Establishment of the prediction models of malignant brain edema after vascular recanalization in anterior circulation acute large vascular occlusion stroke / 国际外科学杂志

Objective:Based on Logistic regression and XGBoost algorithm, the prediction model of malignant brain edema (MBE) after vascular recanalization of anterior circulation acute great vessel occlusive stroke (ALVOS) was constructed, and the prediction performance was compared.Methods:A retrospective selection of 382 patients with anterior circulation ALVOS who underwent early endovascular treatment (EVT) in our hospital from March 2014 to June 2020 and successfully recanalized the occluded blood vessel was selected. The patients were divided into the training group ( n=267) and the test group ( n=115) according to the ratio of 7∶3 by the random number table method. According to whether the patients had MBE after successful recanalization of the occluded blood vessels, the training group was divided into the MBE group ( n=41) and non-MBE group ( n=226). The baseline data, treatment and brain computed tomography perfusion(CTP) results of MBE group and non-MBE group in training group and test group were compared respectively, including age, admission score of National Institutes of Health Stroke Scale (NIHSS), grade of cerebral collateral circulation, cerebral blood volume, and so on. Logistic regression model and XGBoost algorithm model were used to screen the predictors of MBE in ALVOS patients with occluded vessels successfully recanalized, and the discrimination and calibration of the two models were compared. The measurement data conforming to the normal distribution were expressed as mean ± standard deviation ( ± s), and the independent sample t test was used for comparison between the two groups. Non-normally distributed measurement data were represented by M ( Q1, Q3), using independent sample Mann-Whitney U test. The chi-square test was used to compare the count data between groups. Results:There was no significant difference in baseline data, treatment status, and cranial computed tomography perfusion (CTP) imaging results of the training group and the test group ( P>0.05). The age, admission systolic blood pressure, admission NIHSS score, proportion of hypertension, proportion of cerebral collateral circulation 0-2, proportion of thrombus removal times> 3 times, time from onset to recanalization, and cerebral blood volume (CBV) of MBE group were (68.95±8.04) years old, (146.71±22.73) mmHg, 17(13, 21) min, 87.80%, 82.93%, 68.29%, (365.64±87.83) min, (32.56±5.73) mL/100 g, obvious higher than the non-MBE group [(60.27±7.13) years old, (137.92±19.58) mmHg, 14(10, 18) points, 73.01%, 60.62%, 2.65%, (307.59±74.05) min, (27.49±5.46) mL/100 g] ( P<0.05). The results of Logistic regression model showed that age, NIHSS on admission, grade of cerebral collateral circulation, times of thrombectomy and time from onset to recanalization were the predictors of MBE after successful recanalization of occluded vessels after EVT in patients with anterior circulation ALVOS ( P<0.05). The top five important feature scores of XGBoost algorithm model were cerebral collateral circulation classification 34, embolectomy times 27, onset to vascular recanalization time 25, admission NIHSS score 22, age 16.In the training set, the area under the curve of the Logistic regression model was 0.816(95% CI: 0.749-0.883), and the Hosmer-Lemeshow test showed that χ2=1.547, P=0.438. The area under the curve of the XGBoost algorithm model was 0.856(95% CI: 0.799-0.913), and the Hosmer-Lemeshow test showed that χ2=1.021, P=0.998. Conclusion:Logistic regression model and XGBoost algorithm model had similar prediction performance for MBE after successful recanalization of occluded vessels after EVT in patients with anterior circulation ALVOS, and collateral circulation classification, number of thrombolysis, time from onset to recanalization, NIHSS score on admission, and age could be used as predictors.

Hybrid deep learning model for risk prediction of fracture in patients with diabetes and osteoporosis / 医学前沿

The fracture risk of patients with diabetes is higher than those of patients without diabetes due to hyperglycemia, usage of diabetes drugs, changes in insulin levels, and excretion, and this risk begins as early as adolescence. Many factors including demographic data (such as age, height, weight, and gender), medical history (such as smoking, drinking, and menopause), and examination (such as bone mineral density, blood routine, and urine routine) may be related to bone metabolism in patients with diabetes. However, most of the existing methods are qualitative assessments and do not consider the interactions of the physiological factors of humans. In addition, the fracture risk of patients with diabetes and osteoporosis has not been further studied previously. In this paper, a hybrid model combining XGBoost with deep neural network is used to predict the fracture risk of patients with diabetes and osteoporosis, and investigate the effect of patients' physiological factors on fracture risk. A total of 147 raw input features are considered in our model. The presented model is compared with several benchmarks based on various metrics to prove its effectiveness. Moreover, the top 18 influencing factors of fracture risks of patients with diabetes are determined.

