Prediction of lymph node metastasis in rectal cancer: comparison between shear-wave elastography based ultrasomics and MRI.
According to ESGAR guidelines (European Society of Gastrointestinal and Abdominal Radiology) 2016 recommendations, magnetic resonance imaging (MRI) is considered the gold standard for rectal cancer staging (3,5-7). However, MRI is not perfect for determination of LNs status, and the criteria used to indicate LN metastasis may vary in different institutions (8). A meta-analysis found MRI to be 77% sensitive (95% CI, 69%-84%) and 71% specific (95% CI, 59%-81 %) for detection of LN involvement (9). In addition, endoscopic ultrasonography (EUS) is also a widely used method for patients with rectal cancer, but the accuracy of EUS to identify LN involvement is inferior to that of both computed tomography (CT) and MRI because of a lack of visualization of the entire mesorectum and the difficulty in accurately distinguishing benign from malignant nodes based only on the shape, echo feature and size criterion (10-13). Given these limitations, neither MRI nor EUS is an ideal method for diagnosing LN status in rectal cancer.
Radiomics, extracted from CT, MRI, or positron emission tomography images, uses a set of quantitative features to describe the geometrical structure, intensity distribution and texture of a region of interest (ROI). These features include shape, edge, and texture metrics, which can provide important insights into the tumor phenotype and the interaction of the tumor with its microenvironment (14,15). Similarly, we have applied the concept to computing quantitative ultrasound imaging, a term defined as "ultrasomics" (16). Shear-wave elastography (SWE), an ultrasound elastography technique that provides a real-time two-dimensional (2D) quantifiable image of tissue stiffness (17), has emerged as an efficient tool in detection of malignancies. The report of Wang et al. (18) confirmed that the deep learning of elastography showed a better prediction of liver fibrosis staging compared with transient elastography, 2D-SWE, and serological examinations. Therefore, SWE-based ultrasomics has a promising future in staging and prediction.
To our knowledge, there has been no study that combines SWE and ultrasomics to predict the LN metastasis. In order to build a robust model, we hypothesized that SWE ultrasomics could be a better option in LN status prediction of colorectal cancer patients. Therefore, the purpose of this study was to evaluate the accuracy of LN metastasis identification before surgical resection using SWE ultrasomics of primary tumor and compare it with MRI.
This prospective, single-center study (clinical trial ChiCTR-DDD-16008940) was approved by the Institutional Review Board of our hospital, and has obtained the informed consent from all individual participants included in the study. From August 2017 to August 2018, we enrolled consecutive patients with rectal cancer in our institution. In total, 87 patients with rectal cancer were eventually included in the study according to the inclusion and exclusion criteria. Patients meeting the following criteria were included: a) tumor located less than 15 cm from the anal verge, within the scan feasibility; b) patients deemed operable based on the determination of a multidisciplinary team discussion, and surgical resection performed within 2 weeks after EUS and MRI examinations; c) LNs dissected intraoperatively with pathological confirmation; d) tumor confirmed as rectal adenocarcinoma on histopathology; and e) SWE and MRI both examined. Exclusion criteria were as follows: a) history of chemotherapy or surgical treatment for rectal tumor; b) no pathological assessment of LN metastasis; or c) no MRI. Fig. 1 shows the flowchart. Clinical and laboratory parameters of the study population were recorded.
SWE image acquisition
The EUS and SWE image acquisitions were obtained by a radiologist with at least 5 years of experience in ultrasound examinations and 3 years of experience in SWE scans. The radiologist knew that the patients had rectal lesions but was blinded to the results of the biopsy and any imaging or endoscopy results. Sodium phosphate was used to give an enema to all patients without sedation, 2 hours prior to the inspection. The EUS and SWE images were acquired by using an SE12-3 intra-luminal transrectal high-frequency (8 MHz) probe (SuperSonic Imagine) and the Aixplorer US diagnostic imaging system (SuperSonic Imagine). Before the insertion of the probe, 200-500 mL of 0.9% NaCI was injected into the rectum to fully distend. The settings were adjusted to specified parameters at each examination, including the gain, depth, focus, and time gain compensation. After EUS scanning, the supersonic shear-wave image was superimposed on the B-mode image to obtain a fan-shaped region-of-interest, indicating the beginning of SWE acquisition. Tissue stiffness in the region of interest was represented by a color map ranging from blue (soft tissue) to red (hard tissue). Settings were adjusted to penetration, with the maximum elasticity scale at 90-100 kPa. The transducer was placed close to the tumor without pression, and the image was frozen when signal of the color map was stabilized.
