BMC Medical Informatics and Decision Making最新文献

Background: Atherosclerosis in the carotid artery significantly contributes to embolic events leading to ischemic stroke. Precise identification of unstable carotid plaques through non-invasive imaging, is pivotal for stroke prevention. Artificial intelligence (AI) has demonstrated promise in enhancing the accuracy of plaque risk stratification. This review aims to assess the diagnostic performance of AI algorithms in distinguishing unstable carotid plaques from stable plaques using medical imaging.

Methods: We conducted comprehensive searches in Medline, Embase, Web of Science, IEEE, PubMed, and the Cochrane Library up to June 6, 2023. Eligible studies included those that utilized AI algorithms for identifying unstable carotid plaques from medical images. Binary diagnostic accuracy metrics, including sensitivity, specificity, and Area Under the Curve (AUC), were extracted. QUADAS-AI was used to assess risk of bias of the included studies.

Results: Among the 31 studies reviewed, 14 were subjected to meta-analysis, revealing a pooled sensitivity of 91% (95%CI: 86 - 95%), specificity of 84% (79 - 89%), and AUC of 0.94 (0.91 - 0.95). However, only one study reported external validation, limiting the generalizability of these findings, and substantial heterogeneity was observed (I² > 90%). Subgroup analyses indicated performance variations based on factors such as sample size, type of AI algorithms (machine learning or deep learning), segmentation methods (manual or automatic), and publication year. Despite observed publication bias and study heterogeneity, the findings underscore the promise of AI-driven approaches in carotid plaque risk stratification.

Conclusions: AI algorithms demonstrated favorable diagnostic performance in identifying unstable carotid plaques. Future research should focus on rigorous validation, ensuring generalizability, and enhancing the explainability of AI algorithms to facilitate their translational use.

Clinical trial number: Not applicable.

Background: Accurate and rapid assessment of fluid status of maintenance hemodialysis (MHD) patients and maintaining fluid balance is essential to ensure the quality of dialysis treatment. Currently, clinical methods for assessing ultrafiltration volume are still insufficient, and reliable tools that are more accurate and rapid are needed. The objective of this study was to construct a model for predicting ultrafiltration volume (UF) in MHD patients based on artificial neural network (ANN) algorithms, to validate and evaluate this model, and to investigate the impact of body composition prior to dialysis on UF in MHD patients.

Methods: A total of 319 patients undergoing MHD treatment at our center were enrolled. Basic demographic and clinical characteristics were collected and evaluated using the hemodialysis information system. Body composition was measured on ≥ 3 separate days before dialysis treatment using an Inbody bioimpedance instrument. The target ultrafiltration volume was determined by nephrologists based on the integration of body composition measurements and clinical characteristics, yielding a dataset of 1,205 entries. Heat maps were used to demonstrate the correlation between body composition and UF in MHD patients, and LASSO regression and multifactorial linear regression were used to screen the relevant indicator factors for final inclusion in the model, and Backpropagation Neural Network model (BPNN) was developed using the MATLAB (R2022a) neural network toolbox to establish the projected relationship between UF and pre-dialysis body composition. The effectiveness of the model was assessed based on the coefficient of determination (R²) and root mean square error (RMSE) of the calculated regression.

Results: The artificial neural network model demonstrated an optimal predictive performance metric of R² = 0.965 for forecasting ultrafiltration volume in MHD patients. With an average difference of 0.182 L between observed and predicted values, and highlighted the significant influence of certain body composition indicators on UF in MHD patients.

Conclusion: This study effectively demonstrates the predictive role of an artificial neural network model based on pre-dialysis body composition information in estimating ultrafiltration providing a valuable predictive tool to optimize assessment volume for MHD patients, of ultrafiltration volume in MHD patients.

Background: Acute kidney injury (AKI) has been confirmed to be related to the prognosis of aSAH patients. Evaluating the risk of AKI in the early stage is important to avoid the unfavorable outcome of aSAH patients. However, no study has explored the predictive value of machine learning algorithms for AKI after aSAH. This study was designed to develop a machine learning algorithm-based predictive model for AKI among aSAH patients.

