lanzhou, China,
2
Department of General
Surgery, The First Hospital of Lanzhou University, Lanzhou, China,
3
School of Stomatology, Lanzhou
University, Lanzhou, China
Background: To assess the predictive value of radiomics for preoperative
lymph node metastasis (LMN) in patients with biliary tract cancers (BTCs).
Methods: PubMed, Embase, Web of Science, Cochrane Library databases, and
four Chinese databases [VIP, CNKI, Wanfang, and China Biomedical Literature
Database (CBM)] were searched to identify relevant studies published up to
February 10, 2022. Two authors independently screened all publications for
eligibility. We included studies that used histopathology as a gold standard
and radiomics to evaluate the diagnostic efficacy of LNM in BTCs patients.
The quality of the literature was evaluated using the Radiomics Quality Score
(RQS) and the Quality Assessment of Diagnostic Accuracy Studies 2
(QUADAS-2). The diagnostic odds ratio (DOR), sensitivity, specificity, positive
likelihood ratio (PLR), negative likelihood ratio (NLR), and area under the
receiver operating characteristic curve (AUC) were calculated to assess the
predictive validity of radiomics for lymph node status in patients with BTCs.
Spearman correlation coefficients were calculated, and Meta-regression and
subgroup analyses were performed to assess the causes of heterogeneity.
Results: Seven studies were included, with 977 patients. The pooled sensitivity,
specificity and AUC were 83% [95% confidence interval (CI): 77%, 88%], 78%
(95% CI: 71, 84) and 0.88 (95% CI: 0.85, 0.90), respectively. The substantive
heterogeneity was observed among the included studies (I
2
= 80%, 95%CI:
58,100). There was no threshold effect seen. Meta-regression showed that
tumor site contributed to the heterogeneity of specificity analysis (P< 0.05).
Imaging methods, number of patients, combined clinical factors, tumor site,
model, population, and published year all played a role in the heterogeneity
of the sensitivity analysis (P< 0.05). Subgroup analysis revealed that magnetic
resonance imaging (MRI) based radiomics had a higher pooled sensitivity
Abbreviations
BTCs: biliary tract cancers; LMN: lymph node metastasis; CBM: China Biomedical Literature Database;
RQS: Radiomics Quality Score; QUADAS-2: Quality Assessment of Diagnostic Accuracy Studies 2;
DOR: diagnostic odds ratio; PLR: positive likelihood ratio; NLR: negative likelihood ratio; AUC: area
under the receiver operating characteristic curve; CI: confidence interval; MRI: magnetic resonance
imaging; CT: computed tomography; ICC: intrahepatic cholangiocarcinoma; PET-CT: positron
emission tomography/computed tomography; PRISMA: Preferred Reporting Items for Systematic
Reviews and Meta-Analyses; TP: true positive; TN: true negative; FP: false positive; FN: and false
negative; sROC: summary receiver operating characteristic; ROI: region of interest; ML: logistic
regression; LR: machine learning
TYPE Systematic Review
PUBLISHED 06 January 2023
|
DOI 10.3389/fsurg.2022.1045295
Frontiers in Surgery 01 frontiersin.org
than contrast-computed tomography (CT), whereas the result for pooled specificity was
the opposite.
Conclusion: Our meta-analysis showed that radiomics provided a high level of
prognostic value for preoperative LMN in BTCs patients.
KEYWORDS
biliary tract cancers, lymph node metastasis, radiomics, diagnosis, meta-analysis
Introduction
Biliary tract cancers (BTCs) are malignant tumors derived
from biliary epithelial cells, including intrahepatic and
extrahepatic cholangiocarcinoma and gallbladder cancer (1–3).
The global incidence is increasing (4). Radical resection is the
only option to prolong survival in patients with BTCs (5).
Unfortunately, less than 35% of patients are suitable for early
surgery (6,7). Recurrence after curative resection remains
high, with intrahepatic cholangiocarcinoma (ICC) reaching
50%–70% (8). According to previous studies, lymph node
metastasis (LMN) is the most relevant adverse prognostic
factor after BTCs surgery (9). In addition, LMN is a relative
contraindication for liver transplantation (10). Therefore, it is
crucial to evaluate the accurate status of lymph nodes by non-
introductive means, which is key for BTCs to guide treatment
and determine prognosis.
Medical imaging methods play an important role in
assessing the status of lymph nodes in BTCs. Conventional
imaging examinations include ultrasonography, computer
tomography (CT), positron emission tomography/computed
tomography (PET-CT), and magnetic resonance imaging
(MRI) (11,12). Nowadays, preoperative assessment of lymph
node status remains difficult. Razumilava et al. Studies have
shown that CT has a sensitivity of 30%–50% in diagnosing
lymph node status (7). There is no consensus on the
evaluation of the preoperative lymph node status of BTCs by
current detection methods (13,14). A label non-invasive
detection method called radiomics has recently been utilized
to predict chemotherapy response, lymph node metastasis,
and tumor classification (15). it extracts high-dimensional
radiomics information from medical images and combines
machine learning algorithms for clinical decision-making
(16,17). A systematic review and meta-analysis by Huang
et al. found that radiomics might be an effective tool for
assessing preoperative microvascular invasion in hepatocellular
carcinoma (17). Several published studies have employed a
radiomics model to predict LNM in BTCs (18–20). Due to
differences in imaging modality, study methodology, sample
size imaging modalities, research methods, sample size and so
on, the reported diagnostic efficiency ranged from 68% to
98% in the above studies. Therefore, the performance of
radiomics for preoperative LMN identification in clinical
practice remains uncertain.
