A Semi-Supervised Learning Framework for Classifying Colorectal Neoplasia Based on the NICE Classification.

Labelling medical images is an arduous and costly task that necessitates clinical expertise and large numbers of qualified images. Insufficient samples can lead to underfitting during training and poor performance of supervised learning models. In this study, we aim to develop a SimCLR-based semi-supervised learning framework to classify colorectal neoplasia based on the NICE classification. First, the proposed framework was trained under self-supervised learning using a large unlabelled dataset; subsequently, it was fine-tuned on a limited labelled dataset based on the NICE classification. The model was evaluated on an independent dataset and compared with models based on supervised transfer learning and endoscopists using accuracy, Matthew's correlation coefficient (MCC), and Cohen's kappa. Finally, Grad-CAM and t-SNE were applied to visualize the models' interpretations. A ResNet-backboned SimCLR model (accuracy of 0.908, MCC of 0.862, and Cohen's kappa of 0.896) outperformed supervised transfer learning-based models (means: 0.803, 0.698, and 0.742) and junior endoscopists (0.816, 0.724, and 0.863), while performing only slightly worse than senior endoscopists (0.916, 0.875, and 0.944). Moreover, t-SNE showed a better clustering of ternary samples through self-supervised learning in SimCLR than through supervised transfer learning. Compared with traditional supervised learning, semi-supervised learning enables deep learning models to achieve improved performance with limited labelled endoscopic images.

The development and validation of automated machine learning models for predicting lymph node metastasis in Siewert type II T1 adenocarcinoma of the esophagogastric junction.

Background: Lymph node metastasis (LNM) is considered an essential prognosis factor for adenocarcinoma of the esophagogastric junction (AEG), which also affects the treatment strategies of AEG. We aimed to evaluate automated machine learning (AutoML) algorithms for predicting LNM in Siewert type II T1 AEG. Methods: A total of 878 patients with Siewert type II T1 AEG were selected from the Surveillance, Epidemiology, and End Results (SEER) database to develop the LNM predictive models. The patients from two hospitals in Suzhou were collected as the test set. We applied five machine learning algorithms to develop the LNM prediction models. The performance of predictive models was assessed using various metrics including accuracy, sensitivity, specificity, the area under the curve (AUC), and receiver operating characteristic (ROC) curve. Results: Patients with LNM exhibited a higher proportion of male individuals, a poor degree of differentiation, and submucosal infiltration, with statistical differences. The deep learning (DL) model demonstrated relatively good accuracy (0.713) and sensitivity (0.868) among the five models. Moreover, the DL model achieved the highest AUC (0.781) and sensitivity (1.000) in the test set. Conclusion: The DL model showed good predictive performance among five AutoML models, indicating the advantage of AutoML in modeling LNM prediction in patients with Siewert type II T1 AEG.

Association of the fat mass index with hepatic steatosis and fibrosis: evidence from NHANES 2017-2018.

Limited population-based studies discuss the association between fat mass index (FMI) and the risk of liver diseases. This investigation utilized data from the National Health and Nutrition Examination Survey (NHANES) to examine the linkage between the FMI and liver conditions, specifically steatosis and fibrosis. The study leveraged data from NHANES's 2017-2018 cross-sectional study, employing an oversampling technique to deal with sample imbalance. Hepatic steatosis and fibrosis were identified by vibration-controlled transient elastography. Receiver operating curve was used to assess the relationship of anthropometric indicators, e.g., the FMI, body mass index (BMI), weight-adjusted-waist index (WWI), percentage of body fat (BF%), waist-to-hip ratio (WHR), and appendicular skeletal muscle index (ASMI), with hepatic steatosis and fibrosis. In this study, which included 2260 participants, multivariate logistic regression models, stratified analyses, restricted cubic spline (RCS), and sharp regression discontinuity analyses were utilized. The results indicated that the WHR and the FMI achieved the highest area under the curve for identifying hepatic steatosis and fibrosis, respectively (0.720 and 0.726). Notably, the FMI presented the highest adjusted odds ratio for both hepatic steatosis (6.40 [4.91-8.38], p = 2.34e-42) and fibrosis (6.06 [5.00, 7.37], p = 5.88e-74). Additionally, potential interaction effects were observed between the FMI and variables such as the family income-to-poverty ratio, smoking status, and hypertension, all of which correlated with the presence of liver fibrosis (p for interaction < 0.05). The RCS models further confirmed a significant positive correlation of the FMI with the controlled attenuation parameter and liver stiffness measurements. Overall, the findings underscore the strong link between the FMI and liver conditions, proposing the FMI as a potential straightforward marker for identifying liver diseases.

