Advertisement

Radiomic features based on MRI for prediction of lymphovascular invasion in rectal cancer

  • Yu Fu
  • Xiangchun Liu
  • Qi Yang
  • Jianqing Sun
  • Yunming Xie
  • Yiying Zhang
  • Huimao ZhangEmail author
Original Article
  • 34 Downloads

Abstract

Purpose

To investigate the value of radiomics in predicting lymphovascular invasion (LVI) status of rectal cancer based on MRI.

Materials and methods

The retrospective study included 188 patients based on MRI with histologically confirmed rectal cancer and evaluated LVI status. Clinical factors and image data were collected, and radiomics features were extracted from multi-region (tumor and mesorectum) and single region (tumor), respectively, on T2WI and DWI. Spearman correlation analysis and the LASSO algorithm were used for radiomic feature extraction and selection; preliminarily selection of an optimal classifier by the results of the fivefold cross-validation performance in the six preselected specific machine learning classifier. Multi-regional and single-regional predictive models were both built and evaluated by calculating the area under the ROC curve (AUC) and corresponding accuracy, specificity, sensitivity, etc.

Results

A Ridge Classification model was constructed with 21 features (2 clinical features, 10 radiomics features from mesorectum region, and 9 radiomics features from tumor region) selected by Spearman correlation and LASSO analysis. The multi-regional model shows a good performance in the differentiation of the status of LVI in training data sets (AUC = 0.87, accuracy = 0.79). The model was further validated in the testing data sets, giving an AUC and an accuracy of 0.74 and 0.68, respectively. Furthermore, the performance of single-regional model (AUC = 0.72, accuracy = 0.67) is lower compared to the values given by the multi-regional model.

Conclusion

The radiomics model which we developed demonstrates that multi-regional radiomics features based on multiparametric MRI are useful for preoperatively predicting lymphovascular invasion in patients with rectal cancer.

Keywords

Rectal cancer Machine learning Radiomics Lymphovascular MRI 

Abbreviations

AUC

Area under the ROC curve

EMVI

Extramural venous invasion

LASSO

Least absolute shrinkage and selection operator

LVI

Lymphovascular invasion

MRF

Mesorectal fascia

ROC

Receiver-operating characteristic

Introduction

Colorectal cancer (CRC) is the third most common malignant tumor in the world [1, 2], and about one-third to 44% of CRC are occurred in the rectum [3]. The National Comprehensive Cancer Network (NCCN) Guidelines consider lymphovascular invasion (LVI), which is defined as the presence of tumor cells in the lymphatic vessels or blood vessels or both, as a significant negative factor in treatment options and prognostication in rectal cancer [4, 5]. Several investigations have revealed that patients with LVI may associate with lymph-node metastasis and benefit from adjuvant systemic therapy [6, 7, 8, 9]. Hence, it becomes increasingly important to evaluate LVI status preoperatively, so that patients with LVI might benefit from radical surgery and adjuvant treatments [4, 6, 10, 11].

Currently, the LVI status is evaluated by histopathology after resection, which provides no accuracy preoperative evaluation to a treatment option. The biopsy may provide the LVI status before surgery; however, the limited specimen fails to provide the information of the whole tumor [12, 13]. In addition to histopathology, rectal magnetic resonance imaging (MRI) is also an important means in tumor evaluation, which is noninvasively [3, 4, 14]. Previous studies on rectal MRI have demonstrated that the diagnostic performance of extramural venous invasion (EMVI) by MRI was good [15, 16]. However, the intramural blood vessels and lymphatic vessel invasion status, which are parts of the LVI, are failed to evaluate by MRI [17, 18]. Therefore, is there a way to accurately evaluate the LVI status in rectal cancer before treatment?

Radiomics is an emerging method for extracting quantitative features from medical imaging and assisting clinical decision to improve diagnostic, prognostic, and predictive accuracy [13, 19, 20, 21]. The central hypothesis that drives the development of radiomics is based on the tumor microenvironment description, which helps to assess the biological characteristics of the tumor [12, 19]. Rectal MRI is essential for pre- and post-treatment assessment of rectal cancer, as it provides anatomic structures and their relationship with the tumor with a high-spatial resolution [22, 23]. Rectal MRI-based radiomics have been used for treatment response [24], lymph-node metastasis [25], and prognostic evaluation [26]. However, MRI-based radiomics for LVI prediction remains underinvestigated in rectal cancer.

Therefore, the aim of this study was to develop a radiomics model for prediction of lymphovascular invasion in rectal cancer based on MRI.