Thyroid disorder classification using machine learning models

Introduction@#Thyroid hormones are produced by the thyroid gland and are essential for regulating the basal metabolic rate. Abnormalities in the levels of these hormones lead to two classes of thyroid diseases – hyperthyroidism and hypothyroidism. Detection and monitoring of these two general classes of thyroid diseases require accurate measurement and interpretation of thyroid function tests. The clinical utility of machine learning models to predict a class of thyroid disorders has not been fully elucidated. @*Objective@#The objective of this study is to develop machine learning models that classify the type of thyroid disorder on a publicly available thyroid disease dataset extracted from a machine learning data repository. @*Methods@#Several machine learning algorithms for classifying thyroid disorders were utilized after a series of pre-processing steps applied on the dataset. @*Results@#The best performing model was obtained by with XGBoost with a 99% accuracy and showing very good recall, precision, and F1-scores for each of the three thyroid classes. Generally, all models with the exception of Naïve Bayes did well in predicting the negative class generating over 90% in all metrics. For predicting hypothyroidism, XGBoost, decision tree and random forest obtained the most superior performance with metric values ranging from 96-100%. On the other end in predicting hyperthyroidism, all models have lower classification performance as compared to the negative and hypothyroid classes Needless to say, XGBoost and random forest did obtain good metric values ranging from 71-89% in predicting hyperthyroid class. @*Conclusion@#The findings of this study were encouraging and had generated useful insights in the application and development of faster automated models with high reliability which can be of use to clinicians in the assessment of thyroid diseases. The early and prompt clinical assessment coupled with the integration of these machine learning models in practice can be used to determine prompt and precise diagnosis and to formulate personalized treatment options to ensure the best quality of care to our patients.

Predictive analysis of postoperative death in patients with Stanford B acute aortic dissection by XGBoost model / 国际外科学杂志

Objective:To investigate the analysis of postoperative death in patients with Stanford B acute aortic dissection (AAD) by XGBoost model.Methods:A retrospective study was conducted on 226 patients with Stanford type B AAD diagnosed in Yunnan Wenshan People′s Hospital from February 2012 to June 2019, including 126 males and 100 females, with an average age of (61.24±4.25) years. According to the outcome of discharge, the patients were divided into survival group ( n=129) and death group ( n=97), in which those who automatically gave up treatment and left the hospital were regarded as the death group. If the patients were admitted to Yunnan Wenshan People′s Hospital for many times during the study period, only the clinical data diagnosed as Stanford B AAD for the first time were selected for the study. The clinical data and hematological indexes of the subjects were collected, and the XGBoost model was used to predict the rapid diagnosis of postoperative death in patients with Stanford B AAD, and compared with the traditional Logistic regression model. Results:In the XGBoost model, the influencing factors were ranked according to the degree of importance. The top 6 factors were hypertension, neutrophil-to-lymphocyte(NLR), C-reactive protein (CRP), white blood cell count(WBC), D-dimer and heart rate. Hypertension and NLR had the greatest influence on postoperative death in patients with Stanford B AAD. Using receiver operator charateristic curve to compare the prediction performance of the two models, it was found that the prediction efficiency of the XGBoost algorithm was significantly stronger than that of the Logistic regression model in the training set, while the two models were equivalent in the verification set. The prediction models constructed by the two methods eventually included independent variables such as hypertension, NLR, CRP, WBC, D-dimer, heart rate, systolic blood pressure, diastolic blood pressure, surgical treatment and so on.Conclusions:XGBoost model can be used to predict the postoperative death of patients with Stanford B AAD. Its diagnostic performance is better than Logistic regression model in training set and equivalent to the latter in verification set. Hypertension and NLR are the most important predictors of postoperative mortality in patients with Stanford B type AAD.