The purpose of this meta-analysis was to determine the
diagnostic efficacy of radiomics for preoperative LMN
prediction in patients with BTCs.
Materials and methods
This meta-analysis was conducted following the Preferred
Reporting Items for Systematic Reviews and Meta-Analyses
(PRISMA) statement recommended by the Cochrane
Collaboration (21). This study was prospectively registered in
PROSPERO (CRD42022333874).
Literature search
Two authors (YM and YH) independently searched
PubMed, Embase, Web of Science, Cochrane Library, and
four Chinese Databases [VIP, CNKI, Wanfang, and Chinese
BioMedical Literature Databases (CBM)] to determine the
studies published as of February 10, 2022. The search formula
was as follows: [(lymph node metastasis) OR (lymph node)
OR(LMN)] and [(Biliary Tract Cancer) OR (intrahepatic
cholangiocarcinoma) OR (extrahepatic cholangiocarcinoma)
OR (ICC) OR (ECC)] and [(radiomics) OR (machine
learning) OR (deep learning) OR (artificial intelligence) OR
(texture)]. After eliminating duplicate articles, the titles and
abstracts of all remaining articles were reviewed. When it was
ambiguous whether the article was included merely by title
and abstract, the entire publication was downloaded and
reviewed. All studies were independently screened by two
authors (YM and YH). Discuss the inclusion issues if there
were any discrepancies. In order to find other relevant
publications, we also carefully went through the reference lists
for each important study that we had already identified as
well as earlier systematic reviews.
Selection criterion
Inclusion criteria were as follows: (1) diagnosis of BTCs by
pathologic criteria; (2) determination of LMN by pathologic
diagnosis; (3) CT, MRI, PET-CT, or ultrasonography were
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 02 frontiersin.org
performed before surgical resection, liver transplantation, or
other treatments; (4) imaging analysis based on radiomics.
Exclusion criteria are as follows: (1) having received any
treatment (radiotherapy, chemotherapy, or immunotherapy)
before the examination; (2) Patients who received palliative
surgery without lymph node resection; (3) Reviews, editorials,
letters, and animal articles are excluded.
Quality assessment
The Radiomics Quality Score (RQS) and Quality
Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2)
were used by the two authors (YM and YH) to evaluate the
methodological quality and risk of bias of the chosen studies
independently, respectively (22,23).
Data extraction
The data extraction and quality evaluation of the retrieved
research are independently completed by the two authors. We
extracted data about patient characteristics, imaging methods, and
research characteristics from each selected study. Patient
characteristics included the total number of subjects, the number
of subjects with LMN and no-LMN, sensitivity, and specificity.
The number of true positive (TP), true negative (TN), false positive
(FP), and false negative (FN) was calculated according to the
number of LMN, non-LMN, sensitivity, and specificity reported in
each included study. The reference formula was as follows:
sensitivity = TP/(TP+FN), specificity = TN/(FP+TN). The studies
that provide a two-by-two contingency table or sufficient data to
reconstruct such a table are eligible for analysis. The best model
presented in the study was included in our meta-analysis when
there were two or more prediction models based on the same
cohort of patients in one study.
Statistical analysis
Statistical analysis was performed using Stata software
(version 16.0) and Review Manager software (version 5.3).
The sensitivity, specificity, positive likelihood ratio (PLR),
negative likelihood ratio (NLR), and diagnostic odds ratio
(DOR) with their corresponding 95% confidence intervals
(CI) were calculated. A summary receiver operating
characteristic (sROC) curve was plotted and the area under
the curve (AUC) was calculated to demonstrate the diagnostic
value of the joint studies (24). AUC was 0.5–0.7, 0.7–0.9, and
> 0.9, indicating low, medium, and high diagnostic power,
respectively.
We drew forest plots to show the variation among studies
and to detect heterogeneity for the pooled sensitivity and
specificity. The threshold effect resulting in heterogeneity was
assessed using the spearman correlation coefficient. If P> 0.05,
there was no threshold effect. The heterogeneity caused by the
non-threshold effect was measured by Cochrane’s Q-test and
inconsistency index I
2
. When P< 0.05, the difference was
considered significant, and I
2
≥50% was considered moderate
to high heterogeneity among studies (25). Meta-regression
and subgroup analysis were used to study the potential
sources of heterogeneity. We conducted a univariate meta-
regression analysis of some related covariates, including the
tumor site (ICC or no-ICC), combined clinical factors (yes or
no), imaging methods (MRI or CT), number of patients (≥
150 or < 150), QUADAS-2 applicability risk (no or high risk),
model (logistic regression or machine learning), population
(single or multicenter), number of radiomics features (≥300
or < 300), and published year (before 2020 or after 2020).
Additionally, a sensitivity analysis was carried out by
removing one study at a time to assess the effect of a single
study on the overall estimation. Deeks’funnel plot was used
to check publication bias (26).
Clinical utility
A Fagan plot was calculated to assess clinical utility by
indicating the post-test probability (P-post) of LNM when pretest
probabilities (P-pre, suspicion of LNM) were provided (27).
Results
Literature search
The literature search and study selection was shown in
Figure 1. The included studies were published between 2018
and 2021 (four contract-CT based on radiomics studies
(18,28,29) and three MRI based on radiomics studies (19,
30–32) were included in the meta-analysis). A total of 977
BTCs patients were included. Of those, 554 patients (56.8%)
had a pathological diagnosis of no-LMN, while 423 patients
(43.2%) had a pathological diagnosis of LMN. The baseline
characteristics of the included studies was showed in Table 1.
