Objective To develop and compare the predictive performance of five machine learning models for adverse postoperative outcomes in cardiac surgery patients, and to identify key decision factors through SHapley Additive exPlanations (SHAP) interpretability analysis. Methods A retrospective collection of perioperative data (including demographic information, preoperative, intraoperative, and postoperative indicators) with 88 variables was conducted from adult cardiac surgery patients at the First Affiliated Hospital of Xinjiang Medical University in 2023. Adverse postoperative outcomes were defined as the occurrence of acute kidney injury and/or in-hospital mortality during the postoperative hospitalization period following cardiac surgery. Patients were divided into an adverse outcome group and a favorable outcome group based on the presence of adverse postoperative outcomes. After screening feature variables using the least absolute shrinkage and selection operator (LASSO) regression method, five machine learning models were constructed: eXtreme gradient boosting (XGBoost), random forest (RF), gradient boosting machine (GBM), light gradient boosting machine (LightGBM), and generalized linear model (GLM). The dataset was randomly divided into a training set and a test set at a 7 : 3 ratio using stratified sampling, with postoperative outcome as the stratification factor. Model performance was evaluated using receiver operating characteristic curves, decision curve analysis, and F1 Score. The SHAP method was applied to analyze feature contribution. Results A total of 639 patients were included, comprising 395 males and 244 females, with a median age of 62 (55, 69) years. The adverse outcome group consisted of 191 patients, while the favorable outcome group included 448 patients, resulting in an adverse postoperative outcome incidence of 29.9%. Univariate analysis showed no significant differences between the two groups for any variables (P>0.05). Using LASSO regression, 16 feature variables were selected (including cardiopulmonary bypass support time, blood glucose on postoperative day 3, creatine kinase-MB isoenzyme, systemic inflammatory response index, etc.), and five machine learning models (GLM, RF, GBM, LightGBM, XGBoost) were constructed. Evaluation results demonstrated that the XGBoost model exhibited the best predictive performance on both the training set (n=447) and test set (n=192), with area under the curve values of 0.761 [95%CI (0.719, 0.800) ] and 0.759 [95%CI (0.692, 0.818) ], respectively. It also significantly outperformed other models in positive predictive value, and balanced accuracy in the test set. Decision curve analysis further confirmed its clinical utility across various risk thresholds. SHAP analysis indicated that variables such as cardiopulmonary bypass support time, blood glucose on postoperative day 3, creatine kinase-MB isoenzyme, and inflammatory markers (SIRI, NLR, CAR) had high contributions to the prediction. Conclusion The XGBoost model effectively predicts adverse postoperative outcomes in cardiac surgery patients. Clinically, attention should be focused on cardiopulmonary bypass support time, postoperative blood glucose control, and monitoring of inflammatory levels to improve patient prognosis.
Image fusion currently plays an important role in the diagnosis of prostate cancer (PCa). Selecting and developing a good image fusion algorithm is the core task of achieving image fusion, which determines whether the fusion image obtained is of good quality and can meet the actual needs of clinical application. In recent years, it has become one of the research hotspots of medical image fusion. In order to make a comprehensive study on the methods of medical image fusion, this paper reviewed the relevant literature published at home and abroad in recent years. Image fusion technologies were classified, and image fusion algorithms were divided into traditional fusion algorithms and deep learning (DL) fusion algorithms. The principles and workflow of some algorithms were analyzed and compared, their advantages and disadvantages were summarized, and relevant medical image data sets were introduced. Finally, the future development trend of medical image fusion algorithm was prospected, and the development direction of medical image fusion technology for the diagnosis of prostate cancer and other major diseases was pointed out.
The human brain deteriorates as we age, and the rate and the trajectories of these changes significantly vary among brain regions and among individuals. Because neuroimaging data are potentially important indicators of individual's brain health, they are commonly used in brain age prediction. In this review, we summarize brain age prediction model from neuroimaging-based studies in the last ten years. The studies are categorized based on their image modalities and feature types. The results indicate that the prediction frameworks based on neuroimaging holds promise toward individualized brain age prediction. Finally, we addressed the challenges in brain age prediction and suggested some future research directions.