Acquisition of SWE ultrasomics features
The image that displayed the deepest section of tumor infiltration into the rectal wall was selected to place the color map of the SWE. The SWE ultrasomics features of rectal tumor were extracted using AK software (Artificial Intelligence Kit, version X, GE Healthcare). The AK software is an artificial intelligence software for extracting ultrasomics or radiomics features. After importing one SWE image into AK software, the radiologist who was unaware of the final diagnosis, delineated a circular ROI covering the rectal tumor, and ran the program of the software. Then, the software automatically extracted the SWE ultrasomics features from the ROI of the rectal tumor on the SWE image.
A total of 1044 features were finally extracted from each single image of each patient. All features were grouped under the following categories: histogram parameters, grey level co-occurrence matrix (GLCM) parameters, grey level run-length matrix (GLRLM) parameters and texture parameters, which can reflect the main features of the focus of medical images comprehensively.
Intraobserver and interobserver reproducibility of feature extraction
For testing the intraobserver reproducibility, ROIs of 30 patients randomly selected from the training set were delineated by the radiologist no. 1 with at least 5 years of experience who performed the SWE ultrasomics extraction 1 month later. Moreover, to assess the inter-observer reproducibility of SWE ultrasomics feature extraction, ROIs of another 30 patients randomly selected from the training set were delineated by two radiologists (no. 1 and no. 2 with over 5 years of experience) one month later.
The rectal MRI examination was performed using a 3TMRI scanner (MagnetomVerio; Siemens Healthcare) equipped with an 8-channel body-matrix coil. All patients were routinely injected intramuscularly with 20 mg of anisodamine to minimize intestinal peristaltic movement. In order to display the boundary of the tumor more clearly, 30-120 mL of ultrasound gel was injected into the rectum before the initial MRI scan. Gadolinium (gadopentetate dimeglumine, Kangcheng) was injected intravenously using a power injector (Medrad) at 0.2 mmol/kg of bodyweight at a rate of 3.0 mL/s. Conventional rectal MRI included axial, sagittal and coronal T2-weighted turbo spin-echo (T2-TSE). It also included T1-TSE and T2-TSE perpendicular to the long axis of the rectum covering the whole tumor. A prototype diffusion-weighted imaging (DWI) sequence was performed prior to gadolinium injection to acquire the intravoxel incoherent motion data. A total of 14 b values (0, 5, 10, 20, 30, 40, 60, 80, 100, 150,200,400,600, a nd 1000 s/m[m.sup.2]) were a pplied using a single-shot spin-echo echo-planar-imaging sequence.
An abdominal MRI radiologist with more than ten years of experience reassessed the following features of the LNs: shape, signal heterogeneity, and diameter of the largest LN on T2-weighted imaging, whether a high-signal-intensity LN was detected on DWI, and whether there was an enhancement pattern in the arterial phase. Finally, the malignancy of the LNs was estimated. LNs on high-resolution MRI meeting any of the following criteria were diagnosed as metastasis: a) node with an irregular shape and a heterogeneous signal on T2-weighted imaging, b) node with high signal intensity on DWI, and c) node with a high percent enhancement in the arterial phase (5,19-21).
All the enrolled patients underwent surgical resection of the tumor within 2 weeks after EUS, SWE, and MRI examinations. The tissue sections were assessed by a pathologist with more than 10 years of experience in gastroenterological diseases who was blinded to any imaging or endoscopy results. Referring to the guidelines of the American Joint Committee on Cancer, the pathologist determined the depth of tumor infiltration and LN metastasis (22).
Statistical analyses were performed by using R version 3.3.3. (http://www.r-project.org/) and MedCalc (10.4.7.0). All statistical tests were two-sided, and p values of less than 0.05 were considered statistically significant.
Categorical variables were compared with the chi-square test. Continuous variables were compared with t test. Least absolute shrinkage and selection operator (LASSO) was used for data dimension reduction, feature selection, and ultrasomics signature building. This method minimized the residual sum of squares, subject to the sum of the absolute value of the coefficients being less than a tuning parameter. As the tuning parameter gets smaller, this may cause some coefficients to be shrunk towards zero or set to be zero. LASSO was first used to select the most useful predictive features from all features. Then, a regression equation was established based on the regression coefficients and parameters of LASSO. A ultrasomics signature was calculated for each patient via the regression equation. A receiver operating characteristic curve (ROC) analysis was performed, and the area under the receiver operating characteristic curve (AUC) analysis was used to evaluate the diagnostic performance. An optimal cutoff value that maximized the sum of the sensitivity and specificity was used to calculate sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). In addition, the comparison between AUC curves was made by the DeLong test using Medcalc statistical software, which was used to compare the difference in diagnostic performance between SWE ultrasomics and MRI.