Study evaluation
A detailed report of the RQS project scores was shown in
Table 2. The RQS included in the study ranged from 11 to 20
points. The publication with the highest percentage of RQS
was 56.0%. Indicating excellent reproducibility across readers,
the interclass correlation coefficient (ICC) among separate
readers rating publications was 0.972 (95% CI: 0.854–0.997, P
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 03 frontiersin.org
< 0.001). The RQS scores assessed by both readers were
presented in Supplementary materials.
The methodological quality of the studies according to the
QUADAS-2 assessment was illustrated in Figure 2. Two
studies obtained an unclear risk of bias in the index testing
(28,30). Uncertain risk of bias was found in both the flow
and time domains in two studies (19,28). There were
relatively few concerns about the applicability of the three
domains (patient selection, index test, and reference criteria).
No significant publication bias risk was detected by Deeks’
funnel plot analysis (Figure 3;P= 0.77).
Diagnostic accuracy of radiomics
The pooled sensitivity and specificity were 83% (77%, 88%)
and 78% (71%, 84%), respectively, based on radiomics
assessment of lymph node status in each patient. AUC, DOR,
PLR and NLR were 0.88 (0.85, 0.90), 17.82 (11.42, 27.80),
3.80 (2.88, 5.00) and 0.21 (0.16, 0.29), respectively. Figure 4
shows the forest plots for sensitivity and specificity, while
Figure 5 shows the sROC curve.
Heterogeneity assessment
The Spearman correlation coefficient for the threshold effect
was found to be −0.67 and P= 0.45, indicating that there was no
threshold effect. There was considerable heterogeneity among the
studies (overall I² = 80%; 95%CI: 58.00,100.00; P= 0.003). The
forest plots indicated high heterogeneity with I
2
values > 50% for
sensitivity (I² = 65.90%; 95% CI: 38.41, 93.39; P=0.01) and
specificity (I² = 68.00%; 95% CI: 42.55, 93.46; P< 0.01).
Meta-regression
A univariate meta-regression analysis was used to determine
the sources of heterogeneity. The outcomes of subgroup analysis
and univariate meta-regression were displayed in Table 3. The
FIGURE 1
The PRISMA flowchart of the selection procedure.
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 04 frontiersin.org
TABLE 1 The baseline characteristics of the included 7 studies.
Study
ID
Year Tumor
type
Imaging
methods
Sequence Study
design
Population ROI Model Feature
number
Numbers
of patients
LMN No-
LMN
TP FP FN TN Combine
Clinical
Factors
(Yes/No)
Ji et al. 2019 ICC Contrast-CT Arterial
phase
retrospective Single center Tumor
parenchyma
LASSO
and LR
105 103 45 48 39 15 6 43 Yes
Yang
et al.
2020 ECC MRI T1WI,
T2WI, DWI
retrospective Single center Tumor
parenchyma
MI and
RF
300 100 73 27 58 5 15 22 No
Xu et al. 2019 ICC MRI T1WI retrospective Single center Tumor
parenchyma
mRMR
and
SVM
491 106 47 59 42 25 5 34 Yes
Liu et al. 2021 GBC Contrast-CT Venous
phase
retrospective Multicenter Tumor
parenchyma
LASSO
and LR
293 209 84 125 73 28 11 97 Yes
Ji et al. 2018 BTCs Contrast-CT Venous
phase
retrospective Single center Tumor
parenchyma
LASSO
and LR
93 177 35 35 16 7 19 28 Yes
Yao
et al.
2020 ECC MRI T1WI,
T2WI, DWI
and ADC
retrospective Single center Approximately 1–
2 mm from the
edge of the tumor
PSO and
SVM
120 110 79 31 68 6 11 25 No
Huang
et al.
2019 ICC Contrast-CT Complete
sequence
retrospective Single center NA RF and
LR
832 172 51 121 45 21 6 100 Yes
BTCs, biliary tract cancers; ECC, extrahepatic cholangiocarcinoma; ICC, intrahepatic cholangiocarcinoma; ROI: region of interest; GBC, gallbladder cancer; LASSO, least absolute shrinkage, and selection operator; SVM, support
vector machine; mRMR, minimum redundancy maximum relevance; RF, random forest; LR, logistic regression.
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 05 frontiersin.org
results showed that the sources of heterogeneity in the sensitivity
analysis, in addition to QUADAS, included number of radiomics
features, tumor site, imaging methods, number of patients,
combined clinical factors, model, population, and publication
year (P< 0.05). Additionally, tumor site contributed to the
heterogeneity in the specificity analysis (P< 0.05).
Subgroup analysis
In terms of tumor site, the sensitivity (84%; 95% CI: 76, 93
vs. 83%; 95% CI: 76, 90) and specificity (78%; 95% CI: 68, 87 vs.
78%; 95% CI: 70, 85) were basically equivalent among studies
with ICC studies (n= 3) and no-ICC studies (n= 4). In
comparison to radiomics combined with clinical risk factors,
radiomics alone had better sensitivity (86%; 95% CI: 77, 94
vs. 82%; 95% CI: 76, 89) and specificity (81%; 95% CI: 69,
94 vs. 77%; 95%; CI: 70, 85). Moreover, the pooled
sensitivity (86%; 95% CI: 81, 92 vs. 80%; 95% CI: 72, 87)
and specificity (80%; 95% CI: 70, 89 vs. 77%; 95% CI: 69,
85) of studies published after 2020 (n= 3) was relatively high
than that of earlier studies (n= 4). Regardless of the
QUADAS risk, the specificity (78%;95% CI: 63, 93 vs. 78%;
95% CI: 71, 85) was roughly same for both. Furthermore,
QUADAS high-risk studies (n=3;88%;95%CI:78,98)had
a marginally greater sensitivity than no high-risk studies (n=4;
82%; 95% CI: 77, 88).