Development and Validation of Multimodal Models to Predict the 30-Day Mortality of ICU Patients Based on Clinical Parameters and Chest X-Rays.

We aimed to develop and validate multimodal ICU patient prognosis models that combine clinical parameters data and chest X-ray (CXR) images. A total of 3798 subjects with clinical parameters and CXR images were extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and an external hospital (the test set). The primary outcome was 30-day mortality after ICU admission. Automated machine learning (AutoML) and convolutional neural networks (CNNs) were used to construct single-modal models based on clinical parameters and CXR separately. An early fusion approach was used to integrate both modalities (clinical parameters and CXR) into a multimodal model named PrismICU. Compared to the single-modal models, i.e., the clinical parameter model (AUC = 0.80, F1-score = 0.43) and the CXR model (AUC = 0.76, F1-score = 0.45) and the scoring system APACHE II (AUC = 0.83, F1-score = 0.77), PrismICU (AUC = 0.95, F1 score = 0.95) showed improved performance in predicting the 30-day mortality in the validation set. In the test set, PrismICU (AUC = 0.82, F1-score = 0.61) was also better than the clinical parameters model (AUC = 0.72, F1-score = 0.50), CXR model (AUC = 0.71, F1-score = 0.36), and APACHE II (AUC = 0.62, F1-score = 0.50). PrismICU, which integrated clinical parameters data and CXR images, performed better than single-modal models and the existing scoring system. It supports the potential of multimodal models based on structured data and imaging in clinical management.

Few-shot learning for the classification of intestinal tuberculosis and Crohn's disease on endoscopic images: A novel learn-to-learn framework.

Background and aim: Standard deep learning methods have been found inadequate in distinguishing between intestinal tuberculosis (ITB) and Crohn's disease (CD), a shortcoming largely attributed to the scarcity of available samples. In light of this limitation, our objective is to develop an innovative few-shot learning (FSL) system, specifically tailored for the efficient categorization and differential diagnosis of CD and ITB, using endoscopic image data with minimal sample requirements. Methods: A total of 122 white-light endoscopic images (99 CD images and 23 ITB images) were collected (one ileum image from each patient). A 2-way, 3-shot FSL model that integrated dual transfer learning and metric learning strategies was devised. Xception architecture was selected as the foundation and then underwent a dual transfer process utilizing oesophagitis images sourced from HyperKvasir. Subsequently, the eigenvectors derived from the Xception for each query image were converted into predictive scores, which were calculated using the Euclidean distances to six reference images from the support sets. Results: The FSL model, which leverages dual transfer learning, exhibited enhanced performance metrics (AUC 0.81) compared to a model relying on single transfer learning (AUC 0.56) across three evaluation rounds. Additionally, its performance surpassed that of a less experienced endoscopist (AUC 0.56) and even a more seasoned specialist (AUC 0.61). Conclusions: The FSL model we have developed demonstrates efficacy in distinguishing between CD and ITB using a limited dataset of endoscopic imagery. FSL holds value for enhancing the diagnostic capabilities of rare conditions.

Automated machine learning for the identification of asymptomatic COVID-19 carriers based on chest CT images.