The inverse problem of diffuse optical tomography (DOT) is ill-posed. Traditional method cannot achieve high imaging accuracy and the calculation process is time-consuming, which restricts the clinical application of DOT. Therefore, a method based on stacked auto-encoder (SAE) was proposed and used for the DOT inverse problem. Firstly, a traditional SAE method is used to solved the inverse problem. Then, the output structure of SAE neural network is improved to a single output SAE, which reduce the burden on the neural network. Finally, the improved SAE method is used to compare with traditional SAE method and traditional levenberg-marquardt (LM) iterative method. The result shows that the average time to solve the inverse problem of the method proposed in this paper is only 1.67% of the LM method. The mean square error (MSE) value is 46.21% lower than the traditional iterative method, 61.53% lower than the traditional SAE method, and the image correlation coefficient(ICC) value is 4.03% higher than the traditional iterative method, 18.7% higher than the traditional SAE method and has good noise immunity under 3% noise conditions. The research results in this article prove that the improved SAE method has higher image quality and noise resistance than the traditional SAE method, and at the same time has a faster calculation speed than the traditional iterative method, which is conducive to the application of neural networks in DOT inverse problem calculation.
ObjectiveThis study aimed to develop early mortality risk prediction models for patients with severe acute pancreatitis (SAP) based on eight machine learning algorithms, and to identify the major risk factors. MethodsClinical data of SAP patients diagnosed at West China Hospital of Sichuan University between January 2020 and August 2023, were retrospectively collected and randomly divided into a training set (n=878) and a validation set (n=376) in a 7∶3 ratio. Eight machine learning algorithms, including random forest, logistic regression, support vector machine, multilayer perceptron, XGBoost, Gaussian naive Bayes, CatBoost, and AdaBoost, were applied to construct early mortality prediction models for SAP. The models were evaluated using the area under curve (AUC), decision curve analysis (DCA), Shapley additive explanations (SHAP), and other indexes. ResultsA total of 1 254 SAP patients were finally included in this study, with an early mortality rate of 15.8% (198/1 254). The random forest algorithm demonstrated the best predictive performance in both the training and validation sets, with AUCs of 0.913 and 0.844, respectively. In the DCA, random forest also yielded the greatest net benefit. SHAP analysis ranked seven key predictors of early mortality in SAP by importance: age, body mass index, heart rate, need for assisted ventilation, hemoglobin, interleukin-6, and lactate dehydrogenase, with the need for assisted ventilation being the most critical predictor.ConclusionThe random forest model developed in this study can assist clinicians in more accurately identifying high-risk SAP patients at an early stage, thereby enabling timely interventions to reduce early mortality.
Objective To construct and compare risk prediction models for skip metastasis in papillary thyroid carcinoma (PTC) patients with lateral lymph node metastasis (N1b) by using multiple machine learning algorithms, and to provide clinical guidance through model interpretation and visualization. MethodsA retrospective analysis of 573 N1b PTC patients who were admitted between November 2011 and August 2024 in Zhongshan Hospital Affiliated to Xiamen University and undergone primary surgery were conducted. Patients were randomly divided into training (n=402) and testing (n=171) sets according to 7∶3 ratio by using R package caret. The training set was used to build the model, and the test set was used for model validation. Five machine learning models including logistic regression (LR), decision tree (DT), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost) by using 10-fold cross-validation on the training set to determine hyperparameters, then refited the models and validate them on the test set. Model performance was evaluated via area under the curve (AUC). Shapley additive explanations (SHAP) was employed for interpretability, and the optimal model was deployed as a web-based calculator using R Shiny. ResultsThe overall skip metastasis rate was 12.7% (73/573) in N1b PTC patients, with 12.9% (52/402) in the training set and 12.3% (21/171) in the testing set (P>0.05 for baseline comparisons). Eleven predictors (age, age≥55, sex, maximum tumor diameter, maximum tumor diameter≤1 cm, upper pole involvement, multifocality, unilateral lobe involvement, extrathyroidal extension, capsular invasion, and Hashimoto thyroiditis) were used to develop the model. Each model’s AUC of the training set: XGBoost, 0.824±0.070 ［95%CI (0.780, 0.868)］; LR, 0.802±0.065 ［95%CI (0.762, 0.842)］; DT, 0.773±0.141 ［95%CI (0.685, 0.861)］; RF, 0.767±0.068 ［95%CI (0.725, 0.809)］; SVM, 0.647±0.103 ［95%CI (0.583, 0.711)］. Each model’s AUC of the testing set: XGBoost, 0.777 ［95%CI (0.667, 0.887); LR, 0.769 ［95%CI (0.655, 0.883)］; DT, 0.737 ［95%CI (0.615, 0.858)］; RF, 0.757 ［95%CI (0.649, 0.865)］; SVM, 0.674 ［95%CI (0.522, 0.826)］. XGBoost was the optimum model which achieved the highest AUC in both training and testing sets. SHAP analysis identified the top six predictors: upper pole involvement (mean absolute SHAP: 0.249), maximum tumor diameter (0.119), extrathyroidal extension (0.078), age (0.065), unilateral lobe involvement (0.018), and capsular invasion (0.013). The XGBoost-based web calculator was accessible. ConclusionsThe XGBoost model demonstrates superior predictive performance among five machine learning algorithms. The developed web-based calculator offers clinical utility for assessing skip metastasis risk in N1b PTC patients.