Among different imaging methods, MRI (n= 3) had higher
sensitivity (87%; 95% CI: 81, 93 vs. 80%; 95% CI: 73, 87), but the
specificity of contract-CT (81%; 95% CI: 75, 86) better than
MRI (71%; 95% CI: 60, 82). In addition, multicenter center
studies (n= 1) exhibited greater sensitivity (88%; 95% CI: 78,
98 vs. 82%; 95% CI: 77, 88) than s single-center studies
(n= 6). The pooled sensitivity for more radiomics features
(n= 3) was higher (88%; 95% CI: 81, 94 vs. 80%; 95% CI: 74,
87), whereas the trend for pooled specificity was the opposite
(75%; 95% CI: 65, 85 vs. 80%; 95% CI: 73, 88). Similarly,
studies with 150 patients or fewer (n= 4) had a greater pooled
sensitivity (84%; 95% CI: 77, 91) but a lower specificity
(77%; 95% CI: 67, 86) than studies with more than 150
patients (n= 3; sensitivity, 82%; 95% CI: 74, 91; specificity,
79%; 95% CI: 71, 87). For modeling methods, machine
learning (n= 3) approach had exhibited better sensitivity
(87%; 95% CI: 81, 93)] than logistic regression (n= 4; 79%;
95% CI: 73, 86) and lower specificity [72%; 95% CI: 60, 85)
than logistic regression (79%; 95% CI: 71, 87).
Sensitivity analyses
No significant changes were observed when each included
study was eliminated from the analysis one by one. The
results of sensitivity analyses for each study are shown in
Table 4.
Clinical utility
Using radiomics studies would increase the posttest
probability to 49 from 20% with a PLR of 4 when the pretest
was positive and would reduce the posttest probability to 5%
with an NLR of 0.21 when the pretest was negative (Figure 6).
TABLE 2 RQS elements and the mean rating of our eligible studies.
RQS scoring
item
Interpretation Average
score
Image Protocol + 1 for well documented protocols, + 1
for publicly available protocols
2.00
Multiple
Segmentations
+ 1 if segmented multiple times
(different physicians, algorithms, or
perturbation of regions of interest)
0.64
Phantom Study + 1 if texture phantoms were used for
feature robustness assessment
0.00
Multiple Time Points + 1 multiple time points for feature
robustness assessment
0.00
Feature Reduction −3 if nothing, + 3 if either feature
reduction or correction for multiple
testing
3.00
Non Radiomics + 1 if multivariable analysis with non-
radiomics features
0.71
Biological Correlates +1 if present 0.00
Cut-off + 1 if cutoff either pre-defined or at
median or continuous risk variable
reported
0.57
Discrimination and
Resampling
+ 1 for discrimination statistic and
statistical significance, + 1 if resampling
applied
1.50
Calibration + 1 for calibration statistic and
statistical significance, + 1 if resampling
applied
1.21
Prospective + 7 for prospective validation within a
registered study
0.00
Validation −5 if no validation/+2 for internal
validation/+3 for external validation/+4
two external validation datasets or
validation of previously published
signature/+5 validation on ≥3 datasets
from >1 institute
2.14
Gold Standard + 2 for comparison to gold standard 1.42
Clinical Utility + 2 for reporting potential clinical
utility
1.14
Cost-effectiveness + 1 for cost-effectiveness analysis 0.00
Open Science + 1 for open-source scans, +1 for open-
source segmentations, + 1 for open-
source code, + 1 open-source
representative segmentations and
features
2.14
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 06 frontiersin.org
Discussion
The seven studies were included in the meta-analysis. The
diagnostic efficacy of radiomics for preoperative LMN in
BTCs patients was judged by combining diagnostic effect size
and fitting the sROC curve. Analyze the heterogeneity of the
studies and their sources, and identify factors that may affect
the results. Finally, through sensitivity analysis, publication
bias and clinical application value were detected to evaluate
the credibility of this meta-analysis.
Our meta-analysis showed high sensitivity (83%; 95%
CI: 77%, 88%), specificity (78%; 95% CI: 71, 84) and AUC
(0.88; 95% CI: 0.85, 0.90). In addition, the likelihood ratio
and post-test probability indicate that the post-test probability
increases from 20% to 49% when the current test is positive
and the PLR is 4; when the current measurement is negative
and the NLR is 0.21, the post-test probability is reduced to
5%. This further indicates that radiomics is helpful to improve
the accuracy of predicting LMN in BTCs patients. All things
considered, radiomics can assist us in possible resolution
treatment protocols for BTCs LMN patients before surgery,
increase the survival rate of BTCs patients, and decrease the
probability of recurrence.
Currently, using visual observation and interpretation of
medical images to evaluate lymph node metastasis in BTCs is
still challenging (33). Radiomics, as a personalized assessment
tool, has proved to be a promising non-invasive tumor lymph
node evaluation to overcome the limitations of visual
assessment of lymph node images by imaging physicians (34).