BACKGROUND: Asymptomatic COVID-19 carriers with normal chest computed tomography (CT) scans have perpetuated the ongoing pandemic of this disease. This retrospective study aimed to use automated machine learning (AutoML) to develop a prediction model based on CT characteristics for the identification of asymptomatic carriers. METHODS: Asymptomatic carriers were from Yangzhou Third People's Hospital from August 1st, 2020, to March 31st, 2021, and the control group included a healthy population from a nonepizootic area with two negative RT‒PCR results within 48 h. All CT images were preprocessed using MATLAB. Model development and validation were conducted in R with the H2O package. The models were built based on six algorithms, e.g., random forest and deep neural network (DNN), and a training set (n = 691). The models were improved by automatically adjusting hyperparameters for an internal validation set (n = 306). The performance of the obtained models was evaluated based on a dataset from Suzhou (n = 178) using the area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 score. RESULTS: A total of 1,175 images were preprocessed with high stability. Six models were developed, and the performance of the DNN model ranked first, with an AUC value of 0.898 for the test set. The sensitivity, specificity, PPV, NPV, F1 score and accuracy of the DNN model were 0.820, 0.854, 0.849, 0.826, 0.834 and 0.837, respectively. A plot of a local interpretable model-agnostic explanation demonstrated how different variables worked in identifying asymptomatic carriers. CONCLUSIONS: Our study demonstrates that AutoML models based on CT images can be used to identify asymptomatic carriers. The most promising model for clinical implementation is the DNN-algorithm-based model.

Automated machine learning for early prediction of acute kidney injury in acute pancreatitis.

BACKGROUND: Acute kidney injury (AKI) represents a frequent and grave complication associated with acute pancreatitis (AP), substantially elevating both mortality rates and the financial burden of hospitalization. The aim of our study is to construct a predictive model utilizing automated machine learning (AutoML) algorithms for the early prediction of AKI in patients with AP. METHODS: We retrospectively analyzed patients who were diagnosed with AP in our hospital from January 2017 to December 2021. These patients were randomly allocated into a training set and a validation set at a ratio of 7:3. To develop predictive models for each set, we employed the least absolute shrinkage and selection operator (LASSO) algorithm along with AutoML. A nomogram was developed based on multivariate logistic regression analysis outcomes. The model's efficacy was assessed using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Additionally, the performance of the model constructed via AutoML was evaluated using decision curve analysis (DCA), feature importance, SHapley Additive exPlanations (SHAP) plots, and locally interpretable model-agnostic explanations (LIME). RESULTS: This study incorporated a total of 437 patients who met the inclusion criteria. Out of these, 313 were assigned to the training cohort and 124 to the validation cohort. In the training and validation cohorts, AKI occurred in 68 (21.7%) and 29(23.4%) patients, respectively. Comparative analysis revealed that the AutoML models exhibited enhanced performance over traditional logistic regression (LR). Furthermore, the deep learning (DL) model demonstrated superior predictive accuracy, evidenced by an area under the ROC curve of 0.963 in the training set and 0.830 in the validation set, surpassing other comparative models. The key variables identified as significant in the DL model within the training dataset included creatinine (Cr), urea (Urea), international normalized ratio (INR), etiology, smoking, alanine aminotransferase (ALT), hypertension, prothrombin time (PT), lactate dehydrogenase (LDH), and diabetes. CONCLUSION: The AutoML model, utilizing DL algorithm, offers considerable clinical significance in the early detection of AKI among patients with AP.

Development and validation of a multimodal model in predicting severe acute pancreatitis based on radiomics and deep learning.

OBJECTIVE: Aim to establish a multimodal model for predicting severe acute pancreatitis (SAP) using machine learning (ML) and deep learning (DL). METHODS: In this multicentre retrospective study, patients diagnosed with acute pancreatitis at admission were enrolled from January 2017 to December 2021. Clinical information within 24 h and CT scans within 72 h of admission were collected. First, we trained Model α based on clinical features selected by least absolute shrinkage and selection operator analysis. Second, radiomics features were extracted from 3D-CT scans and Model ß was developed on the features after dimensionality reduction using principal component analysis. Third, Model γ was trained on 2D-CT images. Lastly, a multimodal model, namely PrismSAP, was constructed based on aforementioned features in the training set. The predictive accuracy of PrismSAP was verified in the validation and internal test sets and further validated in the external test set. Model performance was evaluated using area under the curve (AUC), accuracy, sensitivity, specificity, recall, precision and F1-score. RESULTS: A total of 1,221 eligible patients were randomly split into a training set (n = 864), a validation set (n = 209) and an internal test set (n = 148). Data of 266 patients were for external testing. In the external test set, PrismSAP performed best with the highest AUC of 0.916 (0.873-0.960) among all models [Model α: 0.709 (0.618-0.800); Model ß: 0.749 (0.675-0.824); Model γ: 0.687 (0.592-0.782); MCTSI: 0.778 (0.698-0.857); RANSON: 0.642 (0.559-0.725); BISAP: 0.751 (0.668-0.833); SABP: 0.710 (0.621-0.798)]. CONCLUSION: The proposed multimodal model outperformed any single-modality models and traditional scoring systems.

