| Objective:Urolithiasis is a common disease with high incidence in modern society.Although the surgical treatment of urolithiasis is developing rapidly,its high incidence and recurrence rate still bring great medical and economic burden to patients and social system,which suggests that it is important to study the etiology,prevention,and control of recurrence of urinary stones.The stone composition plays an important role in the prevention and treatment of urolithiasis.As the "pathology" of stones,it directly guides the diagnosis of the etiology of stones,and it guides the prevention and treatment measures such as diet and drugs.The compositions of urinary stones are diverse,and the infrared spectroscopy has been able to accurately analyze the composition of stones.However,its disadvantage lies in that the stone specimens must be obtained by invasive means such as surgery.For stones patients without surgery,in vitro analysis methods cannot obtain information of stone compositions,so it is difficult to carry out medical prevention and accurate intervention for urinary stones in advance.Metabolic assessment is an important method for the etiological diagnosis of urinary stones.Clinical indicators such as physiological and biochemical indexes of patients have certain suggestive ability for the stone compositions,which can provide information for the diagnosis of stone compositions in vivo.Imaging is another effective means for the stones in vivo.In recent years,dual-energy CT has been initially applied in the identification of stone compositions in vivo because of its ability to distinguish the atomic number of substance elements.However,its ability to identify other compositions except uric acid stones is weak and has certain limitations.Nowadays,with the rapid development of artificial intelligence technology,auxiliary diagnostic systems based on various high-throughput data have been more and more involved in clinical decision-making.High-throughput data can be extracted in CT images including grayscale and texture,forming a new type of omics research called "radiomics",which has been widely applied in the field of tumors and also has great application prospects in the field of urinary stones.This study intends to systematically study and solve the problem of accurate analysis of urinary stone compositions in vivo from the three dimensions of clinical features,dual-energy CT features and radiomics features.Materials and Methods:(1)Retrospectively collect patients who were confirmed to have urinary stones from 2010 to 2019 in the Department of Urology,West China Hospital,Sichuan University,and who underwent infrared spectroscopy to analyze the composition of the stones after surgery.Collect the clinical features and the incidence of postoperative complications of patients,and analyze the differences in clinical features and surgical prognosis of patients with different urinary stone compositions;(2)By the end of December 2020,studies using dual-energy CT to analyze urinary stone compositions in vivo were retrieved and included.The patient population was patients with urinary stones who planned to undergo surgical lithotomy.The accuracy of dual-energy CT in the diagnosis of urinary stone compositions in vivo was analyzed by infrared spectroscopy or X-ray crystal diffraction as the gold standard.Use Quality Assessment of Diagnostic Accuracy Studies to evaluate the quality of the included studies,and extract the four grids’ data.Use Stata14.0 software and bivariate mixed-effects model for diagnostic meta-analysis,draw summary receiver characteristic operating curve and calculate the corresponding AUC value;(3)Patients with urinary stones who underwent dual-energy CT scans before surgery in the Department of Urology,West China Hospital,Sichuan University from September 2017 to May 2020,and obtained stone specimens for infrared spectroscopy were included.Collecting four dual-energy parameters,including dual-energy ratio,energy spectrum slope,electron density and effective atomic number under multi-mode dual-energy CT.Analyze the dual-energy feature differences among various stone compositions,and construct a multi-class logistic regression model.Draw the ROC curve of each category,calculate the AUC value to evaluate the diagnostic efficacy.(4)The patient cohort in this part is the same as that in Chapter 3.Collect the original files of dual-energy CT images.Use ITK-SNAP software to perform layer-by-layer three-dimensional ROI segmentation to create radiomics mask labels.Use python’s pyradiomics library for high-throughput radiomics feature extraction.Meanwhile,19 clinical features and 4 dual-energy CT features in Chapter 3 are collected.After features selection made by the Lasso regression model,a multi-class logistic regression model,a multi-class nonlinear support vector machine model,a random forest model,and a fusion model of the above three models are established respectively.The ROC curve and AUC of each classifier are calculated to evaluate the diagnostic efficiency.Results:(1)Based on the strict inclusion and exclusion criteria,1063 patients were finally enrolled,including 697 pure Ca Ox stones,287 Ca Ox+Ca P stones,47 UA stones,24 STR stones,and 6 CYS stones(2 calcium hydrogen phosphate stones were excluded due to the small sample size).In the analysis of general features,there are statistical differences in gender,age,and BMI of different stone compositions(P<0.05): the overall proportion of males is 68%,while the proportion of females in the Ca Ox+Ca P group is slightly higher(38.3%).The proportion of females in the STR group is extremely high(66.7%).The patients of UA stones are older(mean age of 53.4 years);while the patients of CYS stones are younger(mean age of 36.5 years),and their BMI is significantly smaller(21.3±3.9).In terms of metabolic features,the total cholesterol and low-density lipoprotein contents of patients in the CYS group were higher than those of