Materials and methods

Patients

This study was approved by the ethics committee of The First Hospital of Jilin University, and the informed consent requirement was waived. We retrospectively evaluated patients with rectal cancer in our hospital between January 2016 and December 2018. Inclusion criteria were as follows (a) histologically confirmed rectal adenocarcinoma; (b) rectal MRI were performed before surgery within 2 weeks; (c) LVI were assessed by histopathology after resection. The exclusion criteria included a history of (a) preoperative chemoradiotherapy (CRT), radiotherapy, chemotherapy, or distant metastases, considering that the preoperative treatment maybe changed the LVI status; (b) poor MRI quality; (c) lack of clinic information, such as pretreatment carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9(CA19-9). Finally, we enrolled a total of 188 patients, 80 LVI +, and 108 LVI −. We randomly divide patients into training and testing cohorts in a 2:1 ratio.

Clinicopathologic data, including age, gender, the level of carcinoembryonic antigen (CEA), and carbohydrate antigen19-9 (CA 19-9), were derived from medical records. Laboratory analysis of CEA and CA19-9 was tested within 1 week before surgery. The threshold value for the CEA level was ≤ 5 ng/mL and >5 ng/mL, and the threshold value for the CA19-9 level was ≤ 39U/mL and >39U/mL, according to the normal range used in clinics.

Measurement of conventional radiology evaluation indicators: including the location of primary tumor, lesion involvement length, tumor thickness (measured on oblique axis T2WI), extramural depth of invasion (measured on oblique axis T2WI), mesorectal fascia (MRF, > 1 mm diagnostic negative, ≤ 1 mm positive), and maximum lymph-node short diameter (measured on-axis T2WI).

The data enrolled flowchart of the study is shown in Fig. 1.
Fig. 1

Flowchart showing the numbers of the included and excluded patients in the study

MRI data acquisition

The enrolled rectal MRIs were all performed on the same MR scanner (3.0T, Philips Ingenia, The Netherlands). And glycerine enema was required for rectal cleansing before scanning. To reduce abdomen motility, 20 mg of anisodamine was injected intramuscularly 30 min before MRI scanning. All patients underwent a rectal MRI protocol including sagittal, axial, oblique axial, and coronal T2-weighted images and DWI. High-resolution T2WI images were obtained using turbo spin-echo with a repetition time (TR) = 3500 ms, echo time (TE) = 100 ms, the field of view (FOV) = 180 × 180 mm, echo train length  = 29, matrix = 288 × 256, thickness = 3.0 mm, and gap = 0.3 mm. DWI images were obtained with 2 b factors (0 and 1000 s/mm2), and TR = 2800 ms, TE = 70 ms, FOV = 340 × 340 mm, matrix = 256 × 256, thickness = 4.0 mm, and gap = 1.0 mm. All MRI images were retrieved from the picture archiving and communication system (PACS) for tumor masking and radiomic feature extraction.

Tumor masking

Two radiologists (Dr. Fu and Dr. Liu with 8 and 3 years of experience in rectal cancer radiology diagnosis, respectively) who blinded to the histopathology results segmented the volumes of interest (VOIs) on high-spatial-resolution T2WI and DWI via IntelliSpace Discovery (Philips, Best, The Netherlands). The volumes of interest (VOIs) were defined as follows: (a) the volumes of the whole primary tumor and excluding the intestinal lumen, which was manually drawn on each slice based on T2WI (slightly high signal) and DWI (high signal, b value of 1000 s/mm2), which were drawn along the contour of the tumor; (b) the volume of the mesorectal region on unfat suppressed T2WI, which was between the MRF (the thin-low-signal intensity surrounding the mesorectum) and the outer edge of the tumor and rectal wall. The schematic diagram of the tumor and mesorectal region segmentation is shown in Fig. 2.
Fig. 2

Segmentation result of tumor and mesorectum on T2WI and DWI in an LVI-positive patient. Images in a 51-year-old male, LVI-positive rectal cancer. ac VOIs of primary tumor on T2WI; df VOIs of primary tumor on DWI; gi VOIs of mesorectum on T2WI

Radiomic feature extraction and selection

We have used three VOIs in radiomic feature calculation. The radiomic feature was analyzed by Philips Radiomics Tool (Philips Healthcare, China, the core feature calculation is based on pyRadiomics [27]). The extracted features are shown in Table 1.
Table 1

Extracted features by Philips Radiomics software using pyRadiomics

Indexes

Introduction

Feature number

Direct features

Including first-order statistics features, shape-based features, gray-level co-occurrence matrix features, gray-level size zone matrix features, gray-level run-length matrix features, neighbouring gray tone difference matrix features, and gray-level dependence matrix features

105

Indirect features

Calculated based on direct features, through the algorithm of square, square root, logarithm, and exponential

368

Wavelet transform features

Information about the frequency of similar SIs and describes the wavelet transform of the pixels in the ROI

720

Laplacian of Gaussian filtered features

Description of texture based on the images filtered by Laplacian of Gaussian

460

First, at the feature normalization step, we used the Min–Max scaling algorithm (Eq. 1):
$$ X_{\text{normal}} = \frac{{X - X_{\rm{min} } }}{{X_{\rm{max} } - X_{\rm{min} } }}. $$
(1)

Next, a Spearman correlation analysis of radiomic feature and the label were done. Features with the coefficient lower than absolute value 0.2 or the p value greater than 0.05 were removed accordingly because of the low correlation between these radiomic features and pathological labels.