ObjectiveTo explore the relationship between circadian rhythm genes and the occurrence, development, prognosis, and tumor microenvironment (TME) of lung adenocarcinoma (LUAD). MethodsThe Cancer Genome Atlas data were used to evaluate the expression, copy number variation, and somatic mutation frequency of circadian gene sets in LUAD. GO, KEGG, and GSEA enrichment analyses were used to explore the potential mechanisms by which circadian rhythm genes affected LUAD progression. Cox regression, least absolute shrinkage and selection operator regression, support vector machine recursive feature elimination, and random forest screened circadian genes and established prognostic models, and on this basis constructed nomogram to predict patients' 1-, 3-, and 5-year survival rates. Kaplan-Meier survival curves, receiver operating characteristic (ROC) curves, and time-dependent ROC curves were drawn to evaluate the predictive ability of the model, and the external dataset of GEO further verified the prognostic value of the prediction model. In addition, we evaluated the association of the prognostic model with immune cells and immune checkpoint genes. Single cell RNA sequencing (scRNA-seq) analysis was used to explore the molecular characteristics between prognostically relevant circadian genes and different immune cell populations in TME. ResultsDifferentially expressed circadian rhythm genes were mainly enriched in biological processes related to cGMP-PKG signaling pathway, lipid and atherosclerosis, and JAK-STAT signaling pathway. Seven circadian rhythm genes: LGR4, CDK1, KLF10, ARNTL2, RORA, NPAS2, PTGDS were screened out, and a RiskScore model was established. According to the median RiskScore, samples were divided into a high-risk group and a low-risk group. Compared with patients in the low-risk group, patients in the high-risk group showed a poorer prognosis (P<0.001). Immunological characterization analysis showed that there were differences in the infiltration of multiple immune cells between the low-risk group and high-risk group. Most immune checkpoint genes had higher expression levels in the high-risk group than those in the low-risk group, and RiskScore was positively correlated with the expression of CD276, TNFSF4, PDCD1LG2, CD274, and TNFRSF9, and negatively correlated with the expression of CD40LG and TNFSF15. The scRNA-seq analysis showed that RORA and KLF10 were mainly expressed in natural killer cells. ConclusionThe prognostic model based on seven feature circadian rhythm genes has certain predictive value for predicting survival of LUAD patients. Dysregulated expression of circadian genes may regulate the occurrence, progression as well as prognosis of LUAD through affecting TME, which provides a possible direction for finding potential strategies for treating LUAD from the perspective of mechanism by which circadian disorder affects immune cells.
ObjectiveTo construct a multimodal imaging radiomics model based on enhanced CT features to predict tumor regression grade (TRG) in patients with locally advanced rectal cancer (LARC) following neoadjuvant chemoradiotherapy (NCRT). MethodsA retrospective analysis was conducted on the Database from Colorectal Cancer (DACCA) at West China Hospital of Sichuan University, including 199 LARC patients treated from October 2016 to October 2023. All patients underwent total mesorectal excision after NCRT. Clinical pathological information was collected, and radiomics features were extracted from CT images prior to NCRT. Python 3.13.0 was used for feature dimension reduction, and univariate logistic regression (LR) along with Lasso regression with 5-fold cross-validation were applied to select radiomics features. Patients were randomly divided into training and testing sets at a ratio of 7∶3 for machine learning and joint model construction. The model’s performance was evaluated using accuracy, sensitivity, specificity, and the area under the curve (AUC). Receiver operating characteristic curve (ROC), confusion matrices, and clinical decision curves (DCA) were plotted to assess the model’s performance. ResultsAmong the 199 patients, 155 (77.89%) had poor therapeutic outcomes, while 44 (22.11%) had good outcomes. Univariate LR and Lasso regression identified 8 clinical pathological features and 5 radiomic features, including 1 shape feature, 2 first-order statistical features, and 2 texture features. LR, support vector machine (SVM), random forest (RF), and eXtreme gradient boosting (XGBoost) models were established. In the training set, the AUC values of LR, SVM, RF, XGBoost models were 0.99, 0.98, 1.00, and 1.00, respectively, with accuracy rates of 0.94, 0.93, 1.00, and 1.00, sensitivity rates of 0.98, 1.00, 1.00, and 1.00, and specificity rates of 0.80, 0.67, 1.00, and 1.00, respectively. In the testing set, the AUC values of 4 models were 0.97, 0.92, 0.96, and 0.95, with accuracy rates of 0.87, 0.87, 0.88, and 0.90, sensitivity rates of 1.00, 1.00, 1.00, and 0.95, and specificity rates of 0.50, 0.50, 0.56, and 0.75. Among the models, the XGBoost model had the best performance, with the highest accuracy and specificity rates. DCA indicated clinical benefits for all 4 models. ConclusionsThe multimodal imaging radiomics model based on enhanced CT has good clinical application value in predicting the efficacy of NCRT in LARC. It can accurately predict good and poor therapeutic outcomes, providing personalized clinical surgical interventions.