Deep learning to predict esophageal variceal bleeding based on endoscopic images.

OBJECTIVE: Esophageal varix (EV) bleeding is a particularly serious complications of cirrhosis. Prediction of EV bleeding requires extensive endoscopy experience; it remains unreliable and inefficient. This retrospective cohort study evaluated the feasibility of using deep learning (DL) to predict the 12-month risk of EV bleeding based on endoscopic images. METHODS: Six DL models were trained to perform binary classification of endoscopic images of EV bleeding. The models were subsequently validated using an external test dataset, then compared with classifications performed by two endoscopists. RESULTS: In the validation dataset, EfficientNet had the highest accuracy (0.910), followed by ConvMixer (0.898) and Xception (0.875). In the test dataset, EfficientNet maintained the highest accuracy (0.893), which was better than the endoscopists (0.800 and 0.763). Notably, one endoscopist displayed higher recall (0.905), compared with EfficientNet (0.870). When their predictions were assisted by artificial intelligence, the accuracies of the two endoscopists increased by 17.3% and 19.0%. Moreover, statistical agreement among the models was dependent on model architecture. CONCLUSIONS: This study demonstrated the feasibility of using DL to predict the 12-month risk of EV bleeding based on endoscopic images. The findings suggest that artificial intelligence-aided diagnosis will be a useful addition to cirrhosis management.

Public Imaging Datasets of Gastrointestinal Endoscopy for Artificial Intelligence: a Review.

With the advances in endoscopic technologies and artificial intelligence, a large number of endoscopic imaging datasets have been made public to researchers around the world. This study aims to review and introduce these datasets. An extensive literature search was conducted to identify appropriate datasets in PubMed, and other targeted searches were conducted in GitHub, Kaggle, and Simula to identify datasets directly. We provided a brief introduction to each dataset and evaluated the characteristics of the datasets included. Moreover, two national datasets in progress were discussed. A total of 40 datasets of endoscopic images were included, of which 34 were accessible for use. Basic and detailed information on each dataset was reported. Of all the datasets, 16 focus on polyps, and 6 focus on small bowel lesions. Most datasets (n = 16) were constructed by colonoscopy only, followed by normal gastrointestinal endoscopy and capsule endoscopy (n = 9). This review may facilitate the usage of public dataset resources in endoscopic research.

A nomogram based on platelet-to-lymphocyte ratio for predicting lymph node metastasis in patients with early gastric cancer.

Background: Preoperative assessment of the presence of lymph node metastasis (LNM) in patients with early gastric cancer (EGC) remains difficult. We aimed to develop a practical prediction model based on preoperative pathological data and inflammatory or nutrition-related indicators. Methods: This study retrospectively analyzed the clinicopathological characteristics of 1,061 patients with EGC who were randomly divided into the training set and validation set at a ratio of 7:3. In the training set, we introduced the least absolute selection and shrinkage operator (LASSO) algorithm and multivariate logistic regression to identify independent risk factors and construct the nomogram. Both internal validation and external validation were performed by the area under the receiver operating characteristic curve (AUC), C-index, calibration curve, and decision curve analysis (DCA). Results: LNM occurred in 162 of 1,061 patients, and the rate of LNM was 15.27%. In the training set, four variables proved to be independent risk factors (p < 0.05) and were incorporated into the final model, including depth of invasion, tumor size, degree of differentiation, and platelet-to-lymphocyte ratio (PLR). The AUC values were 0.775 and 0.792 for the training and validation groups, respectively. Both calibration curves showed great consistency in the predictive and actual values. The Hosmer-Lemeshow (H-L) test was carried out in two cohorts, showing excellent performance with p-value >0.05 (0.684422, 0.7403046). Decision curve analysis demonstrated a good clinical benefit in the respective set. Conclusion: We established a preoperative nomogram including depth of invasion, tumor size, degree of differentiation, and PLR to predict LNM in EGC patients and achieved a good performance.

