| Rectal cancer(RC)is one of the most common malignant tumors.Its treatment decision and prognosis are different according to tumor stage,classification and neoadjuvant treatment effect.At present,there is a lack of biological markers to predict some pathological characteristics(classification,vascular invasion,etc.)and neoadjuvant chemoradiotherapy treatment effect of RC Preoperatively,so it could not effectively classify the risk and perform personal treatment accurately.Radiomics,which could extract high-throughput data from medical images,noninvasively predict tumor biological information and reflect tumor heterogeneity,has become a new hotspot.We would like to use ultrasound-based and contrast enhanced ultrasound(CEUS)-based radiomics technology to explore the application in the staging,typing and treatment of RC from the following three parts.Part one Ultrasound-based radiomics for preoperative prediction of stages and lymphovascular invasion in rectal cancer.Objective To create and test an ultrasound-base radiomics model for predicting TNM stages and lymphovascular invasion(LVI)preoperatively in patients with rectal cancer(RC).Methods A total of 629 patients with RC underwent surgical resection in our hospital,with complete preoperative ultrasound examination and postoperative pathological information.We analyzed the differences in clinical parameters of different substages of rectal cancer TNM stages,and evaluated the risk factors of different labels using logistic regression analysis.Radiomics models were established to predict TNM stages and LVI.Radiomics features were extracted from the gray ultrasound images.LASSO regression analysis was used for reducing data dimension,selecting feature,and building radiomics signature.The receiver operating characteristic curve(ROC)analysis and area under the ROC curve(AUC)were used to evaluate the models.Results(1)Ultrasound-based radiomics predicting T stage of RC: A total of629 patients with RC were included in this study,including 126 patients with T1-2 stage and 503 patients with T3-4 stage.Using ultrasound to diagnosis T stage of RC,the AUC,sensitivity,specificity and accuracy were 0.691,0.909,0.476 and 0.822,respectively.Using ultrasound-based radiomics to predicted T stage of RC,the AUC,sensitivity,specificity and accuracy were 0.699,0.887,0.263 and 0.762,respectively in testing set.The diagnostic efficiency of T staging of RC using ultrasound is better than the prediction efficiency of T staging of RC using Ultrasound-based radiomics.Chi-square test and logistic regression analysis showed that RC with higher T stage was more likely to be associated with lymph node metastasis,distant metastasis,bigger tumor size and higher CEA.(2)Ultrasound-based radiomics predicting N stage of RC: A total of 629 patients with RC were included in this study,including 236 patients with N0 stage and393 patients with N1-2 stage.Using ultrasound to diagnosis N stage of RC,the AUC,sensitivity,specificity,and accuracy were 0.658,0.539,0.780,and 0.630,respectively.Using ultrasound-based radiomics to predicted N stage of RC,the AUC,sensitivity,specificity and accuracy were 0.591,0.805,0.324 and 0.624,respectively in testing set.Comparing to use ultrasound to diagnosis N stage of RC,Ultrasound-based radiomics predicting N stage of RC with slightly higher sensitivity and lower specificity.Chi-square test and logistic regression analysis showed that RC with lymph node metastasis was more likely to have higher T stage and distant metastasis.(3)Ultrasound-based radiomics predicting M stage of RC: A total of 629 patients with RC were included in this study,including 532 patients with M0 stage and 97 patients with M1 stage.The AUC,sensitivity and specificity were0.862,0.603 and 0.946,respectively in training set;the AUC,sensitivity and specificity were 0.626,0.276 and 0.862,respectively in testing set.Chi-square test and logistic regression analysis showed that RC with distant metastasis was more likely to have higher T stage,lymph node metastasis,bigger tumor size and higher CEA,CA125,CA242.