The reproducibility of the intra-and interobserver was assessed by the intraclass correlation coefficient (ICC). An ICC [greater than or equal to]0.75 suggested high consistency, 0.50-0.74 suggested moderate ate consistency, and <0.50 suggested low consistency. The p values of less than 0.05 indicate differences between ICC and zero.
All 87 patients (male 49, female 38) were confirmed by pathology, including 40 (46.0%) with LN metastasis and 47 (54.0%) without LN metastasis (Table 1). The average age of the patients with LN metastasis was 58 years (range, 29-77 years), including 21 males (52.5%) and 19 females (47.5%). The average age of the patients without LN metastasis was 60 years (range, 38-78 years), including 28 males (59.5%) and 19 females (40.4%). There were no significant differences between patients with and without LN metastasis in terms of gender, age, tumor size, tumor location, serological examination (CEA and CA19-9 level), fecal occult blood test (hemoglobin in the stool, transferrin in the stool) and stool routine tests (white blood cells and red blood cells in the stool) (all p > 0.05). Of the patients with LN metastasis, only one patient was T1/2, and 39 patients were T3/4 (Fig. 2). Of the patients without LN metastasis, 22 patients were T1/2, and 25 patients were T3/4 (Fig. 3). A significant difference of T stage between the two groups of patients (p < 0.001) was found. LN metastasis was more common in T3/4 patients than in T1/2 patients.
The intraobserver reproducibility of the SWE ultrasomics features extracted from the ROI was high, with ICCs ranging from 0.717 (p = 0.235) to 0.986 (p = 0.000). The interobserver reproducibility of the features extracted from the ROI was also moderate to high, with ICCs ranging from 0.663 (p = 0.548) to 0.958 (p = 0.000). The features with low reproducibility that had intra- or interobserver ICC of <0.75 were excluded.
On the basis of the LASSO regression model, 1044 features of each patient were reduced to 13 selected predictors (Table 2). Among them, 2 features in histogram parameters, 2 features in GLCM parameters, 1 feature in GLCM parameters and 8 features in GLRLM parameters were extracted. Then, the calculation of SWE ultrasomics signature for each patient was performed using the following constructed formula based on the LASSO regression coefficients and 13 selected predictors:
SWE ultrasomics signature=4.980560 +0.000002*RelativeDeviation-0.003763 * Percentile90+23.82352*InverseDif-ferenceMoment_AllDirection_offset3_SD-0.525355*GLCMEntropy_angle45 _offset5-0.001977*Inertia_AllDirection_ offset7_SD-0.003099* ShortRunHighGrey LevelEmphasis_AllDirection_offset1_ SD+0.013506*RunLengthNonuniformi-ty_AllDirection_offset3_SD+1571.633* ShortRunLowGreyLevelEmphasis_AllDi-rection_offset5_SD+497.2829*LongRun LowGreyLevelEmphasis_AllDirection_offset5_SD+837.4631*ShortRunLow-GreyLevelEmphasis_AllDirection_off-set7_SD+473.2587*ShortRunLow-GreyLevelEmphasis_AllDirection_off-set8_SD+21.41898*ShortRunLowGrey-LevelEmphasis_AllDirection_offset9_SD-0.000003*HighIntensityLargeAreaEm-phasis.
The SWE ultrasomics signature correlated with LN metastasis for patients with rectal cancer (p < 0.001). Patients with LN metastasis had higher signature than did patients without LN metastasis (Fig. 4). The average score of a positive LN was 0.313[+ or -]0.498 (-0.567, 1.884), while that of a negative LN was -0.634[+ or -]0.719 (-2.848, 0.774).
Of the 47 patients without LN metastasis, 37 (78.7%) were diagnosed accurately by SWE ultrasomics and 35 (74.5%) by MRI. Of the 40 patients with LN metastasis, 35 (87.5%) were identified accurately by SWE ultrasomics, while only 31 (77.5%) were identified by MRI. Fifteen (17.2%) patients and 21 (24.1%) patients were misdiagnosed by SWE ultrasomics and MRI, respectively (Table 3).