Our findings demonstrate that radiomics can increase to 83%
the sensitivity of preoperative BTCs lymph node metastatic
evaluation. The radiomics method, which offers crucial
supplementary information on imaging phenotypes and may
contain a wealth of data, maybe the cause. This includes
FIGURE 2
Methodological quality assessment of the included studies based on the QUADAS-2 scale. (A) Individual studies, (B) summary.
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 07 frontiersin.org
texture features that reflect the pattern or spatial distribution of
voxel intensity in the region of interest (ROI), which are
connected to tumor heterogeneity (35,36). Wavelet features
can also offer multi-frequency information to measure tumor
heterogeneity and raise diagnostic accuracy (35,37,38).
Therefore, we have reason to believe that radiomics, which is
used by professional radiologists for ROI segmentation and
high-throughput feature extraction for disease classification
and prognosis, may be more suitable for the preoperative
evaluation of LMN in patients with BTCs.
Although there was no discernible threshold effect, there
was overall significant heterogeneity between studies (I² = 80%;
95%CI: 58, 100; P= 0.003). We conducted meta-regression to
detect the sources of heterogeneity. Due to the limited
number of included studies, only univariable meta-regression
analysis was performed. The results revealed that factors such
as the tumor site, imaging methods, number of patients,
combined clinical factors, number of radiomics features,
model, population, and published year all contributed to the
heterogeneity of sensitivity analysis. In addition, the
methodologies utilized in each of the included studies varied,
contributing to the heterogeneity in a way that made it
impossible to identify all of its sources.
We used key factors for subgroup analysis. In the study
design subgroup analysis, the imaging methods results showed
that the pooled sensitivity of MRI imaging is better than that
of contract-CT, which is similar to the previous research
results (28,39,40). From the number of features and
modeling methods of the included studies, the extraction of
more radiomics features could improve the pooled sensitivity,
and the possible reason texture features could improve the
accuracy of the model. However, numerous features had an
FIGURE 5
Summary receiver operating characteristic curves (sROC) based on
radiomics for preoperative prediction of LMN in BTCs.
TABLE 3 Univariable meta-regression and subgroup analyses.
Subgroup Category No. of Studies Sensitivity (95% CI) Pvalue Specificity (95% CI) Pvalue
Tumor site ICC 3 0.84 (0.76–0.93) 0.00 0.78 (0.68–0.87) 0.03
No-ICC 4 0.83 (0.76–0.90) 0.78 (0.70–0.85)
Combine clinical factors Yes 5 0.82 (0.76–0.89) 0.00 0.77 (0.70–0.85) 0.06
No 2 0.86 (0.77–0.94) 0.81 (0.69–0.94)
Imaging methods CT 4 0.80 (0.73–0.87) 0.00 0.81 (0.75–0.86) 0.18
MRI 3 0.87 (0.81–0.93) 0.71 (0.60–0.82)
No. of participants ≥150 3 0.82 (0.74–0.91) 0.01 0.79 (0.71–0.87) 0.06
<150 4 0.84 (0.77–0.91) 0.77 (0.67–0.86)
QUADAS High risk 1 0.88 (0.78–0.98) 0.33 0.78 (0.63–0.93) 0.18
No high risk 6 0.82 (0.77–0.88) 0.78 (0.71–0.85)
Model LR 4 0.80 (0.74–0.87) 0.00 0.80 (0.73–0.88) 0.11
ML 3 0.87 (0.81–0.94) 0.75 (0.65–0.85)
Population Single center 6 0.82 (0.77–0.88) 0.02 0.78 (0.71–0.85) 0.19
Multicenter center 1 0.88 (0.78–0.98) 0.78 (0.63–0.93)
No. of features ≥300 3 0.88 (0.81–0.94) 0.00 0.75 (0.65–0.85) 0.11
<300 4 0.80 (0.74–0.87) 0.80 (0.73–0.88)
Published year After 2020 3 0.86 (0.81–0.92) 0.03 0.80 (0.70–0.89) 0.08
Before 2020 4 0.80 (0.72–0.87) 0.77 (0.69–0.85)
ICC, intrahepatic cholangiocarcinoma; MRI, magnetic resonance imaging; CT, computed tomography; QUADAS, quality assessment of diagnostic accuracy studies;
LR, logistic regression; ML, machine learning.
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 09 frontiersin.org
impact on feature selection and model robustness, so its
specificity was lower. Machine learning still needs to be
investigated in terms of feature selection and model
generalization potential because it can increase sensitivity but
decrease specificity.
Studies have demonstrated that in patients with BTCs, the
CA199 and CT lymph node status are independent predictors
of LMN (18,36). Our meta-analysis showed that combined
clinical factors did not improve the diagnostic ability of
radiomics. Therefore, combining clinical features in radiomics
to improve the diagnostic accuracy of LNM needs further
exploration. The diagnostic sensitivity of ICC was slightly
higher than no-ICC, but the overall diagnostic efficacy had no
significant difference. It is reasonable to think that more
multicenter studies in the future will increase the predictive
performance of MLN in patients with BTCs because its
pooled sensitivity was greater than that of single-center
studies. There is only one high-risk trial, however, there is no
obvious heterogeneity in the sensitivity or specificity of the
QUADAS-2 score. The ability to identify LMN in the future
study will be improved by controlling study quality to lessen
bias.
Despite radiomics’strong capacity to predict, the quality of
the included research as a whole range (RQS range from
11 points to 20 points). There is no cost-benefit analysis or
prospective design. One study only received external
validation. The QUADAS-2 data quality assessment revealed
some additional issues. For example, index testing results
from two studies showed uncertain bias risks.
