The Clinical Research Of High-resolution MR,Dynamic Enhanced MR And Radiomics Nomogram In The Diagnosis Of Lymph Node Metastasis Of Rectal Cancer

Part Ⅰ The efficacy of short axis diameter combined with morphological features on high-resolution MRI in the diagnosis of lymph node metastasis of rectal cancer and the incremental value of morphological featuresBackground and purpose:The clinical stage of rectal cancer patients with lymph node metastasis(LNM)belongs to stage Ⅲ,and intensive treatment should be given.Previous studies have used lymph node size or morphological features to diagnose LNM,but the size criteria and morphological features were different in each study.In 2016,the European Society of Gastrointestinal and Abdominal Radiology(ESGAR)proposed a practical criterion for the diagnosis of LNM by size(short axis diameter 5 mm,short axis diameter 9 mm)combined with morphological features(irregular border,round shape and heterogeneous signal).At present,there is no relevant report on the efficacy of the 2016 ESGAR criteria for diagnosing LNM in rectal cancer in China.High-resolution MR(HR-MRI)has the advantages of thin slices,small field of view,and multi-directional scanning,which can measure the size of lymph nodes more accurately and display the morphological characteristics of lymph nodes more clearly.Therefore,our first part of the study intends to investigate the efficacy of 2016 ESGAR criteria for diagnosing LNM in rectal cancer on HR-MRI and the incremental value of morphological features in 2016 ESGAR criteria.Methods:174 rectal cancer patients who underwent HR-MRI before radical resection from December 2016 to September 2021[112 males and 62 females;average age(65.1 ±9.8)years old,range from 33 to 86 years old]were retrospectively enrolled.HR-MRI images were retrospectively re-interpreted by two radiologists to evaluate whether the patient had LNM or not according to three criteria(short axis diameter 5mm,9mm,2016 ESGAR criteria).Based on postoperative pathological results as the "golden standard",patients were divided into patients with LNM and patients without LNM.The value of short axis diameter 5mm,9mm,size combined with morphological features in the diagnosis of LNM of rectal cancer was calculated and compared.χ2 test was used to compare the sensitivity,specificity and accuracy of different criteria in the diagnosis of rectal cancer LNM.Results:Postoperative pathological results confirmed that 66 patients had LNM,108 patients without LNM.The sensitivity,specificity and accuracy of the diagnosis of LNM with the criteria of short axis diameter≥5mm were 53.03%(35/66)、75.00%(81/108)、66.67%(116/174),and 13.64%(9/66)、96.30%(104/108)、64.94%(113/174)for short axis diameter≥9mm.Combined with morphological features,the sensitivity and accuracy were increased to 71.21%(47/66,χ2=4.636 and 44.786,P=0.031and 0)and 77.59%(135/174,χ2=5.160 and 6.792,P=0.023 and 0.009).While the specificity is 81.48%(88/108),which is no statistical difference with the 5mm criteria(χ2=1.332,P=0.248),and was reduced with the 9mm criteria(χ2=12.000,P<0.001).12 and 38 rectal cancer patients with LNM that were missed by the 5mm and 9mm criteria were corrected by combining the morphological features respectively.Conclusion:Size combined with morphological features in the diagnosis of LNM of rectal cancer has medium sensitivity,high specificity and unsatisfactory accuracy.Combining size and the morphological features of lymph nodes on HR-MRI can significantly improve the sensitivity and accuracy of diagnosing LNM in rectal cancer,and help a lot to correct missed cases of size criteria.Part Ⅱ Value of quantitative parameters of dynamic contrast-enhanced MRI combined with high-resolution MRI findings in predicting lymph node metastasis of rectal adenocarcinomaBackground and purpose:Lymph node metastasis(LNM)is an important factor affecting the treatment and prognosis of patients with rectal cancer.Currently,the size and morphological characteristics of lymph nodes are used to evaluate the lymph node status of rectal cancer,but the size and morphological characteristics have some limitations.The result of our first part showed that size combined with morphological features on HR-MRI had moderate sensitivity,high specificity and poor accuracy in diagnosing LNM of rectal cancer.Some scholars have proposed that dynamic contrast-enhanced MRI(DCE-MRI)quantitative parameters of rectal cancer