On MRI, 43 patients were diagnosed with LN metastasis. All of these patients had LNs greater than 5 mm in diameter. There were 23 patients with LNs with irregular shapes, and 43 patients had ill-defined borders and heterogeneous signals on T2-weighted imaging (Fig. 3). Furthermore, 37 patients had nodes with high signal intensity on DWI, and 33 patients had LNs with a higher percent enhancement in the arterial phase.
Among the 21 patients misdiagnosed by MRI, 17 patients were misdiagnosed by MRI but correctly diagnosed by SWE ultrasomics. Among the 17 patients misdiagnosed by MRI, 9 patients had LN metastasis on pathology, one appeared to have a regular shape (Fig. 2), 3 appeared to have well-defined contours, and 5 appeared to have homogeneous signal onT2-weighted imaging. In the group of 8 patients with no LN metastasis on pathology that were misdiagnosed by MRI, 2 appeared to have ill-defined LN borders, 3 appeared to have heterogeneous signals, one appeared to have an irregularly shaped LN with heterogeneous signal (Fig. 3), and one appeared to have an irregularly shaped LN, with ill-defined border and heterogeneous signal on T2-weighted imaging. Another patient appeared to have a high signal intensity LN on DWI.
For the discrimination of LN metastasis, the diagnostic performance of SWE ultrasomics signature was significantly higher than that of MRI (AUC, 0.883 vs. 0.760, p = 0.034) (Fig. 5). The diagnostic accuracy, sensitivity, specificity, PPV and NPV of SWE ultrasomics signature were 82.8%, 87.5%, 78.8%, 77.8%, and 88.1%, respectively, which are higher than those of MRI. The accuracy, sensitivity, specificity, PPV, and NPV of MRI were 75.9%, 77.5%, 74.5%, 72.1%, and 79.6%, respectively (Table 4).
In the present study, we performed SWE ultrasomics to predict LN metastasis in patients with rectal cancer before individualized therapy and compared it with MRI.
LN metastasis is one of the key indications for preoperative neoadjuvant chemotherapy and also an important prognostic factor for local recurrence (23). However, the lack of high sensitivity and specificity of imaging modalities for nodal staging might lead to many patients being overtreated or having delayed treatment with chemoradiotherapy. Previously, the widely accepted diagnostic criterion for malignant nodes was the size of the LN on MRI images (24). The size threshold for LN positivity has been a source of controversy in previous studies. Some studies used the size threshold of >5 mm to indicate positive LNs (25). While for others, a size threshold of >5 mm in short axis with heterogeneous signal or ill-defined margins were used as the diagnostic criteria for positive LNs (26). However, approximately 41.4% of LNs <5 mm are positive for metastasis (27). Other imaging parameters in MRI, such as the border, shape or intrinsic signal of LNs, were analyzed in various studies, but the results have been inconsistent (19, 27, 28). Moreover, these features are subjective, due to a lack of objective quantitative indicators. Different physicians may have different opinions about the same LN, which may affect the accuracy of diagnosis.
Another important diagnostic tool for accurate evaluation of rectal tumors is EUS. In one report, the sensitivity and specificity of EUS in diagnosing nodal involvement were approximately 73.2% and 75.8%, respectively (12). In fact, the determination of LN involvement with EUS is less accurate because of the difficulty in discriminating between inflammatory and metastatic LNs, which leads to misdiagnosis and possible overtreatment. Moreover, the inspection scope of EUS is limited, which precludes full exploration of the pelvic cavity or iliac fossa LNs. These limitations weaken the usefulness of EUS for LN staging.
With the development of radiomics in recent years, these limitations may be solved. Radiomics-based tools have been developed to improve diagnostic, prognostic, and predictive accuracy mainly in cancer disease. Wu et al. (29) developed an MRI-based radiomics signature of bladder tumor for the preoperative prediction of LN metastasis in patients. Similarly, Huang et al. (30) developed a radiomics model of primary tumor to predict LN status in colorectal cancer. Thus, the noninvasive radiomics signature of primary tumor, which makes use of the images we already obtained, could serve as a more convenient method for the prediction of LN metastasis. Recently, Chen et al. (31) developed a multiparametric ultrasomics nomogram to improve pre-therapeutic individualized prediction of LN metastasis. The nomogram incorporated EUS, CT and SWE features, with a C-index of 0.857-0.872. However, the study did not compare their diagnostic efficiency with other imaging methods, such as MRI.