(4)Ultrasound-based radiomics predicting LVI of RC: A total of 203 patients with RC were included in this study,including 170 patients with LVI negative and 33 with LVI positive.The AUC,sensitivity and specificity were 0.849,0.760 and 0.750,respectively in training set;the AUC,sensitivity and specificity were 0.781,0.750 and 0.790 respectively in testing set.Conclusions The higher T,N and M stages of RC are correlated with tumor size and CEA.Using ultrasound-based radiomics model to predict TNM stages of RC,it has better diagnostic efficiency in a in training set and lower diagnostic efficiency in testing set,suggesting that the TNM stages of RC should be based on direct ultrasound imaging rather than ultrasound-based radiomics.However,for LVI which could not be visually displayed in ultrasound image,the ultrasound-based radiomics model shows better prediction performance,suggesting that ultrasound-based radiomics could predict LVI preoperatively in patients with RC and provide reference for clinical treatment decisions.Part two Application of radiomics model based on ultrasound and contrast-enhanced ultrasound in preoperative prediction of rectal mucinous adenocarcinomaObjective To develop ultrasound and contrast-enhanced ultrasound(CEUS)radiomics models for the diagnosis of rectal mucinous adenocarcinoma(MAC).Methods(1)2D ultrasound images,clinical parameters and pathological information of 629 patients with rectal cancer(RC)were retrospectively collected.Divided into MAC and non-MAC groups and compared clinical parameters between groups.We used drawing software to sketch and extract the radiomics features on the ultrasound image of the maximum section of the tumor.Then,divided patients into training and testing sets in a 7:3 ratio.The area under the curve(AUC)of the receiver operating characteristic(ROC)curve was applied to assess model’s performance.(2)A total of 148 patients with RC performing CEUS examination and surgery were included in this retrospective study.Divided into MAC and non-MAC groups and compared clinical parameters between groups.We used drawing software to sketch and extract the radiomics features on the ultrasound image of the maximum section of the tumor.Then,divided patients into training and testing sets in a 7:3 ratio.The area under the curve(AUC)of the receiver operating characteristic(ROC)curve was applied to assess model’s performance.Results(1)Ultrasound-based radiomics predicting MAC: A total of 5936 features were extracted from 629 patients with RC,which were divided into MAC group(n=63)and non-MAC group(n=566).And 13 optimal features were identified for the development of radiomic model,with the AUC value of 0.865,0.655 in the training and testing sets,respectively.Compared with the non-MAC group,chi-square test and logistic regression analysis showed that the tumor in the MAC group was larger(P<0.05)(2)CEUS-based radiomics predicting MAC: 148 patients in the study were divided into MAC group(n=37)and non-MAC group(n=111).A total of 5936 features were extracted,and 149 optimal features were identified for the development of radiomic model,with the AUC value of 0.988,0.885 in the training and testing sets,respectively.Compared with the non-MAC group,chi-square test and logistic regression analysis showed that the tumor marker CA724 in the MAC group was higher(P<0.05).Conclusions Compared with non-MAC group,MAC group has larger tumors and higher CA724 levels,suggesting that larger rectal cancers with higher CA724 levels may be MAC.Ultrasound and CEUS could not directly diagnose rectal MAC,however,Ultrasound-base and CEUS-based radiomic models might be effective supplemental means to diagnosis MAC.Part three The application of ultrasound-based radiomics model in predicting the effect of neoadjuvant chemoradiotherapy for rectal cancerObjective To develop an ultrasound-based radiomics model in predicting the effect of neoadjuvant chemoradiotherapy(NCRT)for rectal cancer(RC).Methods A total of 71 patients with RC performing neoadjuvant chemoradiotherapy(NCRT)and surgery were included in this retrospective study.The patients were divided into effective group(n=39)and ineffective group(n=32),and divided to a training or test set(7:3 ratio).The area of interest of the RC was determined by manually identifying the ultrasound images of the tumor,and radiological features were obtained.Selected the best features and built a radiation model.Useing the area under the curve(AUC)of the receiver operating characteristic(ROC)curve to evaluate the performance of model.Results A total of 5936 features were extracted,and 37 optimal features were identified for the development of radiomic model,with AUC value of 0.878 and0.804 in training set and testing set,respectively.Chi-square test and logistic regression analysis showed that lymph node metastasis was a high-risk factor for the failure of NCRT(P<0.05).Conclusions Rectal cancer with lymph node metastasis may not be suitable for NCRT.Ultrasound-based radiomics model might be an effective complementary method for predicting NCRT outcomes. |
PDF Full Text Request