Prediction model of perioperative risk of deep venous thrombosis in patients with acute multiple knee joint injuries based on Logistic regression and XGBoost algorithm / 国际外科学杂志

Objective:Based on Logistic regression and XGBoost algorithm, the prediction model of perioperative risk of deep venous thrombosis in patients with acute multiple knee joint injuries was constructed, and the prediction performance was compared.Methods:A total of 120 patients with acute multiple injuries around the knee treated in the Department of Orthopaedic Trauma, Guangzhou Panyu District Central Hospital from January 2017 to June 2020 were retrospectively selected. According to the proportion of 7∶3, the patients were randomly divided into training set ( n=84) and test set ( n=36). The prediction models of Logistic regression and XGBoost algorithm were constructed by training set data, to screen the predictors of perioperative deep venous thrombosis in patients with acute multiple injury around knee joint, and the prediction effect of the model was evaluated by test set data. The measurement data conforming to the normal distribution were expressed as mean±standard deviation ( Mean± SD), and the independent t-test was used for comparison between groups; the measurement data of non-normal distribution were expressed as the median (interquartile range) [ M( P25, P75)], the independent sample Mann-Whitney U test was used for comparison between groups; the Chi-square test was used for comparison of enumeration data between groups. Results:The results of Logistic regression model showed that age, hypertension, coronary heart disease, time from injury to operation, D-dimer at 1 day after operation and multiple injuries were predictive factors for perioperative deep venous thrombosis in patients with acute multiple injuries around the knee joint. The top five important feature scores of XGBoost algorithm model were combined multiple injuries (35 points), time from injury to operation (28 points), age (24 points), coronary heart disease (21 points) and D-dimer 1 day after operation (16 points). In the training set, the area under the curve of the Logistic regression model was 0.805 (95% CI: 0.637-0.912), and χ2=1.436, P=0.329 for Hosmer and Lemeshow test. The area under the curve of the XGBoost algorithm model was 0.847(95% CI: 0.651-0.920), and χ2=1.103, P=0.976 for Hosmer and Lemeshow test. Conclusion:Logistic regression model and XGBoost algorithm model are similar in predicting perioperative deep venous thrombosis in patients with acute multiple injuries around the knee, and multiple injuries, time from injury to operation, age, coronary heart disease and D-dimer 1 day after operation can be used as predictive factors.

Prediction of plasma protein binding rate based on machine learning / 中国药科大学学报

@#Predicting the protein binding rate of drugs in plasma is helpful to us in understanding the pharmacokinetic characteristics of drugs, with much value of reference for early research on drug discovery. In this study, plasma protein binding rate information of 2 452 clinical drugs were collected.Two pieces of software, Molecular Operating Environment (MOE) and Mordred, were used to calculate molecular descriptors, which were used as input features of the model.Extreme gradient boosting (XGBoost) algorithm and random forest (RF) algorithm were then used to build a machine learning model.The results showed that, compared with MOE, the prediction performance of the constructed model was better using the molecular descriptor calculated by Mordred as the input of the model.The prediction performance results of the model constructed using the XGBoost algorithm and the RF algorithm were similar, and the R2 of the optimal model were both 0.715.According to the research results, it can be concluded that the drug plasma protein binding rate is closely related to some physical and chemical properties of the drug molecule, such as water solubility, octanol/water partition coefficient and conjugated double bonds.Using these parameters to predict the plasma protein binding rate of drugs has the advantages of convenience and efficiency, which can provide reference for related pharmacokinetic studies.

Prediction of drug-induced cell viability by SAE-XGBoost algorithm based on LINCS-L1000 perturbation signal / 生物工程学报

Different cell lines have different perturbation signals in response to specific compounds, and it is important to predict cell viability based on these perturbation signals and to uncover the drug sensitivity hidden underneath the phenotype. We developed an SAE-XGBoost cell viability prediction algorithm based on the LINCS-L1000 perturbation signal. By matching and screening three major dataset, LINCS-L1000, CTRP and Achilles, a stacked autoencoder deep neural network was used to extract the gene information. These information were combined with the RW-XGBoost algorithm to predict the cell viability under drug induction, and then to complete drug sensitivity inference on the NCI60 and CCLE datasets. The model achieved good results compared to other methods with a Pearson correlation coefficient of 0.85. It was further validated on an independent dataset, corresponding to a Pearson correlation coefficient of 0.68. The results indicate that the proposed method can help discover novel and effective anti-cancer drugs for precision medicine.