Our meta-analysis that used radiomics to predict the lymph
node status of preoperative BTCs offered two advantages. First
of all, this study, which was the first meta-analysis to assess
the diagnostic efficacy of preoperative prediction of lymph
node status in BTCs patients by radiomics evaluation method,
involved seven studies and 977 BTCs patients. Second, we
used subgroup analysis to evaluate the effects of different
factors on the heterogeneity of the studies, providing a guide
for upcoming radiomics research and clinical evaluation.
There are some limitations to our study. First, there are few
qualified radiomics studies, and different medical centers use
various inspection equipment. As a result, research methods
vary from study to study, and imaging methods, ROI, feature
extraction, and modeling methods provide many options.
Second, different imaging methods, number of patients, and
tumor sites may lead to heterogeneity. Therefore, we use
regression analysis to identify the sources of heterogeneity.
Finally, while there are some uncertainties associated with the
TABLE 4 The results of sensitivity analyses for each study.
Study ID sROC Sensitivity (95% CI) Specificity (95% CI) PLR (95% CI) NLR (95% CI) DOR (95% CI)
Ji et al. 0.88 (0.85–0.90) 0.85 (0.79–0.89) 0.76 (0.69–0.82) 3.58 (2.72–4.72) 0.20 (0.11–0.28) 17.88 (10.71–29.87)
Yang et al. 0.87 (0.84–0.90) 0.83 (0.76–0.89) 0.78 (0.70–0.84) 3.74 (2.77–5.03) 0.22 (0.15–0.31) 17.24 (10.61–28.01)
Xu et al. 0.82 (0.79–0.85) 0.82 (0.76–0.87) 0.80 (0.76–0.84) 4.16 (3.40–5.10) 0.22 (0.16–0.30) 18.93 (11.91–30.09)
Liu et al. 0.87 (0.84–0.90) 0.82 (0.75–0.88) 0.78 (0.69–0.85) 3.77 (2.69–5.27) 0.23 (0.16–0.31) 16.74 (10.03–27.95)
Ji et al. 0.90 (0.87–0.92) 0.85 (0.80–0.90) 0.79 (0.70–0.86) 4.07 (2.90–5.73) 0.18 (0.14–0.24) 22.08 (14.52–33.57)
Yao et al. 0.87 (0.84–0.90) 0.83 (0.76–0.88) 0.78 (0.70–0.84) 3.74 (2.77–5.06) 0.22 (0.16–0.31) 17.02 (10.48–27.63)
Huang et al. 0.87 (0.83–0.89) 0.83 (0.76–0.88) 0.77 (0.69–0.83) 3.56 (2.65–4.77) 0.23 (0.16–0.31) 15.79 (10.11–24.67)
sROC, Summary receiver operating characteristic curves; PLR, positive likelihood ratio; NLR, negative likelihood ratio; DOR, diagnostic odds ratio.
FIGURE 6
Fagan plots for assessing the clinical utility.
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 10 frontiersin.org
QUADAS-2 evaluation, the risks of uncertainty may not
significantly affect the results and therefore the overall quality
of the study can be analyzed.
Conclusion
In conclusion, radiomics is a useful tool for predicting LMN
in patients with BTCs. Radiomics study on LMN prediction,
however, is still in its early phases. Further study on the
quality of radiomics is required in the segmentation of ROI,
method repeatability, model building, and overfitting
solutions. To demonstrate the clinical value of radiomics,
further high-quality, multicenter, large-scale prospective trials
are required.
Data availability statement
The original contributions presented in the study are
included in the article/Supplementary Material, further
inquiries can be directed to the corresponding author/s.
Author contributions
Study design: YM, YL, JL, YP, YZ, and WM. Literature
search and study selection: YM, YH, and WM. Data
extraction and quality assessment: YH, YL, QS, and BZ.
Statistical analysis: YH, HL, and JL. Study supervision: YZ,
WM, and XL. WM and YL obtained the research fund.
Editing and review of the manuscript: all authors. All authors
contributed to the article and approved the submitted version.
Funding
This work was supported by the National Natural Science
Foundation of China (Grant No. 82060551); Health industry
scientific research project of Gansu Province (Grant No.
GSWSKY2020-11 sCLU); and Lanzhou Chengguan District
Science and technology planning project (Grant No.
2019RCCX0038).
Conflict of interest
The authors declare that the research was conducted in the
absence of any commercial or financial relationships that could
be construed as a potential conflict of interest.
Publisher’s note
All claims expressed in this article are solely those of the
authors and do not necessarily represent those of their
affiliated organizations, or those of the publisher, the editors
and the reviewers. Any product that may be evaluated in this
article, or claim that may be made by its manufacturer, is not
guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found
online at: https://www.frontiersin.org/articles/10.3389/fsurg.
2022.1045295/full#supplementary-material.