Development of a scoring system for predicting primary resistance to venetoclax plus hypomethylating agents (HMAs) in acute myeloid leukemia patients.

In recent years, one of the most promising advances in the treatment of acute myeloid leukemia (AML) is the combination of a hypomethylating agent (HMA) with the BCL2 inhibitor venetoclax (VEN). To better understand the key factors associated with the response of VEN plus HMA, 212 consecutive AML patients were retrospectively recruited to establish and validate a scoring system for predicting the primary resistance to VEN-based induced therapy. All AML patients were divided randomly into a training set (n = 155) and a validation set (n = 57). Factors were selected using a multivariate logistic regression model, including FAB-M5, myelodysplastic syndrome-secondary acute myeloid leukemia (MDS-sAML), RUNX1-RUNX1T1 and FLT3-ITD mutation (FLT3-ITDm). A nomogram was then constructed including all these four predictors. The nomogram both presented a good performance of discrimination and calibration, with a C-index of 0.770 and 0.733 in the training and validation set. Decision curve analysis also indicated that the nomogram was feasible to make beneficial decisions. Eventually a total scoring system of 8 points was developed, which was divided into three risk groups: low-risk (score 0-2), medium-risk (score 3-4), and high-risk (score 5-8). There was a significant difference in the nonremission (NR) rate of these three risk groups (22.8% vs. 60.0% vs. 77.8%, p < 0.001). After adjustment of the other variables, patients in medium- or high-risk groups also presented a worse event-free survival (EFS) than that in the low-risk group (hazard ratio [HR] = 1.62, p = 0.03). In conclusion, we highlighted the response determinants of AML patients receiving a combination therapy of VEN plus HMAs. The scoring system can be used to predict the resistance of VEN, providing better guidance for clinical treatment.

Application Value of the Automated Machine Learning Model Based on Modified Computed Tomography Severity Index Combined With Serological Indicators in the Early Prediction of Severe Acute Pancreatitis.

BACKGROUND AND AIMS: Machine learning (ML) algorithms are widely applied in building models of medicine due to their powerful studying and generalizing ability. To assess the value of the Modified Computed Tomography Severity Index (MCTSI) combined with serological indicators for early prediction of severe acute pancreatitis (SAP) by automated ML (AutoML). PATIENTS AND METHODS: The clinical data, of the patients with acute pancreatitis (AP) hospitalized in Hospital 1 and hospital 2 from January 2017 to December 2021, were retrospectively analyzed. Serological indicators within 24 hours of admission were collected. MCTSI score was completed by noncontrast computed tomography within 24 hours of admission. Data from the hospital 1 were adopted for training, and data from the hospital 2 were adopted for external validation. The diagnosis of AP and SAP was based on the 2012 revised Atlanta classification of AP. Models were built using traditional logistic regression and AutoML analysis with 4 types of algorithms. The performance of models was evaluated by the receiver operating characteristic curve, the calibration curve, and the decision curve analysis based on logistic regression and decision curve analysis, feature importance, SHapley Additive exPlanation Plot, and Local Interpretable Model Agnostic Explanation based on AutoML. RESULTS: A total of 499 patients were used to develop the models in the training data set. An independent data set of 201 patients was used to test the models. The model developed by the Deep Neural Net (DL) outperformed other models with an area under the receiver operating characteristic curve (areas under the curve) of 0.907 in the test set. Furthermore, among these AutoML models, the DL and gradient boosting machine models achieved the highest sensitivity values, both exceeding 0.800. CONCLUSION: The AutoML model based on the MCTSI score combined with serological indicators has good predictive value for SAP in the early stage.

Identification of gastric signet ring cell carcinoma based on endoscopic images using few-shot learning.