In our study, ultrasomics features were extracted from the color map of the SWE containing the information of tissue stiffness. Red and darker images signify higher stiffness of the lesions, while blue images represent soft tissue. Therefore, we believe that SWE ultrasomics data extracted from the color map of the SWE, which includes characteristics of lesion stiffness, distribution of stiffness, as well as the uniformity and heterogeneity of stiffness, may reflect the biological characteristics of the tumor more comprehensively, rather than the stiffness value measured by the SWE equipment.
Considering the above findings, we proposed a model of SWE-based ultrasomics for the preoperative prediction of LN metastasis in patients with rectal cancer. As indicated in our results, we found that the diagnostic performance of SWE ultrasomics for the detection of LN metastasis in patients with rectal cancer was significantly higher compared to that of MRI. Our results also demonstrated that among the 21 patients misdiagnosed by MRI, 17 patients were correctly identified by SWE ultrasomics. The high accuracy of SWE ultrasomics benefited from the application of numerous quantitative ultrasomics parameters rather than the use of subjective evaluation (e.g., signal, boundary). Therefore, the overall diagnostic efficiency of SWE ultrasomics was higher than that of MRI.
In this study, the selected 13 high-throughput ultrasomics parameters, including histogram parameters, texture parameters, GLCM parameters and GLRLM parameters, are crucial for the evaluation of LN metastasis in rectal cancer. These parameters may fully reflect the biological characteristics of tumors. For example, histogram parameters have been applied to describe the distribution of voxel intensities of the lesion. Textural features calculated from GLCM and GLRLM can quantify intratumor heterogeneity.
Our study limitations included the relatively small sample size; thus, a future multicenter study with a larger sample size is mandatory. In addition, no subgroup analysis, such as the SWE ultrasomics difference between T1/2 and T3/4 tumors, was performed. Finally, EUS is an endoluminal examination. Considering the tolerance and compliance of patients, we did not perform intra- and inter-radiologist reliability on multiple examinations. Beyond that, the comparison between SWE-based ultrasomics and MRI radiomics was not performed in our study. It has been reported that the preoperative radiomic signature based on multiparametric MRI has been used for LN status prediction in rectal cancer (32). The MRI-based radiomics has AUC, accuracy, sensitivity, and specificity of 0.677, 0.610, 0.762, and 0.494, respectively. Therefore, the comparison between SWE based ultrasomics and MRI radiomics will be performed in an upcoming study.
In conclusion, this study demonstrated that SWE ultrasomics is a more accurate predictive tool than MRI for identifying LN metastasis preoperatively. Thus, SWE ultrasomics might be conveniently used to guide preoperative individualized therapies for patients with rectal cancer.
This work was supported by the National Nature Science Foundation of China [grant numbers 81701719, 81701701, 81601500]; Guangdong Finance Foundation [grant number 20160904]; and Training Project for Young Teacher of Sun Yat-sen University [grant number 16YKPY37].
Conflict of interest disclosure
The authors declared no conflicts of interest.
(1.) Brenner H, Kloor M, Pox CP. Colorectal cancer. Lancet 2014; 383:1490-1502.
(2.) Benson AB, 3rd, Venook AP, Al-Hawary MM, et al. Rectal Cancer, version 2.2018, NCCN clinical practice guidelines in oncology. J Natl Compr Canc Netw 2018; 16:874-901.
(3.) Smith N, Brown G. Preoperative staging of rectal cancer. Acta Oncol 2008;47:20-31.
(4.) Chang GJ, Rodriguez-Bigas MA, Skibber JM, Moyer VA. Lymph node evaluation and survival after curative resection of colon cancer: Systematic review. J Natl Cancer Inst 2007; 99:433-441.
(5.) Tudyka V, Blomqvist L, Beets-Tan RG, et al. Eurecca consensus conference highlights about colon & rectal cancer multidisciplinary management: The radiology experts review. Eur J Surg Oncol 2014;40:469-475.
(6.) Engstrom PF, Arnoletti JP, Benson AB, 3rd, et al. NCCN clinical practice guidelines in oncology: Rectal cancer. J Natl Compr Cane Netw 2009; 7:838-881.
(7.) Mortenson MM, Khatri VP, Bennett JJ, Petrelli NJ. Total mesorectal excision and pelvic node dissection for rectal cancer: An appraisal. Surg Oncol Clin North Am 2007; 16:177-197.
(8.) Moreno CC, Sullivan PS, Mittal PK. MRI evaluation of rectal cancer: Staging and restaging. Curr Probl Diagn Radiol 2017; 46:234-241.