Finally, at the dimensionality reduction step, least absolute shrinkage and selection operator (LASSO) algorithm [28] were used.

Radiomic model construction and evaluation

At the model construction and evaluation step, six linear classification algorithms were investigated, including Passive Aggressive Classifier, Perceptron, Ridge Classifier, SGD Classifier, Logistic Regression, and Linear Support Vector Classifier for training and prediction. First, in the model training stage, we used fivefold cross validation to evaluate the performance of six specific machine learning classifiers in the training cohort with ‘accuracy’ as the optimization metric, preliminarily selected the prediction model with the best prediction performance, and then evaluated the model in the training and test cohorts with the area under the ROC curve (AUC), etc.

Statistical analysis

All statistical analyses were performed using SPSS 24.0 (IBM Corp). Chi-square test was used to compare the differences in categorical variables (gender, the location of primary tumor, the level of CEA and CA19-9, and MRF status), while an independent sample t test or Mann–Whitney U test, as appropriate, was used to compare the differences in continues variables (age, lesion involvement length, tumor thickness, and extramural depth of invasion).

Receiver-operating characteristic (ROC) curves were generated to assess the diagnostic performance of the radiomic models in predicting LVI status by calculating the area under the ROC curve (AUC) and corresponding accuracy, specificity, sensitivity, and so on were calculated.

The reported statistical significance levels are all two-sided, with the statistical significance set at 0.05.

Results

Clinical characteristics

In total, 188 patients were identified and comprise the study cohort: 128 males (68%) and 60 females (32%), the age ranged from 24 to 89 years, with an average of 59.61 ± 11.75 years old. The demographic statistics characteristics of patients in the training and testing cohorts are shown in Table 1. As is shown in Table 2, there were significant statistical differences in gender and maximum lymph-node short diameter in the training cohort between the LVI positive and negative groups (P < 0.05).
Table 2

Characteristic of patients in the training and testing cohort

Characteristics

Training cohort

Testing cohort

LVI (+)

n = 53

LVI (−)

n = 72

P

LVI (+)

n = 27

LVI (−)

n = 36

P

Gender, no. (%)

  

0.429a

  

0.184a

 Male

34 (64.2)

51 (70.8)

 

16 (59.3)

27 (75.0)

 

 Female

19 (35.8)

21 (29.2)

 

11 (40.7)

9 (25.0)

 

Age, years

57.0 (51.0, 62.5)

61.5 (52.2, 69.8)

0.036c*

62.8 ± 10.5

59.4 ± 11.7

0.239b

The location of the tumor, no. (%)

  

0.154a

  

0.089a

 Upper

3 (5.7)

2 (2.8)

 

5 (18.5)

1 (2.8)

 

 Middle

31 (58.5)

32 (44.4)

 

10 (37.0)

19 (52.8)

 

 Lower

19 (35.8)

38 (52.8)

 

12 (44.4)

16 (44.4)

 

The tumor involved length (CM)

5.2 ± 1.7

5.1 ± 2.1

0.770b

5.3 ± 2.7

4.9 ± 2.1

0.568b

Tumor thickness (CM)

1.4 (1.1, 1.7)

1.3 (1.1, 1.6)

0.661c

1.2 ± 0.4

1.3 ± 0.5

0.377b

Extramural depth of invasion (MM)

5.0 (3.0, 8.0)

4.0 (1.0, 8.0)

0.260c

4.0 (3.0,7.0)

4.5 (2.5,8.0)

0.650c

Maximum lymph node short diameter (MM)

6.0 (5.0, 9.0)

5.0 (3.0, 6.0)

< 0.001c*

5.9 ± 2.8

4.6 ± 3.0

0.091b

CEA level, no (%)

  

0.382a

  

0.787a

 Normal

29 (54.7)

45 (62.5)

 

21 (77.8)

29 (80.6)

 

 Abnormal

24 (45.3)

27 (37.5)

 

6 (22.2)

7 (19.4)

 

CA19-9 level, no (%)

  

0.081a

  

0.572a

 Normal

43 (81.1)

66 (91.7)

 

25 (92.6)

35 (97.2)

 

 Abnormal

10 (18.9)

6 (8.3)

 

2 (7.4)

1 (2.8)

 

MRF, no. (%)

  

0.139a

  

0.578a

 Normal

30 (56.6)

50 (69.4)

 

21 (77.8)

30 (83.3)

 

 Abnormal

23 (43.4)

22 (30.6)

 

6 (22.2)

6 (16.7)

 

The threshold value for CEA level was 5 ng/mL and > 5 ng/mL, and the threshold value for CA 19-9 level was 39 U/mL and > 39 U/mL, according to the normal range used in clinics

LVI − lymphovascular invasion negative, LVI + lymphovascular invasion positive, CEA carcinoembryonic antigen, MRF mesorectal fascia, CA19-9 carbohydrate antigen 19-9

*P value < 0.05

aChi-square test, data are number of patients, with percentages in parentheses

bIndependent sample t test, data are mean ± SD

cMann–Whitney U test, data are median, with interquartile range in parentheses

No significant differences in LVI prevalence were found between the two cohorts (P = 0.952). Overall, 42.4% and 42.9% of cases were LVI positive in the training and testing cohorts, respectively.