primary lesions have the potential to predict LNM of rectal cancer.Therefore,our second part intends to investigate the value of DCE-MRI quantitative parameters of rectal cancer primary lesions combined with high-resolution MRI(HR-MRI)imaging findings in predicting LNM of rectal adenocarcinoma.Methods:98 rectal adenocarcinoma patients[62 males and 36 females;mean age(64.9±8.3)years old,range from 41 to 81 years old]who underwent DCE-MRI and HR-MRI examination before radical resection in our hospital from January 2019 to September 2021 were retrospectively enrolled.The HR-MRI findings were re-interpreted by two radiologists.Rectal primary tumor was outlined in TISSUE 4D software by two radiologists,and the quantitative parameters(Ktrans,kep,Ve,iAUC)of DCE-MRI were measured.Taking the postoperative pathological results as the "golden standard",the patients were divided into patients with LNM and patients without LNM.The differences of clinical risk factors,HR-MRI findings and DCE-MRI quantitative parameters between the two groups were analyzed.The factors predicting LNM were screened by univariate and multivariate logistic regression analysis.Receiver operating characteristic curve(ROC)were used to calculate the efficiency in predicting LNM.Delong test was used to compare the area under the curve(AUC)of different parameters and combined model in predicting LNM in rectal adenocarcinoma.Chi-square test was used to compare the sensitivity and specificity of different parameters and combined model in predicting LNM in rectal adenocarcinoma.Results:Postoperative pathological results confirmed that 36 patients with LNM and 62 patients without LNM.Compared with the patients without LNM,the patients with LNM were younger(62.4±8.9 vs 66.4±7.6,P=0.019)and had higher CEA abnormality rate(16/20 vs 14/48,P=0.024),higher HR-MRI reported T-stage(24/12 vs 25/37,P=0.012),higher HR-MRI reported N-stage(24/12 vs 14/48,P<0.001),higher HR-MRI reported positive rate of mesorectal fasciae(mrMRF)involvement(10/26 vs 4/58,P=0.004),and higher HR-MRI reported positive rate of extramural vascular invasion(mrEMVI)(13/23 vs 11/51,P=0.041),lower Ktrans value[0.247(0.190,0.330)min-1 vs 0.362(0.245,0.502)min-1,P=0.015].The results of univariate logistic regression analysis showed that age<59.5 years old,abnormal CEA,HR-MRI reported T3 stage,HR-MRI reported positive N stage,HR-MRI reported positive MRF involvement,HR-MRI reported positive EMVI and DCE quantitative parameter Ktrans<0.298 min-1 were risk factors for LNM of rectal adenocarcinoma[odds ratio(OR)value were 0.331,2.743,2.960,6.857,5.577,2.261 and 0.210 respectively,P values were 0.043,0.026,0.013,<0.001,0.007,0.045 and 0.001respectively.Multivariate logistic regression analysis showed that HR-MRI reported positive N stage(OR=5.574,95CI:2.149～14.462,P<0.001)and Ktrans<0.298 min-1(OR=0.273,95CI:0.105～0.709,P=0.008)were independent predictors of LNM in rectal adenocarcinoma.Area under the curve(AUC)of the logistic model with positive HR-MRI reported N stage combined with Ktrans<0.298 min-1 in predicting LNM were 0.787(95%CI:0.694～0.881),with a sensitivity of 88.9%and a specificity of 56.5%.DeLong test showed that the logistic model combined with HR-MRI reported N stage,and Ktrans had higher AUC in predicting LNM status than HR-MRI reported N stage alone,and Ktrans alone(Z value were 2.625 and 2.708 respectively,and P values were 0.001 and 0.001 respectively).Compared with HR-MRI reported N stage alone and Ktrans alone,the sensitivity of combined model was improved(χ2=5.143 and 4.126,P values were 0.023 and 0.042 respectively).The specificity is lower than that of mrN stage alone(χ2=6.158,P=0.013),while there is no statistical difference compared with the specificity of Ktrans alone(χ2=1.679,P=0.195).Conclusions:HR-MRI reported positive N stage and Ktrans<0.298 min-1 were independent predictors of LNM in rectal adenocarcinoma.Compared with HR-MRI reported N stage alone and Ktrans alone,the combined model had the highest AUC and the highest sensitivity in predicting LNM.By using the combined model,we could screen more patients with LNM and thus benefited from preoperative neoadjuvant chemoradiotherapy.Part Ⅲ High resolution T2 weighted images and apparent diffusion coefficient maps based radiomics nomogram for prediction of lymph node metastasis in rectal adenocarcinomaBackground and purpose:The clinical stage of rectal cancer patients with