(9.) Al-Sukhni E, Milot L, Fruitman M, et al. Diagnostic accuracy of MRI for assessment of T category, lymph node metastases, and circumferential resection margin involvement in patients with rectal cancer: A systematic review and meta-analysis. Ann Surg Oncol 2012; 19:2212-2223.
(10.) Cartana ET, Parvu D, Saftoiu A. Endoscopic ultrasound: Current role and future perspectives in managing rectal cancer patients. J Gastrointestin Liver Dis 2011; 20:407-413.
(11.) Cosgrove D, Piscaglia F, Bamber J, et al. Efsumb guidelines and recommendations on the clinical use of ultrasound elastography. Part 2: Clinical applications. Ultraschall Med 2013; 34:238-253.
(12.) Puli SR, Reddy JB, Bechtold ML, Choudhary A, Antillon MR, Brugge WR. Accuracy of endoscopic ultrasound to diagnose nodal invasion by rectal cancers: A meta-analysis and systematic review. Ann Surg Oncol 2009; 16:1255-1265.
(13.) Herzog U, von Flue M, Tondelli P, Schuppisser JP. How accurate is endorectal ultrasound in the preoperative staging of rectal cancer? Dis Colon Rectum 1993;36:127-134.
(14.) Kumar V, Gu Y, Basu S, et al. Radiomics: The process and the challenges. Magn Reson Imaging 2012; 30:1234-1248.
(15.) Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: The bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 2017; 14:749-762.
(16.) Li W, Huang Y, Zhuang BW, et al. Multiparametric ultrasomics of significant liver fibrosis: A machine learning-based analysis Eur Radiol 2018.
(17.) Bamber J, Cosgrove D, Dietrich CF, et al. Efsumb guidelines and recommendations on the clinical use of ultrasound elastography. Part 1: Basic principles and technology. Ultraschall Med 2013;34:169-184.
(18.) Wang K, Lu X, Zhou H, et al. Deep learning radiomics of shear wave elastography significantly improved diagnostic performance for assessing liver fibrosis in chronic hepatitis B: A prospective multicentre study. Gut 2019; 68:729-741.
(19.) Brown G, Richards CJ, Bourne MW, et al. Morphologic predictors of lymph node status in rectal cancer with use of high-spatial-resolution MR imaging with histopathologic comparison. Radiology 2003; 227:371 -377.
(20.) Liu L, Liu M, Yang Z, He W, Wang Z, Jin E. Correlation of MRI-detected extramural vascular invasion with regional lymph node metastasis in rectal cancer. Clin Imaging 2016; 40:456-460.
(21.) Tang Y, Rao S, Yang C, Hu Y, Sheng R, Zeng M. Value of MRI morphologic features with ptl -2 rectal cancer in determining lymph node metastasis. J Surg Oncol 2018; 118:544-550.
(22.) Edge SB, Compton CC. The american joint committee on cancer: The 7th edition of the ajcc cancer staging manual and the future of tnm. Ann Surg Oncol 2010; 17:1471 -1474.
(23.) Hermanek P, Merkel S, Fietkau R, Rodel C, Hohenberger W. Regional lymph node metastasis and locoregional recurrence of rectal carcinoma in the era of tme [corrected] surgery. Implications for treatment decisions. Int J Colorectal Dis 2010; 25:359-368.
(24.) Ogawa S, Itabashi M, Hirosawa T, Hashimoto T, Bamba Y, Kameoka S. Lateral pelvic lymph node dissection can be omitted in lower rectal cancer in which the longest lateral pelvic and perirectal lymph node is less than 5 mm on MRI. J Surg Oncol 2014; 109:227-233.
(25.) lannicelli E, Di Renzo S, Ferri M, et al. Accuracy of high-resolution MRI with lumen distention in rectal cancer staging and circumferential margin involvement prediction. Korean J Radiol 2014; 15.
(26.) Park JS JY-J, Choi G-S, et al. Accuracy of preoperative MRI in predicting pathology stage in rectal cancers: Node-for-node matched histopathology validation of MRI features. Dis Col Rectum 2014; 57:32-38.
(27.) Liu Y, Wang R, Ding Y, et al. A predictive nomogram improved diagnostic accuracy and interobserver agreement of perirectal lymph nodes metastases in rectal cancer. Oncotarget 2016; 7:14755-14764.
(28.) Klessen C, Rogalla P, Taupitz M. Local staging of rectal cancer: The current role of MRI. Eur Radiol 2007; 17:379-389.