Radiomic feature extraction and selection

For each VOI, a total of 1653 three-dimensional (3D)-based radiomic features were extracted. These radiomic features quantified tumor characteristics using tumor size and shape, intensity statistics, and texture. For each patient, we integrated all of the 4959 radiomic features from three VOIs together.

We extracted radiomic features from multi-region (tumor and mesorectum) and single region (tumor), respectively, to investigate whether multi-regional radiomic model could improve the predictive performance in LVI. After the Spearman correlation analysis and LASSO algorithm, 21 radiomic features were retained for constructing the multi-regional radiomic model, including 2 clinical features (the location of primary tumor and maximum lymph-node short diameter), 10 radiomic features from mesorectum region, and 9 radiomic features from tumor region. In single-regional radiomic models, 10 radiomic features were retained (Table 3).
Table 3

Feature coefficients of trained model

Feature name

Coefficient

With mesorectum

Without mesorectum

Location

− 0.550270779

− 0.650599274

Maximum lymph-node short diameter (mm)

0.810826003

0.913718234

mesorectum-T2WI-ExponentialGLCM-exponential-Imc2

− 0.407578931

Non-available

mesorectum-T2WI-ExponentialGLDM-exponential-LargeDependenceLowGrayLevelEmphasis

0.537410184

Non-available

mesorectum -T2WI-LogarithmGLSZM-logarithm-GrayLevelNonUniformityNormalized

0.854873667

Non-available

mesorectum -T2WI-LogarithmGLSZM-logarithm-SizeZoneNonUniformityNormalized

1.169559029

Non-available

mesorectum -T2WI-ShapeBased-Flatness

0.344792771

Non-available

mesorectum -T2WI-SquareFirstOrder-square-Minimum

− 0.5321271

Non-available

mesorectum -T2WI-SquareGLCM-square-Imc2

− 0.095079393

Non-available

mesorectum -T2WI-SquareRootGLCM-squareroot-ClusterShade

− 0.81680237

Non-available

mesorectum -T2WI-WaveletFirstOrder-wavelet-HLH-Median

0.688704501

Non-available

mesorectum -T2WI-WaveletNGTDM-wavelet-LHL-Strength

− 0.394190462

Non-available

tumor-DWI-WaveletGLCM-wavelet-HLL-MCC

− 0.662314866

− 1.055273156

tumor-T2WI-ShapeBased-SphericalDisproportion

0.187089993

0.297837686

tumor-T2WI-WaveletFirstOrder-wavelet-HLL-Mean

0.688141186

0.941876692

tumor-T2WI-WaveletFirstOrder-wavelet-HLL-Median

0.54697935

0.828889871

tumor-T2WI-WaveletFirstOrder-wavelet-HLL-Minimum

− 0.370569768

− 0.801286924

tumor-T2WI-WaveletFirstOrder-wavelet-LLH-Kurtosis

0.420803299

0.649334246

tumor-T2WI-WaveletGLCM-wavelet-HHH-Correlation

0.687261356

1.013091721

tumor-T2WI-WaveletGLCM-wavelet-HLL-Idn

0.257655009

Non-available

tumor-T2WI-WaveletGLRLM-wavelet-HLL-ShortRunHighGrayLevelEmphasis

0.465780302

0.336917221

Intercept

− 1.484964

− 1.190307866

Radiomic model construction and evaluation

We constructed multi-regional and single-regional radiomic models, and then compared their predictive performance. In the model training stage, we use the results of fivefold cross validation as the performance of a specific machine learning classifier. Ridge Classifier used in the feature extraction of multi-region and linear SVC used in the feature extraction of single region were found to produce the most accurate model on fivefold cross-validation data set, respectively (Fig. 3). In this linear models, the coefficients of each radiomic features are shown in Table 4.
Fig. 3

Accuracy ranking from six models from multi-region (a) and single region (b) trained by different classifiers on fivefold cross-validation data set

Table 4

Performance of multi- and single-regional radiomic models on fivefold cross validation, training, and testing data sets