Methods: The outcome of this study was the AKI confirmed using the KDIGO criteria. The predictive value of seven machine learning algorithms for the AKI among aSAH patients was explored and verified using the 5-fold cross-validation. The predictive efficiency of machine learning algorithms-based predictive models was evaluated by the area under the receiver operating characteristics curve (AUC). The Shapley Additive explanation method was performed to visualize the importance of features incorporated in machine learning algorithms-based predictive models.

Results: 711 aSAH patients were enrolled with an AKI incidence of 7.7%. The AKI group had higher WFNS (p = 0.011), Hunt Hess (p = 0.006), and lower Glasgow Coma Scale (GCS) (p = 0.004). The multiple aneurysm was more frequently observed in the AKI group (p = 0.027). The AKI group had longer length of ICU stay (p < 0.001), length of hospital stay (p < 0.001), and higher mortality (p < 0.001). Three algorithms performed well in predicting the AKI in the training dataset including the random forest (AUC = 1.000), AdaBoost (AUC = 0.954), and XGBoost (AUC = 0.947). The random forest performed the best in the validation dataset with an AUC of 0.724. The top ten features in the random forest algorithm were GCS, mean blood pressure, initial serum creatinine, cystatin C level, albumin, neutrophil, lactate dehydrogenase, glucose, white blood cell, and sodium.

Conclusions: The random forest model demonstrated superior performance in predicting AKI in aSAH patients, achieving a high AUC value, predictive accuracy, and remarkable stability. This model could help clinicians evaluate the risk of AKI in the early stage and guide therapeutic options among aSAH patients.

Background: Adolescent depression is a major public health concern. Physical health indicators are rarely included in risk tools. We examined whether adding dry eye disease (DED) to psychosocial and behavioral factors improves prediction of depressive symptoms.

Methods: We analyzed 2,076 adolescent questionnaires (94.5% response) covering ocular health, sleep, electronic device use, social support, and demographics. Five machine-learning classifiers were trained with cross-validation and evaluated for discrimination and calibration.

Results: Models that included DED achieved strong discrimination (AUC ≈ 0.84) and good calibration, with highest accuracy for no and severe depression and lower performance for mild/moderate categories.

Conclusions: Integrating ocular health with psychosocial factors enhances machine-learning prediction of adolescent depression and may support earlier, school-based identification and referral. Given the low-cost, questionnaire-based inputs and favorable calibration, this approach shows promise for population screening and targeted prevention, pending external validation and prospective testing.

Background: Making precise treatment decisions in esophageal cancer is essential for enhancing patient outcomes and avoiding overtreatment. Traditional approaches relying on special features or shallow learning models often fail to capture the complex, multi-scale patterns embedded in PET/CT imaging data. Recent advances in deep learning provide an opportunity to build more robust, data-driven systems for predictive modeling in oncology.

Methods: We propose a novel deep learning model that integrates convolutional and transformer-based components based on PET/CT data to support treatment decisions for esophageal cancer. The architecture introduces a Convolutional Feature Extractor with split-based residual blocks for efficient local feature capture, a Multi-scale Pooling module for spatial context aggregation, and an Multilayer Perceptron block for predicting. The model was evaluated using several performance metrics such as AUCROC, F1 score, Balanced Accuracy and benchmarked against state-of-the-art convolutional and transformer backbones such as ConvNeXt and Vision Transformer.

Results: The proposed model achieved superior performance across all evaluation metrics, including an AUCROC of 0.9935 and a Balanced Accuracy of 0.9630, outperforming existing models. These results validate the effectiveness of combining local-global representation learning through custom-designed modules. In addition, we conducted ablation studies to further demonstrate the individual contributions and effectiveness of each component within the proposed architecture. By systematically removing or replacing specific modules such as the Convolutional Feature Extractor and Multi-scale Pooling, we observed consistent performance degradation, which highlights the necessity and complementary roles of these components in achieving optimal predictive accuracy.

Conclusions: This study presents a novel hybrid deep learning architecture that enhances treatment decision support for esophageal cancer by leveraging multi-scale spatial encoding. The empirical evidence demonstrates that tailored architectural innovations significantly improve predictive accuracy over existing methods.