(29.) Wu S, Zheng J, Li Y, et al. Development and validation of an MRI-based radiomics signature for the preoperative prediction of lymph node metastasis in bladder cancer. EBioMedicine 2018; 34:76-84.
(30.) Huang Y-q, Liang C-h, He L, et al. Development and validation of a radiomics nomogram for preoperative prediction of lymph node metastasis in colorectal cancer. J Clin Oncol 2016; 34:2157-2164.
(31.) Chen LD, Liang JY, Wu H, et al. Multiparametric radiomics improve prediction of lymph node metastasis of rectal cancer compared with conventional radiomics. Life Sci 2018; 208:55-63.
(32.) Meng X, Xia W, Xie P, et al. Preoperative radiomic signature based on multiparametric magnetic resonance imaging for noninvasive evaluation of biological characteristics in rectal cancer. Eur Radiol 2019; 29:3200-3209.
Meng-Fei Xian (*) (iD) Xin Zheng (*) (iD) Jian-Bo Xu (iD) Xin Li (iD) Li-Da Chen (iD) Wei Wang (iD)
From the Department of Medical Ultrasonics (M.F.X.), East division of the First Affiliated Hospital of Sun Yat-sen University, Guangzhou, China; Department of Medical Ultrasonics (X.Z., L.D.C. [??] firstname.lastname@example.org, W.W. [??] email@example.com), Ultrasonics Artificial Intelligence X-Lab, Institute of Diagnostic and Interventional Ultrasound, The First Affiliated Hospital of Sun Yat-Sen University, Guangzhou, China; Department of Gastrointestinal Surgery (J.B.X.). The First Affiliated Hospital of Sun Yat-Sen University, Guangzhou, China; Research Center of GE Healthcare (X.L.), Shanghai., China.
(*) Meng-Fei Xian and Xin Zheng contributed equally to tis work.
Received 7 February 2020; revision requested 23 March 2020; last revision received 27 May 2020; accepted 2 June 2020.
Published online 19 April 2021.
You may cite this article as: Xian MF, Zheng X, Xu JB, Li X, Chen LD, Wang W. Prediction of lymph node metastasis in rectal cancer: comparison between shear-wave elastography and MRI.Diagn Interv Radiol 2021;27:424-431
* Ultrasomics based on shear-wave elastography was used for preoperative prediction of LN metastasis in patients with rectal cancer.
* Patients with LN metastasis had higher signature than patients without LN metastasis.
* SWE ultrasomics showed superior diagnostic performance compared with MRI for identifying the LN status.
* Ultrasomics can guide preoperative adjuvant chemoradiation of rectal cancer patients.
Table 1. Basic characteristics of patients Characteristics LN metastasis (+) Number of patients, n (%) 40 (46.0) Gender, n (%) Male 21 (52.5) Female 19 (47.5) Age (years), mean[+ or -]SD (range) 58[+ or -]13 (29-77) Histologic T stage, n (%) T1/2 1 (2.5) T3/4 39 (97.5) Size <3 cm 7 (17.5) 3-6 cm 31 (77.5) >6 cm 2 (5.0) Location, n (%) Upper 17 (42.5) Middle 14 (35.0) Lower 9 (22.5) CEA, n (%) Normal (< 5 [micro]g/L) 21 (52.5) Abnormal (> 5 [micro]g/L) 19 (47.5) CA19-9 level, (%) Normal (<35 U/mL) 38 (95.0) Abnormal (>35 U/mL) 2 (5.0) Hemoglobin in stool, n (%) Normal 5 (12.5) Abnormal 35 (87.5) Transferrin in stool, n (%) Normal 8 (20.0) Abnormal 32 (80.0) White blood cells in stool, n (%) Normal 34 (85.0) Abnormal 6 (15.0) Red blood cells in stool, n (%) Normal 32 (80.0) Abnormal 8 (20.0) Characteristics LN metastasis (-) p Number of patients, n (%) 47 (54.0) Gender, n (%) 0.507 Male 28 (59.6) Female 19 (40.4) Age (years), mean[+ or -]SD (range) 61[+ or -]10 (38-78) 0.196 Histologic T stage, n (%) 0.000 (*) T1/2 22 (46.8) T3/4 25 (53.2) Size 0.189 <3 cm 16 (34.1) 3-6 cm 30 (63.8) >6 cm 1 (2.1) Location, n (%) 0.124 Upper 13 (27.6) Middle 14 (29.8) Lower 20 (42.6) CEA, n (%) 0.138 Normal (< 5 [micro]g/L) 32 (68.0) Abnormal (> 5 [micro]g/L) 15 (32.0) CA19-9 level, (%) 0.209 Normal (<35 U/mL) 47 (100) Abnormal (>35 U/mL) 0 (0) Hemoglobin in stool, n (%) 0.775 Normal 6 (12.8) Abnormal 41 (87.2) Transferrin in stool, n (%) 0.702 Normal 11 (23.4) Abnormal 36 (76.6) White blood cells in stool, n (%) 0.989 Normal 40 (85.1) Abnormal 7 (14.9) Red blood cells in stool, n (%) 0.360 Normal 41 (87.2) Abnormal 6 (12.8) The p value is derived from the comparison between the LN metastasis (+) and the LN metastasis (-), using the student's t-test for age or chi-square test for other clinicopathologic variables. LN, lymph node; SD, standard deviation; CEA, carcinoembryonic antigen; CA19-9, cancer antigen 19-9. (*) p < 0.05.