Models

Fivefold cross validation

Training data sets

Testing data sets

Multi-region

Single region

Multi-region

Single region

Multi-region

Single region

AUC

0.82

0.79

0.87

0.81

0.74

0.72

Accuracy

0.78

0.74

0.79

0.76

0.68

0.67

F1

0.73

0.69

0.82

0.80

0.70

0.70

Sensitivity

0.72

0.68

0.83

0.81

0.77

0.73

Specificity

0.82

0.79

0.75

0.70

0.61

0.60

PPV

0.75

0.71

0.81

0.76

0.64

0.67

NPV

0.80

0.77

0.77

0.76

0.74

0.67

FPR

0.18

0.21

0.26

0.30

0.39

0.40

FNR

0.28

0.32

0.17

0.19

0.23

0.27

FDR

0.25

0.29

0.19

0.24

0.36

0.33

PPV positive predictive value, NPV negative predictive value, FPR false-positive rate, FNR false-negative rate, FDR false discovery rate

As is shown in Fig. 4, in the predictive multi-regional radiomic model, the mean AUC of the ROC curves of fivefold validation is 0.82; the AUC of training data sets and testing data sets is 0.87 and 0.74, respectively. In the predictive single-regional radiomic model, the mean AUC of the ROC curves of fivefold validation is 0.79; the AUC of training data sets and testing data sets is 0.81 and 0.72, respectively. For more performance index, please see Table 4.
Fig. 4

ROC curves of radiomic models: a ROC of multi-regional model in fivefold cross validation; b ROCs of multi-regional model in training and testing data sets; c ROC of single-regional model in fivefold cross validation; d ROCs of single-regional model in training and testing data sets

The AUC, accuracy, F1, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the multi-regional model is better than that of the single-regional model, except PPV in testing data sets.

Discussion

This retrospective study investigated a radiomic model based on MRI for the preoperative prediction of LVI status in patients with rectal cancer. We constructed and compared the single and multi-regional radiomic models for discriminate LVI invasion of rectal cancer patients. Our results suggest that MRI radiomic signature is significantly correlated with LVI. The multi-regional radiomic model was performed better than that of the single-regional model in preoperative prediction of LVI, with acceptable accuracy.

LVI is regarded as an important negative factor in treatment options and prognostication in rectal cancer from multiple statistically published trials. The previous study which was reported by Liu et al. found that the DCE-MRI-based radiomic features and LVI status were correlated in breast cancer, and indicated that the radiomic features were effective in predicting the LVI status of patients with invasive breast cancer before surgery [29]. However, it is still a challenge for LVI preoperative prediction in rectal cancer. In the clinic, rectal MRI was suggested to reflect EMVI in rectal cancer, which is only one part of LVI [15, 16, 17, 18]. Kim Y et al., who have investigated the visually assessed features, considered that the LVI presents when the mesorectal perivascular infiltrative signal was visible on pelvic MR imaging, and the sensitivity of MR-reported LVI status was 68.2% [30]. While the radiomic model in our study showed a better predictive performance than MRI-reported LVI status by Kim. The better performance in our research might be due to that the radiomic features which was derived from multi-regional VOIs in multiparametric MR images could provide comprehensive information on LVI status, including intramural, extramural blood vessels, and lymphatic vessels in rectal cancer.

Previous studies showed multi-regional MRI radiomics allowing for a more comprehensive characterization of the tumor heterogeneity. This may offer potential to improve the prediction performance [31, 32]. LVI, which is defined as the presence of tumor cells in the lymphatic vessels or blood vessels or both, include intramural, extramural blood vessels, and lymphatic vessels. In addition to the region of the tumor, the surrounding mesorectal tissues may also exhibit abnormal microscopic changes in the microvascular and lymphatic networks, extracellular matrix, and interstitial pressure, which cannot be ignored [33, 34]. Hence, we investigated whether multi-regional radiomics, including both tumor and mesorectum, could provide more features to discriminate LVI positive from LVI-negative lesions. When the current multi-regional radiomics signature was introduced into the prediction model, the performance improved than that of the single-regional model [34]. This suggests that the multi-regional radiomic signature could enhance the prediction of LVI in rectal cancer patients. In addition, our study used a 3 D VOI radiomic features by segmenting the tumor and mesorectum layer-by-layer. The 3D features provided more comprehensive information about lesions and improved the prediction accuracy of radiomic analysis compared with 2D features. Previous studies have shown that the 3D VOI provided more information about the heterogeneity of the whole lesion than the 2D region of interest (ROI) [35, 36].

How to select a modeling method is important for the performance of the radiomic model. Hence, a variety of machine learning methods should be used and the implementation should be fully documented [13], and then compare the performance of different algorithms. In our study, Ridge Classifier used in the features extraction of multi-region and linear SVC used in the feature extraction of single region were found to produce the most accurate model on fivefold cross-validation data set, respectively, can predict the LVI and maybe assist clinical decision-making.