lymph node metastasis(LNM)belongs to stage Ⅲ,and intensive treatment should be given.Diagnosis of rectal cancer LNM based on lymph node size and morphological features on high-resolution magnetic resonance imaging(HR-MRI)has some limitations.Dynamic contrast-enhanced MRI(DCE-MRI)quantitative parameters of rectal cancer primary lesions combined with HR-MRI findings can improve the sensitivity of diagnosing LNM,but there are few centers that routinely carry out DCE-MRI in clinical practice and the critical values of DCE-MRI quantitative parameters for predicting LNM in rectal cancer are not uniform.Radiomics is an emerging research hotspot,analyzing and further decoding tumor heterogeneity by extracting high-throughput quantitative radiomics features from medical images.Therefore,our third part intends to establish and verify a radiomics model based on T2-weighted images(T2WI)and apparent diffusion coefficient(ADC)maps in predicting LNM of rectal adenocarcinoma.Methods:165 rectal adenocarcinoma patients in our hospital from December 2016 to September 2021 were retrospectively enrolled.The enrolled patients were randomly divided into training set(n=116)and verification set(n=49)according to the ratio of 7:3.The lesions were delineated on high-resolution oblique axial T2WI and high b value diffusion weighted images(DWI)layer by layer by using 3D-slicer software.Volume of interest(VOI)of the primary rectal cancer lesion obtained on the high-b value DWI were copied to the ADC maps,and 1595 radiomics features were extracted on the oblique axial T2WI and ADC maps respectively.Firstly,the radiomics features were initially screened by maximum correlation and minimum redundancy,then LASSO regression analysis was performed to screen the features again and radiomics signature was calculated.Univariate analysis was conducted on the radiomics signature,clinical risk factors and HR-MRI findings.Multivariate logistic analysis was carried out to investigate the final feature subset and thus the predicting model was established.Results:Postoperative pathological results confirmed that 61 patients with LNM and 104 patients without LNM.Compared with the patients without LNM,the patients with LNM were younger(62.5 ± 10.6 vs 65.6±8.8,P=0.046),had higher CEA abnormality rate(28/33 vs 22/82,P=0.002),higher HR-MRI reported T-stage(40/21 vs 45/59,P=0.009),higher HR-MRI reported N-stage(40/21 vs 22/82,P<0.001),higher HR-MRI reported positive rate of mesorectal fasciae(mrMRF)involvement(18/43 vs 13/91,P=0.013),and higher HR-MRI reported positive rate of extramural vascular invasion(mrEMVI)(21/40 vs 12/92,P=0.001).The results of univariate logistic regression analysis showed that age,abnormal CEA,HR-MRI reported T3 stage,HR-MRI reported positive N stage,HR-MRI reported positive MRF involvement and HR-MRI reported positive EMVI were risk factors for predicting LNM of rectal adenocarcinoma[odds ratio(OR)value were 0.967,3.163,2.497,7.100,2.930 and 4.025,P values were 0.049,0.010,0.006,<0.001,0.008 and 0.001 respectively.Multivariate logistic regression analysis showed that abnormal CEA(OR=1.121,95%CI:1.070～1.736,P=0.047),HR-MRI reported positive N stage(OR=3.336,95%CI:1.179～9.434,P=0.023)and radiomics signature(OR=42.015,95%CI:8.494～207.819,P<0.001)were independent predictors of LNM in rectal adenocarcinoma(OR values were 1.121、3.336 and 42.015 respectively,P values were 0.047,0.023 and<0.001 respectively).The predictive nomogram of LNM included CEA,mrN stage and radiomics signature.The predictive nomogram showed good predictive performance[area under curve(AUC)=0.91;95%confidence interval[CI],0.84-0.96]which was better than clinic-radiological model(AUC,0.77;95%CI,0.66-0.88;Delong test,P=0.002)or radiomics signature-only model(AUC,0.82;95%CI,0.73-0.92;Delong test,P=0.016)in training cohort.Application of the nomogram in the validation cohort still exhibited good performance(AUC,0.88;95CI,0.77-0.98).The sensitivity,specificity and accuracy of the combined model in predicting LNM was 0.82,0.90 and 0.87 in training cohort and 0.86,0.82 and 0.84 in validation cohort.The result of decision curve analysis(DCA)indicated that using nomogram to predict LNM gains more benefit when the threshold probability from 0 to 1.0 for a doctor or a patient than "treat all”or "treat none".Conclusion:The nomogram combined preoperative CEA,mrN stage with radiomics signature could accurately predict LNM of rectal cancer.