Researches on the illness risk of essential hypertension complicated with coronary heart disease based on machine learning algorithm / 解放军医学杂志

[Abstract] Objective To study a model of screening the risk factors of essential hypertension complicated with coronary heart disease and establishing the individual risk classification, and provide a computer-aided diagnostic methods for disease occurrence. Methods To collect 70 clinical information including 2791 patients with essential hypertension complicated with coronary heart disease and 2135 patients with simple essential hypertension diagnosed from January 1, 2014 to May 31, 2019 in Chongqing Medical University medical big data platform, screen out the indicators with statistical differences in single factor analysis. With R3.6.1 to construct logistic regression classification model and 3 machine learning models of BP neural network, random forest and extreme gradient rise (XGBoost), then compare the relevant parameters of various models and select the optimal classification model. Results According to the univariate analysis, 44 indexes with statistical difference were selected and included in logistic regression classification model and machine learning model. The classification accuracy in test set of logistic regression classification model, BP neural network model, random forest model, XGBoost model was 0.852, 0.968, 0.966 and 0.976, respectively, and the area under the work characteristic curve (AUC) of the subjects was 0.853, 0.970, 0.967 and 0.977, respectively. Applying XGBoost model with optimal performance to clinical practice of cardiology in the University Town Hospital of Chongqing Medical University. The diagnostic sensitivity was 1.000, specificity was 0.912, accuracy was 0.926, and AUC was 0.956. Conclusion Establishment of XGBoost model has a good auxiliary diagnostic function for essential hypertension complicated with coronary heart disease, and has achieved good results in clinical practice.

Prediction and feature selection for fatal gastrointestinal bleeding recurrence in hospital via machine learning / 中华危重病急救医学

Objective To propose a method of prediction for fatal gastrointestinal bleeding recurrence in hospital and a method of feature selection via machine learning models. Methods 728 digestive tract hemorrhage samples were extracted from the first aid database of PLA General Hospital, and 343 patients among them were diagnosed as fatal gastrointestinal bleeding recurrence in hospital. A total of 64 physiological or laboratory indicators were extracted and screened. Based on the ten-fold cross-validation, Logistic regression, AdaBoost and XGBoost were used for classification prediction and comparison. XGBoost was used to search sequence features, and the key indicators for predicting fatal gastrointestinal bleeding recurrence in hospital were screened out according to the importance of the indicators during training. Results Logistic regression, AdaBoost and XGBoost all get better F1.5 score under each feature input dimension, among which XGBoost had the best effect and the highest score, which was able to identify as many patients as possible who might have fatal gastrointestinal bleeding recurrence in hospital. Through XGBoost iteration results, the Top 30 indicators with high importance for predicting fatal gastrointestinal bleeding recurrence in hospital were ranked. The F1.5 scores of the first 12 key indicators peaked at iteration (0.893), including hemoglobin (Hb), calcium (CA), red blood cell count (RBC), mean platelet volume (MPV), mean erythrocyte hemoglobin concentration (MCH), systolic blood pressure (SBP), platelet count (PLT), magnesium (MG), lymphocyte (LYM), glucose (GLU, blood gas analysis), glucose (GLU, blood biochemistry) and diastolic blood pressure (DBP). Conclusions Logistic regression, AdaBoost and XGBoost could achieve the purpose of early warning for predicting fatal gastrointestinal bleeding recurrence in hospital, and XGBoost is the most suitable. The 12 most important indicators were screened out by sequential forward selection.

Identification of Curcuma herbs using XGBoost algorithm in electronic nose odor fingerprint / 中国中药杂志

This article aims to identify four commonly applied herbs from Curcuma genus of Zingiberaceae family,namely Curcumae Radix( Yujin),Curcumae Rhizoma( Ezhu),Curcumae Longae Rhizoma( Jianghuang) and Wenyujin Rhizoma Concisum( Pianjianghuang). The odor fingerprints of those four herbal medicines were collected by electronic nose,respectively. Meanwhile,XGBoost algorithm was introduced to data analysis and discriminant model establishment,with four indexes for performance evaluation,including accuracy,precision,recall,and F-measure. The discriminant model was established by XGBoost with positive rate of returning to 166 samples in the training set and 69 samples in the test set were 99. 39% and 95. 65%,respectively. The top four of the contribution to the discriminant model were LY2/g CT,P40/1,LY2/Gh and LY2/LG,the least contributing sensor was T70/2. Compared with support vector machine,random forest and artificial neural network,XGBoost algorithms shows better identification capacity with higher recognition efficiency. The accuracy,precision,recall and F-measure of the XGBoost discriminant model forecast set were 95. 65%,95. 25%,93. 07%,93. 75%,respectively. The superiority of XGBoost in the identification of Curcuma herbs was verified. Obviously,this new method could not only be suitable for digitization and objectification of traditional Chinese medicine( TCM) odor indicators,but also achieve the identification of different TCM based on their odor fingerprint in electronic nose system. The introduction of XGBoost algorithm and more excellent algorithms provide more ideas for the application of electronic nose in data mining for TCM studies.