There are some limitations to this study. First, the sample was divided into training and testing cohorts, but lacked of external testing validation. It likely led to overfitting. And all the enrolled MRIs were performed on the same MR scanner, which may also reduce the robustness of the prediction models. In the future, the study cohort should mixed different MRI scanners’ data sets to enhance robustness. Moreover, a multicenter study with a larger sample size and external validation is warranted. Second, the study did not evaluate T1 W and enhanced MR images, and only the VOIs of T2WI and DWI were calculated. However, in clinic, the T2WI and DWI play vital role in tumor evaluation, which have a proven high diagnostic accuracy [23, 37]. Third, LVI status was only classified as positive or negative in this study. LVI status was further categorized into four grades based on the number of lymphovascular structures invaded, according to Jass classification (expanding vs infiltrative) [17, 18]. Further study should investigate the relationship between radiomic feature scores with grades of LVI. Finally, this research was a retrospective study. Therefore, there is an inevitable selectivity bias. In the future, we will design a prospective study of radiomic data related to rectal cancer.

In conclusion, the radiomic model which we developed demonstrates that multi-regional and multiparametric radiomic features based on MRI are useful tools for preoperatively predicting lymphovascular invasion in patients with rectal cancer.

Notes

Acknowledgements

This study has received funding by Jilin Province Science and Technology Department Science and Technology Innovation Talents Cultivation Program (20180519008JH), Jilin Provincial Department of Finance (2018SCZWSZX-02, P8022740001, JLSCZD2019-062), Jilin Province Development and Reform Commission (2017C020), and Prevention and Control of Major Diseases Science and Technology Action Plan of China [ZX07-c20160036].