With the innovation and breakthrough of key technologies in smart medicine, actively exploring smart emergency measures and methods with artificial intelligence as the core technology is helpful to improve the ability of emergency medical team to diagnose and treat acute and critical diseases. This paper reviews the application status of artificial intelligence in pre-hospital and in-hospital diagnosis and treatment capabilities and system construction, expounds on the challenges it faces and possible coping strategies, and provides a reference for the in-depth integration and development of “artificial intelligence + emergency medicine” education, research and production during the new wave of scientific and technological revolution.
Objective To develop and validate a composite model (PAH score) based on dual-center data, integrating logistic regression and machine learning approaches, to improve the preoperative differential diagnostic efficacy for appendiceal mucinous neoplasms (AMNs). MethodsA dual-center retrospective case-control design was adopted. The study included 108 AMNs patients and 230 healthy controls from The 900th Hospital of Joint Logistics Support Force (January 2014 to November 2024) and Sanming First Hospital Affiliated to Fujian Medical University (December 2018 to December 2023) for feature screening and model construction. Additionally, 258 patients with pathologically confirmed chronic appendicitis (CA) from the same period were included as the differential validation group. Predictors were screened using leastabsolute shrinkage and selection operator combined with traditional logistic regression, and four machine learning algorithms—random forest, support vector machine, gradient boosting, and decision tree—were applied to rank feature importance. Core variables consistently identified by both approaches were integrated to construct a logistic regression model. Based on the model results, the PAH score was formulated, and its performance in distinguishing AMNs from CA was validated. An online visualization platform for AMNs risk prediction was subsequently developed. ResultsBaseline characteristics were balanced between the AMNs group and healthy control group, as well as between the AMNs group and CA group (P>0.05). Multivariate logistic regression identified prognostic nutritional index (PNI, OR=0.81), albumin-to-globulin ratio (AGR, OR=0.37), and hemoglobin to red blood cell distribution width ratio (HRR, OR=0.36) as independent predictors of AMNs (all P<0.001). All four machine learning algorithms consistently ranked PNI, AGR, and HRR as the top three important features. Based on these findings, a PAH model was constructed, and the PAH score was calculated using the standardized regression coefficient weighting method as follows: PAH score=20.8–0.21×PNI–0.99×AGR–1.01×HRR. The model demonstrated excellent discriminative ability for AMNs, with an area under the curve (AUC) of 0.918. The Hosmer-Lemeshow test indicated good calibration between predicted and observed probabilities (P=0.925). Decision curve analysis (DCA) showed significant net clinical benefit within the risk threshold range of 0.05–0.95. Bootstrap internal validation confirmed robust model performance (AUC=0.911). The median PAH score was significantly higher in the AMNs group than that of the CA group (MD=1.78, P<0.001). For distinguishing AMNs from CA, the PAH score achieved an AUC of 0.758. At the optimal cutoff value (–1.00), sensitivity was 70%, specificity was 76%, and accuracy rate was 74%. The Hosmer-Lemeshow test again confirmed good calibration (P=0.106), and Bootstrap validation indicated stable performance (AUC=0.783). DCA further demonstrated considerable net benefit within the threshold range of 0.20–0.95. ConclusionsThe PAH score developed in this study effectively predicts the risk of AMNs and accurately differentiates AMNs from CA, showing promising clinical application potential. However, as an exploratory study, further validation through multicenter, large-sample, prospective studies with diverse control groups is needed to enhance the generalizability and stability of the scoring system.