BACKGROUND: Deep learning (DL) models perform poorly when there are limited gastric signet ring cell carcinoma (SRCC) samples. Few-shot learning (FSL) can address the small sample problem. METHODS: EfficientNetV2-S was first pretrained on ImageNet and then pretrained on esophageal endoscopic images (i.e., base classes: normal vs. early cancer vs. advanced cancer) using transfer learning. Second, images of gastric benign ulcers, adenocarcinoma and SRCC, i.e., novel classes (n = 50 per class), were included. Image features were extracted as vectors using the dual pretrained EfficientNetV2-S. Finally, a k-nearest neighbor classifier was used to identify SRCC. The above proposed 3-way 3-shot FSL framework was conducted in three rounds. RESULTS: Dual pretrained FSL performed better than single pretrained FSL, endoscopists and traditional EfficientNetV2-S models. Dual pretrained FSL obtained the highest accuracy (79.4%), sensitivity (68.8%), recall (68.8%), precision (69.3%) and F1-score (0.691), and the senior endoscopist achieved the highest specificity of 93.6% when identifying SRCC. The macro-AUC and F1-score for dual pretraining (0.763 and 0.703, respectively) were higher than those for single pretraining (0.656 and 0.537, respectively), along with endoscopists and traditional EfficientNetV2-S models. The 2-way 3-shot FSL also performed better. CONCLUSION: The proposed FSL framework showed practical performance in the differentiation of SRCC on endoscopic images, suggesting the potential of FSL in the computer-aided diagnosis for rare diseases.

[Application of machine learning model based on XGBoost algorithm in early prediction of patients with acute severe pancreatitis].

OBJECTIVE: To establish a machine learning model based on extreme gradient boosting (XGBoost) algorithm for early prediction of severe acute pancreatitis (SAP), and explore its predictive efficiency. METHODS: A retrospective cohort study was conducted. The patients with acute pancreatitis (AP) who admitted to the First Affiliated Hospital of Soochow University, the Second Affiliated Hospital of Soochow University and Changshu Hospital Affiliated to Soochow University from January 1, 2020 to December 31, 2021 were enrolled. Demography information, etiology, past history, and clinical indicators and imaging data within 48 hours of admission were collected according to the medical record system and image system, and the modified CT severity index (MCTSI), Ranson score, bedside index for severity in acute pancreatitis (BISAP) and acute pancreatitis risk score (SABP) were calculated. The data sets of the First Affiliated Hospital of Soochow University and Changshu Hospital Affiliated to Soochow University were randomly divided into training set and validation set according to 8 : 2. Based on XGBoost algorithm, the SAP prediction model was constructed on the basis of hyperparameter adjustment by 5-fold cross validation and loss function. The data set of the Second Affiliated Hospital of Soochow University was served as independent test set. The predictive efficacy of the XGBoost model was evaluated by drawing the receiver operator characteristic curve (ROC curve), and compared it with the traditional AP related severity score; variable importance ranking diagram and Shapley additive explanation (SHAP) diagram were drawn to visually explain the model. RESULTS: A total of 1 183 AP patients were enrolled finally, of which 129 (10.9%) developed SAP. Among the patients from the First Affiliated Hospital of Soochow University and Changshu Hospital Affiliated to Soochow University, there were 786 patients in the training set and 197 in the validation set; 200 patients from the Second Affiliated Hospital of Soochow University were used as the test set. Analysis of all three datasets showed that patients who advanced to SAP exhibited pathological manifestation such as abnormal respiratory function, coagulation function, liver and kidney function, and lipid metabolism. Based on the XGBoost algorithm, an SAP prediction model was constructed, and ROC curve analysis showed that the accuracy for prediction of SAP reached 0.830, the area under the ROC curve (AUC) was 0.927, which was significantly improved compared with the traditional scoring systems including MCTSI, Ranson, BISAP and SABP, the accuracy was 0.610, 0.690, 0.763, 0.625, and the AUC was 0.689, 0.631, 0.875, and 0.770, respectively. The feature importance analysis based on the XGBoost model showed that the top ten items ranked by the importance of model features were admission pleural effusion (0.119), albumin (Alb, 0.049), triglycerides (TG, 0.036), Ca2+ (0.034), prothrombin time (PT, 0.031), systemic inflammatory response syndrome (SIRS, 0.031), C-reactive protein (CRP, 0.031), platelet count (PLT, 0.030), lactate dehydrogenase (LDH, 0.029), and alkaline phosphatase (ALP, 0.028). The above indicators were of great significance for the XGBoost model to predict SAP. The SHAP contribution analysis based on the XGBoost model showed that the risk of SAP increased significantly when patients had pleural effusion and decreased Alb. CONCLUSIONS: A SAP prediction scoring system was established based on the machine automatic learning XGBoost algorithm, which can predict the SAP risk of patients within 48 hours of admission with good accuracy.