References

  1. 1.
    Arnold M, Sierra MS, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global patterns and trends in colorectal cancer incidence and mortality. Gut. 2017;66(4):683–91.  https://doi.org/10.1136/gutjnl-2015-310912.CrossRefPubMedGoogle Scholar
  2. 2.
    Ferlay J, Soerjomataram I, Dikshit R, Eser S, Mathers C, Rebelo M, et al. Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer. 2015;136(5):E359–86.  https://doi.org/10.1002/ijc.29210.CrossRefPubMedPubMedCentralGoogle Scholar
  3. 3.
    Glynne-Jones R, Wyrwicz L, Tiret E, Brown G, Rodel C, Cervantes A et al. Rectal cancer: esmo clinical practice guidelines for diagnosis, treatment and follow-up. Ann Oncol. 2017;28(suppl_4):iv22–iv40.  https://doi.org/10.1093/annonc/mdx224.CrossRefGoogle Scholar
  4. 4.
    Benson AB, 3rd, Venook AP, Al-Hawary MM, Cederquist L, Chen YJ, Ciombor KK et al. Rectal cancer, Version 2.2018, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw. 2018;16(7):874–901.  https://doi.org/10.6004/jnccn.2018.0061.CrossRefGoogle Scholar
  5. 5.
    Ale Ali H, Kirsch R, Razaz S, Jhaveri A, Thipphavong S, Kennedy ED, et al. Extramural venous invasion in rectal cancer: overview of imaging, histopathology, and clinical implications. Abdom Radiol (NY). 2019;44(1):1–10.  https://doi.org/10.1007/s00261-018-1673-2.CrossRefPubMedGoogle Scholar
  6. 6.
    Al-Sukhni E, Attwood K, Gabriel EM, LeVea CM, Kanehira K, Nurkin SJ. Lymphovascular and perineural invasion are associated with poor prognostic features and outcomes in colorectal cancer: a retrospective cohort study. Int J Surg. 2017;37:42–9.  https://doi.org/10.1016/j.ijsu.2016.08.528.CrossRefPubMedGoogle Scholar
  7. 7.
    Huh JW, Kim HR, Kim YJ. Lymphovascular or perineural invasion may predict lymph node metastasis in patients with T1 and T2 colorectal cancer. J Gastrointest Surg. 2010;14(7):1074–80.  https://doi.org/10.1007/s11605-010-1206-y.CrossRefPubMedGoogle Scholar
  8. 8.
    Barresi V, Reggiani Bonetti L, Vitarelli E, Di Gregorio C, Ponz de Leon M, Barresi G. Immunohistochemical assessment of lymphovascular invasion in stage I colorectal carcinoma: prognostic relevance and correlation with nodal micrometastases. Am J Surg Pathol. 2012;36(1):66–72.  https://doi.org/10.1097/pas.0b013e31822d3008.CrossRefGoogle Scholar
  9. 9.
    Cienfuegos JA, Rotellar F, Baixauli J, Beorlegui C, Sola JJ, Arbea L, et al. Impact of perineural and lymphovascular invasion on oncological outcomes in rectal cancer treated with neoadjuvant chemoradiotherapy and surgery. Ann Surg Oncol. 2015;22(3):916–23.  https://doi.org/10.1245/s10434-014-4051-5.CrossRefPubMedGoogle Scholar
  10. 10.
    Lee JH, Jang HS, Kim JG, Cho HM, Shim BY, Oh ST, et al. Lymphovascular invasion is a significant prognosticator in rectal cancer patients who receive preoperative chemoradiotherapy followed by total mesorectal excision. Ann Surg Oncol. 2012;19(4):1213–21.  https://doi.org/10.1245/s10434-011-2062-z.CrossRefPubMedGoogle Scholar
  11. 11.
    The Eighth Edition AJCC cancer staging manual. https://www.springer.com/us/book/9783319406176.
  12. 12.
    Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RG, Granton P, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441–6.  https://doi.org/10.1016/j.ejca.2011.11.036.CrossRefPubMedPubMedCentralGoogle Scholar
  13. 13.
    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(12):749–62.  https://doi.org/10.1038/nrclinonc.2017.141.CrossRefPubMedGoogle Scholar
  14. 14.
    Beets-Tan RGH, Lambregts DMJ, Maas M, Bipat S, Barbaro B, Curvo-Semedo L, et al. Magnetic resonance imaging for clinical management of rectal cancer: updated recommendations from the 2016 European Society of Gastrointestinal and Abdominal Radiology (ESGAR) consensus meeting. Eur Radiol. 2018;28(4):1465–75.  https://doi.org/10.1007/s00330-017-5026-2.CrossRefPubMedGoogle Scholar
  15. 15.
    Smith NJ, Barbachano Y, Norman AR, Swift RI, Abulafi AM, Brown G. Prognostic significance of magnetic resonance imaging-detected extramural vascular invasion in rectal cancer. Br J Surg. 2008;95(2):229–36.  https://doi.org/10.1002/bjs.5917.CrossRefPubMedGoogle Scholar
  16. 16.
    Bae JS, Kim SH, Hur BY, Chang W, Park J, Park HE, et al. Prognostic value of MRI in assessing extramural venous invasion in rectal cancer: multi-readers’ diagnostic performance. Eur Radiol. 2019;29(8):4379–88.  https://doi.org/10.1007/s00330-018-5926-9.CrossRefPubMedGoogle Scholar
  17. 17.
    Jass JR, Love SB, Northover JM. A new prognostic classification of rectal cancer. Lancet (London, England). 1987;1(8545):1303–6.  https://doi.org/10.1016/s0140-6736(87)90552-6.CrossRefGoogle Scholar
  18. 18.
    Betge J, Pollheimer MJ, Lindtner RA, Kornprat P, Schlemmer A, Rehak P, et al. Intramural and extramural vascular invasion in colorectal cancer: prognostic significance and quality of pathology reporting. Cancer. 2012;118(3):628–38.  https://doi.org/10.1002/cncr.26310.CrossRefPubMedGoogle Scholar
  19. 19.
    Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures. They are data. Radiology. 2016;278(2):563–77.  https://doi.org/10.1148/radiol.2015151169.CrossRefPubMedGoogle Scholar
  20. 20.
    Limkin EJ, Sun R, Dercle L, Zacharaki EI, Robert C, Reuze S, et al. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann Oncol. 2017;28(6):1191–206.  https://doi.org/10.1093/annonc/mdx034.CrossRefPubMedGoogle Scholar