Table 2. The 13 predictors selected from the 1044 SWE ultrasomics features Parameters Selected features Histogram parameters RelativeDeviation Percentile90 GLCM parameters lnverseDifferenceMoment_AIIDirection_offset3_SD GLCMEntropy_angle45_offset5 Texture parameter lnertia_AIIDirection_offset7_SD GLRLM parameters ShortRunHighGreyLevelEmphasis_AIIDirection _offset1_SD RunLengthNonuniformity_AIIDirection_offset3_SD ShortRunLowGreyLevelEmphasis_AIIDirection _offset5_SD ShortRunLowGreyLevelEmphasis_AIIDirection _offset7_SD ShortRunLowGreyLevelEmphasis_AIIDirection _offset8_SD ShortRunLowGreyLevelEmphasis_AIIDirection _offset9_SD LongRunLowGreyLevelEmphasis_AIIDirection _offset5_SD HighlntensityLargeAreaEmphasis Parameters Coefficients Histogram parameters 0.000002 0.003763 GLCM parameters 23.82352 0.525355 Texture parameter 0.001977 GLRLM parameters 0.003099 0.013506 1571.633 497.2829 837.4631 473.2587 21.41898 0.000003 There were four kinds of SWE ultrasomics parameters listed in column 1 which were extracted by AK software in our study, and the four parameters contain 13 selected predictors in column 2. The coefficients of regression model are listed in column 3.
Table 3. The diagnostic results of SWE ultrasomics and MRI Pathology LN metastasis (+), n=40 LN metastasis (-), n=47 SWE ultrasomics, n (%) LN metastasis (+) 35 (87.5) 10(21.3) LN metastasis (-) 5(12.5) 37 (78.7) MRI, n (%) LN metastasis (+) 31 (77.5) 12(25.5) LN metastasis (-) 9(22.5) 35 (74.5) Total SWE ultrasomics, n (%) LN metastasis (+) 45 LN metastasis (-) 42 MRI, n (%) LN metastasis (+) 43 LN metastasis (-) 44 Variables represent the number and percentage of cases diagnosed by SWE/MRI or pathology. SWE, shear-wave elastography; MRI, magnetic resonance imaging; LN, lymph node.
Table 4. The diagnostic performance of SWE ultrasomics and MRI AUC Accuracy 95% CI Sensitivity Specificity SWE ultrasomics 0.883 82.8% 0.732,0.900 87.5% 78.8% MRI 0.760 75.9% 0.655,0.844 77.5% 74.5% PPV NPV SWE ultrasomics 77.8% 88.1% MRI 72.1% 79.6% SWE, shear-wave elastography; MRI, magnetic resonance imaging; AUC, area under the ROC curve; 95% CI, 95% confidence interval; PPV, positive predictive value; NPV, negative predictive value.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||MODALITY-BASED (US, CT, MRI, PET-CT) IMAGING: ORIGINAL ARTICLE; magnetic resonance imaging|
|Author:||Xian, Meng-Fei; Zheng, Xin; Xu, Jian-Bo; Li, Xin; Chen, Li-Da; Wang, Wei|
|Publication:||Diagnostic and Interventional Radiology|
|Date:||May 1, 2021|
|Previous Article:||Discontinuous moving shot technique for conformal thermal ablation in an ex vivo porcine liver model.|
|Next Article:||MRI of lateral hindfoot impingement.|