other groups.Among the infection features of blood and urine,the positive rates of urinary nitrite for Ca Ox stones,Ca Ox+Ca P stones,UA stones,STR stones,and CYS stones were 2.6%,7.3%,2.1%,12.5%,and 0%,respectively;The positive rates of urine pus cells were3.9%,6.6%,2.1%,16.7%,and 0%;the positive rates of urine culture were 13.1%,22.7%,14.8%,38.9%,and 0%.The post-mortem examination showed that the positive rate of urine nitrite and urine culture in the infectious stone group was higher than the other three groups(and the STR group> Ca Ox+Ca P group);the positive rate of urine pus cells in the STR group was higher than other groups.In the analysis of postoperative complications,the postoperative fever rates of Ca Ox stones,Ca Ox+Ca P stones,UA stones,STR stones,and CYS stones were 1.9%,2.8%,3.7%,11.1%,and 0%,respectively.There was a statistically significant difference between them(P=0.045).The postoperative fever rate of the STR group was significantly higher than that of the other groups.(2)Finally,924 patients with 1204 stones(a total of 17 studies)were enrolled in the pooled analysis.The combined sensitivity and specificity of uric acid stones were 0.92(0.85-0.96),0.99(0.98-1.00),and the AUC value of combined SROC curve was 0.99(0.98-1.00).The combined sensitivity and specificity of calcium-containing stones were 1.00(0.98-1.00),0.89(0.79-0.95),and the AUC value of combined SROC curve was 1.00(0.99-1.00).The heterogeneity test did not show significant heterogeneity.The sample size of cystine stones in only 4 articles is more than 5,which is difficult to combine for meta-analysis;the current dual-energy CT software cannot identify magnesium ammonium phosphate stones,and cannot accurately identify subtypes of calcium-containing stones.(3)From September 2017 to May 2020,120 stone patients(169 stones in total)in the Department of Urology,West China Hospital,Sichuan University were enrolled,all of whom underwent dual-energy CT examinations before stone removal.Among them,84 were males,36 were females.After the operation,infrared spectroscopy was used to analyze the stone compositions in all patients,including119 Ca Ox stones,29 Ca Ox+Ca P stones,9 UA stones,5 STR stones,and 6 CYS stones(1 calcium hydrogen phosphate stones were excluded due to the small sample size).The differences in the dual energy ratio,energy spectrum slope,electron density and effective atomic number among the stone compositions are statistically significant(P<0.001).The multi-class logistic regression model constructed by four dual-energy CT features showed that the AUC values of Ca Ox stones,Ca Ox+Ca P stones,UA stones,STR stones,CYS stones were [0.727,0.670,0.876,0.395,0.961].(4)The patient’s basic information and stone composition information are the same as the third part.After radiomics feature extraction and feature selection by Lasso regression,30 features were finally included,including 11 clinical features,4dual-energy CT imaging features,and 15 radiomics features.The logistic regression model,nonlinear support vector machine model and random forest model were established respectively by using the included features.The AUC values of the area under the ROC curve for Ca Ox stones,Ca Ox+Ca P stones,UA stones,STR stones,and CYS stones of the logistic regression model are 0.888,0.802,0.986,0.861,and0.999,respectively;the AUC values of the five types of stones of support vector machine model are 0.911,0.847,0.964,0.953,and 1.000,respectively;the AUC values of the five types of stones of the random forest model are 0.890,0.821,0.959,0.963,and 1.000,respectively.The ensemble model was constructed by the probability voting method,and the final AUC values of the five types of stones are0.916,0.840,0.983,0.969,and 1.000,respectively.Conclusion:1.There are significant differences in the clinical features and prognosis of patients with different stone compositions,which are mainly in the gradient change of urine p H among different compositions,the urine infection index and postoperative fever rate in patients with infectious stones are significantly higher,the patients with uric acid stone are older,etc.,which provides preliminary evidence for the clinical features to participate in prediction of the stone compositions in vivo.2.The current dual-energy CT can accurately distinguish between uric acid/non-uric acid stones and calcium-containing/non-calcium-containing stones,but the high accuracy is based on the negligence of other low-morbidity stones.Dual-energy CT now cannot diagnose magnesium ammonium phosphate stones,and there is no definite conclusion on the ability to recognize cystine stones.It is also difficult to distinguish the subtypes of calcium-containing stones.Also,the dual-energy features currently used for diagnosis are relatively simple(only the dual-energy ratio).3.The multi-class diagnosis model using four dual-energy CT parameters,including dual-energy ratio,energy spectrum slope,electron density,and effective atomic number,has better distinguishing ability for uric acid and cystine,but still cannot distinguish calcium-containing stone subtypes and magnesium ammonium phosphate stones.4.Integrating the three dimensions of information,including clinical features,dual-energy CT multi-parameter features,and radiomics features,a high-dimensional clinical diagnosis model based on artificial intelligence methods have good diagnostic performance of five types of stone compositions.5.We finally obtained a new multi-class clinical diagnostic model with good diagnostic performance and a complete diagnosis of multiple stone compositions,so that the problem of accurate analysis of urinary stone compositions in vivo is expected to be solved on a practical level.Dual-energy CT is expected to become a "one-stop" non-invasive examination of urinary stones,which will contribute to the early and precise intervention and treatment decision-making of the patients with urinary stones. |
PDF Full Text Request