  21. 21.
    Verma V, Simone CB, 2nd, Krishnan S, Lin SH, Yang J, Hahn SM. The rise of radiomics and implications for oncologic management. J Natl Cancer Inst. 2017;109(7).  https://doi.org/10.1093/jnci/djx055.
  22. 22.
    Kaur H, Choi H, You YN, Rauch GM, Jensen CT, Hou P, et al. MR imaging for preoperative evaluation of primary rectal cancer: practical considerations. Radiographics. 2012;32(2):389–409.  https://doi.org/10.1148/rg.322115122.CrossRefPubMedGoogle Scholar
  23. 23.
    Horvat N, Carlos Tavares Rocha C, Clemente Oliveira B, Petkovska I, Gollub MJ. MRI of rectal cancer: tumor staging, imaging techniques, and management. Radiographics. 2019;39(2):367–87.  https://doi.org/10.1148/rg.2019180114.CrossRefGoogle Scholar
  24. 24.
    Liu Z, Zhang XY, Shi YJ, Wang L, Zhu HT, Tang Z, et al. Radiomics analysis for evaluation of pathological complete response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer. Clin Cancer Res. 2017;23(23):7253–62.  https://doi.org/10.1158/1078-0432.CCR-17-1038.CrossRefPubMedGoogle Scholar
  25. 25.
    Yang L, Liu D, Fang X, Wang Z, Xing Y, Ma L, et al. Rectal cancer: can T2WI histogram of the primary tumor help predict the existence of lymph node metastasis? Eur Radiol. 2019.  https://doi.org/10.1007/s00330-019-06328-z.CrossRefPubMedPubMedCentralGoogle Scholar
  26. 26.
    Jalil O, Afaq A, Ganeshan B, Patel UB, Boone D, Endozo R, et al. Magnetic resonance based texture parameters as potential imaging biomarkers for predicting long-term survival in locally advanced rectal cancer treated by chemoradiotherapy. Colorectal Dis. 2017;19(4):349–62.  https://doi.org/10.1111/codi.13496.CrossRefPubMedGoogle Scholar
  27. 27.
    van Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational radiomics system to decode the radiographic phenotype. Can Res. 2017;77(21):e104–7.  https://doi.org/10.1158/0008-5472.can-17-0339.CrossRefGoogle Scholar
  28. 28.
    Sauerbrei W, Royston P, Binder H. Selection of important variables and determination of functional form for continuous predictors in multivariable model building. Stat Med. 2007;26(30):5512–28.  https://doi.org/10.1002/sim.3148.CrossRefPubMedGoogle Scholar
  29. 29.
    Liu Z, Feng B, Li C, Chen Y, Chen Q, Li X, et al. Preoperative prediction of lymphovascular invasion in invasive breast cancer with dynamic contrast-enhanced-MRI-based radiomics. J Magn Reson Imaging. 2019.  https://doi.org/10.1002/jmri.26688.CrossRefPubMedGoogle Scholar
  30. 30.
    Kim Y, Chung JJ, Yu JS, Cho ES, Kim JH. Preoperative evaluation of lymphovascular invasion using high-resolution pelvic magnetic resonance in patients with rectal cancer: a 2-year follow-up study. J Comput Assist Tomogr. 2013;37(4):583–8.  https://doi.org/10.1097/RCT.0b013e31828d616a.CrossRefPubMedGoogle Scholar
  31. 31.
    Li ZC, Bai H, Sun Q, Li Q, Liu L, Zou Y, et al. Multiregional radiomics features from multiparametric MRI for prediction of MGMT methylation status in glioblastoma multiforme: a multicentre study. Eur Radiol. 2018;28(9):3640–50.  https://doi.org/10.1007/s00330-017-5302-1.CrossRefPubMedGoogle Scholar
  32. 32.
    Kickingereder P, Bonekamp D, Nowosielski M, Kratz A, Sill M, Burth S, et al. Radiogenomics of glioblastoma: machine learning-based classification of molecular characteristics by using multiparametric and multiregional MR imaging features. Radiology. 2016;281(3):907–18.  https://doi.org/10.1148/radiol.2016161382.CrossRefPubMedGoogle Scholar
  33. 33.
    Zhou Z, Lu ZR. Molecular imaging of the tumor microenvironment. Adv Drug Deliv Rev. 2017;113:24–48.  https://doi.org/10.1016/j.addr.2016.07.012.CrossRefPubMedGoogle Scholar
  34. 34.
    Chen LD, Liang JY, Wu H, Wang Z, Li SR, Li W, et al. Multiparametric radiomics improve prediction of lymph node metastasis of rectal cancer compared with conventional radiomics. Life Sci. 2018;208:55–63.  https://doi.org/10.1016/j.lfs.2018.07.007.CrossRefPubMedGoogle Scholar
  35. 35.
    Ng F, Kozarski R, Ganeshan B, Goh V. Assessment of tumor heterogeneity by CT texture analysis: can the largest cross-sectional area be used as an alternative to whole tumor analysis? Eur J Radiol. 2013;82(2):342–8.  https://doi.org/10.1016/j.ejrad.2012.10.023.CrossRefPubMedGoogle Scholar
  36. 36.
    Blazic IM, Lilic GB, Gajic MM. Quantitative assessment of rectal cancer response to neoadjuvant combined chemotherapy and radiation therapy: comparison of three methods of positioning region of interest for ADC measurements at diffusion-weighted MR imaging. Radiology. 2017;282(2):418–28.  https://doi.org/10.1148/radiol.2016151908.CrossRefPubMedGoogle Scholar
  37. 37.
    Li J, Liu H, Yin J, Liu S, Hu J, Du F, et al. Wait-and-see or radical surgery for rectal cancer patients with a clinical complete response after neoadjuvant chemoradiotherapy: a cohort study. Oncotarget. 2015;6(39):42354–61.  https://doi.org/10.18632/oncotarget.6093.CrossRefPubMedPubMedCentralGoogle Scholar

Copyright information

© Springer Nature Singapore Pte Ltd. 2019

Authors and Affiliations

  • Yu Fu
    • 1
  • Xiangchun Liu
    • 1
  • Qi Yang
    • 1
  • Jianqing Sun
    • 2
  • Yunming Xie
    • 1
  • Yiying Zhang
    • 1
  • Huimao Zhang
    • 1
    Email author
  1. 1.Department of RadiologyThe First Hospital of Jilin UniversityChangchunChina
  2. 2.Clinical Science TeamPhilips (China) Investment Co., Ltd.ShanghaiChina

Personalised recommendations