Predicting the Recurrence of Common Bile Duct Stones After ERCP Treatment with Automated Machine Learning Algorithms.

BACKGROUND: Recurrence of common bile duct stones (CBDs) commonly happens after endoscopic retrograde cholangiopancreatography (ERCP). The clinical prediction models for the recurrence of CBDs after ERCP are lacking. AIMS: We aim to develop high-performance prediction models for the recurrence of CBDS after ERCP treatment using automated machine learning (AutoML) and to assess the AutoML models versus the traditional regression models. METHODS: 473 patients with CBDs undergoing ERCP were recruited in the single-center retrospective cohort study. Samples were divided into Training Set (65%) and Validation Set (35%) randomly. Three modeling approaches, including fully automated machine learning (Fully automated), semi-automated machine learning (Semi-automated), and traditional regression were applied to fit prediction models. Models' discrimination, calibration, and clinical benefits were examined. The Shapley additive explanations (SHAP), partial dependence plot (PDP), and SHAP local explanation (SHAPLE) were proposed for the interpretation of the best model. RESULTS: The area under roc curve (AUROC) of semi-automated gradient boost machine (GBM) model was 0.749 in Validation Set, better than the other fully/semi-automated models and the traditional regression models (highest AUROC = 0.736). The calibration and clinical application of AutoML models were adequate. Through the SHAP-PDP-SHAPLE pipeline, the roles of key variables of the semi-automated GBM model were visualized. Lastly, the best model was deployed online for clinical practitioners. CONCLUSION: The GBM model based on semi-AutoML is an optimal model to predict the recurrence of CBDs after ERCP treatment. In comparison with traditional regressions, AutoML algorithms present significant strengths in modeling, which show promise in future clinical practices.

Deep learning for pancreatic diseases based on endoscopic ultrasound: A systematic review.

BACKGROUND AND AIMS: Endoscopic ultrasonography (EUS) is one of the main examinations in pancreatic diseases. A series of the studies reported the application of deep learning (DL)-assisted EUS in the diagnosis of pancreatic diseases. This systematic review is to evaluate the role of DL algorithms in assisting EUS diagnosis of pancreatic diseases. METHODS: Literature search were conducted in PubMed and Semantic Scholar databases. Studies that developed DL models for pancreatic diseases based on EUS were eligible for inclusion. This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines and quality assessment of the included studies was performed according to the IJMEDI checklist. RESULTS: A total of 23 studies were enrolled into this systematic review, which could be categorized into three groups according to computer vision tasks: classification, detection and segmentation. Seventeen studies focused on the classification task, among which five studies developed simple neural network (NN) models while twelve studies constructed convolutional NN (CNN) models. Three studies were concerned the detection task and five studies were the segmentation task, all based on CNN architectures. All models presented in the studies performed well based on EUS images, videos or voice. According to the IJMEDI checklist, six studies were recognized as high-grade quality, with scores beyond 35 points. CONCLUSIONS: DL algorithms show great potential in EUS images/videos/voice for pancreatic diseases. However, there is room for improvement such as sample sizes, multi-center cooperation, data preprocessing, model interpretability, and code sharing.

Identification of Asymptomatic COVID-19 Patients on Chest CT Images Using Transformer-Based or Convolutional Neural Network-Based Deep Learning Models.