References
1. Mavros MN, Economopoulos KP, Alexiou VG, Pawlik TM. Treatment and
prognosis for patients with intrahepatic cholangiocarcinoma: systematic review and
meta-analysis. JAMA Surg. (2014) 149:565–74. doi: 10.1001/jamasurg.2013.5137
2. Hundal R, Shaffer EA. Gallbladder cancer: epidemiology and outcome. Clin
Epidemiol. (2014) 6:99–109. doi: 10.2147/CLEP.S37357
3. Khan SA, Clements O, Jin UK, Joseph E. Reply to: “letter regarding [risk factors
for intrahepatic and extrahepatic cholangiocarcinoma: a systematic review and meta-
analysis]”.JHepatol. (2020) 72:95–103. doi: 10.1016/S0168-8278(20)30719-4
4. DeSantis CE, Kramer JL, Jemal A. The burden of rare cancers in the United
States. CA: a Cancer J Clin. (2017) 67:261–72. doi: 10.3322/caac.21400
5. Horgan AM, Amir E, Walter T, Knox JJ. Adjuvant therapy in the treatment of
biliary tract cancer: a systematic review and meta-analysis. J Clin Oncol. (2012)
30:1934–40. doi: 10.1200/JCO.2011.40.5381
6. Chan E, Berlin J. Biliary tract cancers: understudied and poorly understood.
J Clin Oncol. (2015) 33:1845–8. doi: 10.1200/JCO.2014.59.7591
7. Razumilava N, Gores GJ. Cholangiocarcinoma. Lancet (London, England).
(2014) 383:2168–79. doi: 10.1016/S0140-6736(13)61903-0
8. Zhang XF, Beal EW, Bagante F, Chakedis J, Weiss M, Popescu I, et al. Early
versus late recurrence of intrahepatic cholangiocarcinoma after resection with
curative intent. Br J Surg. (2018) 105:848–56. doi: 10.1002/bjs.10676
9. Mao ZY, Guo XC, Su D, Wang LJ, Zhang TT, Bai L. Prognostic factors of
cholangiocarcinoma after surgical resection: a retrospective study of 293
patients. Med Sci Moni. (2015) 21:2375–81. doi: 10.12659/MSM.893586
10. Rizvi S, Khan SA, Hallemeier CL, Kelley RK, Gores GJ. Cholangiocarcinoma
- evolving concepts and therapeutic strategies. Nat Rev Clin Oncol. (2018)
15:95–111. doi: 10.1038/nrclinonc.2017.157
11. Wang Y, Zhou CW, Zhu GQ, Li N, Qian XL, Chong HH, et al. A
multidimensional nomogram combining imaging features and clinical factors to
predict the invasiveness and metastasis of combined hepatocellular
cholangiocarcinoma. Ann Transl Med. (2021) 9:1518. doi: 10.21037/atm-21-2500
12. Zhou Y, Zhou G, Gao X, Xu C, Wang X, Xu P. Apparent diffusion coefficient
value of mass-forming intrahepatic cholangiocarcinoma: a potential imaging
biomarker for prediction of lymph node metastasis. Abdom Radiol (New York).
(2020) 45:3109–18. doi: 10.1007/s00261-020-02458-x
13. Saleh M, Virarkar M, Bura V, Valenzuela R, Javadi S, Szklaruk J, et al.
Intrahepatic cholangiocarcinoma: pathogenesis, current staging, and radiological
findings. Abdom Radiol (New York). (2020) 45:3662–80. doi: 10.1007/s00261-
020-02559-7
14. Blechacz B, Komuta M, Roskams T, Gores GJ. Clinical diagnosis and staging
of cholangiocarcinoma. Nature Reviews Gastroenterology & Hepatol. (2011)
8:512–22. doi: 10.1038/nrgastro.2011.131
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 11 frontiersin.org
15. Huang YQ, Liang CH, He L, Tian J, Liang CS, Chen X, et al. Development
and validation of a radiomics nomogram for preoperative prediction of lymph
node metastasis in colorectal cancer. J Clin Oncol. (2016) 34:2157–64. doi: 10.
1200/JCO.2015.65.9128
16. Liang W, Xu L, Yang P, Zhang L, Wan D, Huang Q, et al. Novel nomogram
for preoperative prediction of early recurrence in intrahepatic
cholangiocarcinoma. Front Oncol. (2018) 8:360. doi: 10.3389/fonc.2018.00360
17. Tang Y, Zhang T, Zhou X, Zhao Y, Xu H, Liu Y, et al. The preoperative
prognostic value of the radiomics nomogram based on CT combined with
machine learning in patients with intrahepatic cholangiocarcinoma. World
J Surg Oncol. (2021) 19:45. doi: 10.1186/s12957-021-02162-0
18. Ji GW, Zhu FP, Zhang YD, Liu XS, Wu FY, Wang K, et al. A radiomics
approach to predict lymph node metastasis and clinical outcome of intrahepatic
cholangiocarcinoma. Eur Radiol. (2019) 29:3725–35. doi: 10.1007/s00330-019-
06142-7
19. Yang C, Huang M, Li S, Chen J, Yang Y, Qin N, et al. Radiomics model of
magnetic resonance imaging for predicting pathological grading and lymph node
metastases of extrahepatic cholangiocarcinoma. Cancer Lett. (2020) 470:1–7.
doi: 10.1016/j.canlet.2019.11.036
20. Tang Y, Yang CM, Su S, Wang WJ, Fan LP, Shu J. Machine learning-based
radiomics analysis for differentiation degree and lymphatic node metastasis of
extrahepatic cholangiocarcinoma. BMC cancer. (2021) 21:1268. doi: 10.1186/
s12885-021-08947-6
21. McInnes MDF, Moher D, Thombs BD, McGrath TA, Bossuyt PM, Clifford
T, et al. Preferred reporting items for a systematic review and meta-analysis of
diagnostic test accuracy studies: the PRISMA-DTA statement. Jama. (2018)
319:388–96. doi: 10.1001/jama.2017.19163
22. Huang J, Tian W, Zhang L, Huang Q, Lin S, Ding Y, et al. Preoperative
prediction power of imaging methods for microvascular invasion in
hepatocellular carcinoma: a systemic review and meta-analysis. Front Oncol.
(2020) 10:887. doi: 10.3389/fonc.2020.00887
23. Lambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van
Timmeren J, et al. Radiomics: the bridge between medical imaging and
personalized medicine. Nat Rev Clin Oncol. (2017) 14:749–62. doi: 10.1038/
nrclinonc.2017.141
24. Moses LE, Shapiro D, Littenberg B. Combining independent studies of a
diagnostic test into a summary ROC curve: data-analytic approaches and some
additional considerations. Stat Med. (1993) 12:1293–316. doi: 10.1002/sim.
4780121403
25. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency
in meta-analyses. BMJ (Clin Res ed). (2003) 327:557–60. doi: 10.1136/bmj.327.
7414.557
26. Deeks JJ, Macaskill P, Irwig L. The performance of tests of publication bias
and other sample size effects in systematic reviews of diagnostic test accuracy was
assessed. J Clin Epidemiol. (2005) 58:882–93. doi: 10.1016/j.jclinepi.2005.01.016
27. Hellmich M, Lehmacher W. A ruler for interpreting diagnostic test results.
Methods Inf Med. (2005) 44:124–6. doi: 10.1055/s-0038-1633930
28. Ji GW, Zhang YD, Zhang H, Zhu FP, Wang K, Xia YX, et al. Biliary
tract cancer at CT: a radiomics-based model to predict lymph node
metastasis and survival outcomes. Radiol. (2019) 290:90–8. doi: 10.1148/radiol.
2018181408
29. Liu X, Liang X, Ruan L, Yan S. A clinical-radiomics nomogram for
preoperative prediction of lymph node metastasis in gallbladder cancer. Front
Oncol. (2021) 11:633852. doi: 10.3389/fonc.2021.633852
30. Xu L, Yang P, Liang W, Liu W, Wang W, Luo C, et al. A radiomics approach
based on support vector machine using MR images for preoperative lymph node
status evaluation in intrahepatic cholangiocarcinoma. Theranostics. (2019)
9:5374–85. doi: 10.7150/thno.34149
31. Yao X, Huang X, Yang C, Hu A, Zhou G, Lei J, et al. A novel approach to
assessing differentiation degree and lymph node metastasis of extrahepatic
cholangiocarcinoma: prediction using a radiomics-based particle swarm
optimization and support vector machine model. JMIR Med Inform. (2020) 8:
e23578. doi: 10.2196/23578
32. Huang C. Development and validation of a preoperative radiomics
nomogram for prediction of lymph node metastasis of intrahepatice
cholangiocarcinoma.: Shanghai, China: Naval Medical University; 2019.
33. Bridgewater J, Galle PR, Khan SA, Llovet JM, Park JW,Patel T, et al. Guidelines
for the diagnosis and management of intrahepatic cholangiocarcinoma. JHepatol.
(2014) 60:1268–89. doi: 10.1016/j.jhep.2014.01.021
34. Yu Y, He Z, Ouyang J, Tan Y, Chen Y, Gu Y, et al. Magnetic resonance
imaging radiomics predicts preoperative axillary lymph node metastasis to
support surgical decisions and is associated with tumor microenvironment in
invasive breast cancer: a machine learning, multicenter study. EBioMedicine.
(2021) 69:103460. doi: 10.1016/j.ebiom.2021.103460
35. Mayerhoefer ME, Materka A, Langs G, Häggström I, Szczypiński P, Gibbs P,
et al. Introduction to radiomics. J Nuc Med. (2020) 61:488–95. doi: 10.2967/
jnumed.118.222893
36. Wang Y, Shao J, Wang P, Chen L, Ying M, Chai S, et al. Deep learning
radiomics to predict regional lymph node staging for hilar cholangiocarcinoma.
Front Oncol. (2021) 11:721460. doi: 10.3389/fonc.2021.721460
37. Rizzo S, Botta F, Raimondi S, Origgi D, Fanciullo C, Morganti AG, et al.
Radiomics: the facts and the challenges of image analysis. Euro Radiol Exp.
(2018) 2:36. doi: 10.1186/s41747-018-0068-z
38. Lafata KJ, Wang Y, Konkel B, Yin FF, Bashir MR. Radiomics: a
primer on high-throughput image phenotyping. Abdominal Radiol (New York).
(2022) 47:2986–3002. doi: 10.1007/s00261-021-03254-x.
39. Hwang J, Kim YK, Park MJ, Lee MH, Kim SH, Lee WJ, et al. Differentiating
combined hepatocellular and cholangiocarcinoma from mass-forming
intrahepatic cholangiocarcinoma using gadoxetic acid-enhanced MRI. J Magn
Res Imag: JMRI. (2012) 36:881–9. doi: 10.1002/jmri.23728
40. Park TG, Yu YD, Park BJ, Cheon GJ, Oh SY, Kim DS, et al. Implication of
lymph node metastasis detected on 18F-FDG PET/CT for surgical planning in
patients with peripheral intrahepatic cholangiocarcinoma. Clin Nucl Med.
(2014) 39:1–7. doi: 10.1097/RLU.0b013e3182867b99
Ma et al. 10.3389/fsurg.2022.1045295
Frontiers in Surgery 12 frontiersin.org
Available via license: CC BY
Content may be subject to copyright.