Novel coronavirus disease 2019 (COVID-19) has rapidly spread throughout the world; however, it is difficult for clinicians to make early diagnoses. This study is to evaluate the feasibility of using deep learning (DL) models to identify asymptomatic COVID-19 patients based on chest CT images. In this retrospective study, six DL models (Xception, NASNet, ResNet, EfficientNet, ViT, and Swin), based on convolutional neural networks (CNNs) or transformer architectures, were trained to identify asymptomatic patients with COVID-19 on chest CT images. Data from Yangzhou were randomly split into a training set (n = 2140) and an internal-validation set (n = 360). Data from Suzhou was the external-test set (n = 200). Model performance was assessed by the metrics accuracy, recall, and specificity and was compared with the assessments of two radiologists. A total of 2700 chest CT images were collected in this study. In the validation dataset, the Swin model achieved the highest accuracy of 0.994, followed by the EfficientNet model (0.954). The recall and the precision of the Swin model were 0.989 and 1.000, respectively. In the test dataset, the Swin model was still the best and achieved the highest accuracy (0.980). All the DL models performed remarkably better than the two experts. Last, the time on the test set diagnosis spent by two experts-42 min, 17 s (junior); and 29 min, 43 s (senior)-was significantly higher than those of the DL models (all below 2 min). This study evaluated the feasibility of multiple DL models in distinguishing asymptomatic patients with COVID-19 from healthy subjects on chest CT images. It found that a transformer-based model, the Swin model, performed best.

Intestinal Barrier Function in the Pathogenesis of Nonalcoholic Fatty Liver Disease.

Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease worldwide. The mechanisms involved in NAFLD onset are complicated and multifactorial. Recent literature has indicated that altered intestinal barrier function is related to the occurrence and progression of liver disease. The intestinal barrier is important for absorbing nutrients and electrolytes and for defending against toxins and antigens in the enteric environment. Major mechanisms by which the intestinal barrier influences the development of NAFLD involve the altered epithelial layer, decreased intracellular junction integrity, and increased intestinal barrier permeability. Increased intestinal permeability leads to luminal dysbiosis and allows the translocation of pathogenic bacteria and metabolites into the liver, inducing inflammation, immune response, and hepatocyte injury in NAFLD. Although research has been directed to NAFLD in recent decades, the pathophysiological changes in NAFLD initiation and progression are still not completely understood, and the therapeutic targets remain limited. A deeper understanding on the correlation between NAFLD pathogenesis and intestinal barrier regulation must be attained. Therefore, in this review, the components of the intestinal barrier and their respective functions and disruptions during the progression of NAFLD are discussed.

Automated Multimodal Machine Learning for Esophageal Variceal Bleeding Prediction Based on Endoscopy and Structured Data.

Esophageal variceal (EV) bleeding is a severe medical emergency related to cirrhosis. Early identification of cirrhotic patients with at a high risk of EV bleeding is key to improving outcomes and optimizing medical resources. This study aimed to evaluate the feasibility of automated multimodal machine learning (MMML) for predicting EV bleeding by integrating endoscopic images and clinical structured data. This study mainly includes three steps: step 1, developing deep learning (DL) models using EV images by 12-month bleeding on TensorFlow (backbones include ResNet, Xception, EfficientNet, ViT and ConvMixer); step 2, training and internally validating MMML models integrating clinical structured data and DL model outputs to predict 12-month EV bleeding on an H2O-automated machine learning platform (algorithms include DL, XGBoost, GLM, GBM, RF, and stacking); and step 3, externally testing MMML models. Furthermore, existing clinical indices, e.g., the MELD score, Child‒Pugh score, APRI, and FIB-4, were also examined. Five DL models were transfer learning to the binary classification of EV endoscopic images at admission based on the occurrence or absence of bleeding events during the 12-month follow-up. An EfficientNet model achieved the highest accuracy of 0.868 in the validation set. Then, a series of MMML models, integrating clinical structured data and the output of the EfficientNet model, were automatedly trained to predict 12-month EV bleeding. A stacking model showed the highest accuracy (0.932), sensitivity (0.952), and F1-score (0.879) in the test dataset, which was also better than the existing indices. This study is the first to evaluate the feasibility of automated MMML in predicting 12-month EV bleeding based on endoscopic images and clinical variables.