Cervical cancer is the leading cause of cancer-related deaths among women worldwide, necessitating early and accurate detection methods. This study introduces a hybrid framework utilizing Vision Transformers (ViT) and ensemble learning-based convolutional neural networks (CNN) models for cervical cancer classification based on Pap smear images. Two prominent datasets, Mendeley LBC and SIPaKMeD, are employed, encompassing nine distinct categories of cervical cell abnormalities. The proposed approach integrates pre-trained CNN models of DenseNet201, Xception, and InceptionResNetV2 to extract high-level features, further fused through ensemble learning. These features are then processed by the ViT-based encoder model designed for improved interpretability and accuracy. Experimental results demonstrate that the hybrid model achieves superior accuracy rates of 97.26%, a recall of 97.27%, a precision of 97.27%, and 96.70% for the F1-score on the Mendeley LBC dataset. For the SIPaKMeD dataset, there was an accuracy of 99.18%, a recall of 99.18%, a precision of 99.15%, and a 99.21% F1-score. On the combined dataset, the model outperformed individual pre-trained models with 95.10% accuracy and a 95.01% F1-score. Moreover, the framework incorporates augmentation with Explainable AI (XAI) techniques, specifically Grad-CAM, to provide transparent and interpretable diagnostic outcomes, enhancing its utility in clinical settings. This research underscores the potential of hybrid AI frameworks in revolutionizing cervical cancer diagnostics by offering accurate, efficient, and interpretable solutions.

Purpose: This scoping review aims to synthesize research on artificial intelligence (AI) in predicting open-heart surgery outcomes, evaluating AI model performance, and identifying gaps in data quality, algorithmic bias, and clinical applicability to guide future advancements in personalized surgical planning and patient outcomes.

Methods: Conducted using the PRISMA-ScR guideline, the review involved a systematic search across PubMed, Web of Science, IEEE, and Scopus. Articles were included if they focused on open-heart surgery, utilized AI methods, and were published in English. Exclusion criteria included non-relevance to open-heart surgery, non-original research, and lack of AI techniques. Data extraction included study details, AI methods, and performance metrics. Descriptive statistics were used for analysis.

Results: Of the 64 included studies, 89.06% were retrospective. The most frequently employed algorithm was logistic regression (n = 41), followed by random forest in 38 studies and XGBoost in 32 studies for data analysis. Most studies focused on predicting postoperative outcomes. Mortality, acute kidney injury, and complications were the outcomes that more studies concentrated on. XGBoost, used in 32 studies, exhibited the best performance in 11 of these studies. Deep learning and hybrid models were underutilized. Major limitations included inconsistent model validation, limited prospective data, and lack of diversity in patient populations.

Conclusion: AI demonstrates promising predictive capabilities in open-heart surgery, particularly through machine learning models. These models can already assist surgeons in real-world practice by supporting real-time risk stratification and personalized decision-making, such as identifying high-risk patients for targeted interventions. However, methodological limitations hinder clinical translation. Future work should emphasize prospective validation, explainable AI, and equitable data representation to enhance model reliability and applicability in real-world settings.

Background: Myocardial infarction (MI) is a life-threatening condition caused by sudden interruption of blood supply to the heart. Electrocardiogram (ECG) is the primary tool for MI diagnosis, but interpretation challenges exist. This study aimed to optimize MI detection by developing a hybrid CNN-GRU Deep Learning model (DLM) based on ECG as a diagnostic support tool.

Methods: This retrospective diagnostic study included a total of 56,354 ECGs, comprising 41,871 from patients diagnosed with (MI) and 14,474 from healthy patients. Each ECG record consisted of a 20-second 15-lead recording per individual, sampled at 1000 Hz. The CNN-GRU model was trained on 85% of these ECGs and validated on the remaining 15%. The CNN-GRU model was executed on the pre-processed data using the Pan-Tompkins algorithm obtained from the PhysioNet website (PTB Diagnostic ECG Database), and all recordings were labelled by expert cardiologists. We examined a new model for classifying ECG heartbeats and found that it can compete with advanced models. The performance of the DLM was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, Macro Average, and Weighted Average.

Results: The area under the receiver operating characteristic curve (AUC) of the CNN-GRU model for MI detection was close to one. The CNN-GRU model achieved excellent performance with 15 leads (ACC = 99.43%, sensitivity = 99.71%, specificity = 98.59%). Using lead II alone, performance improved slightly (ACC = 99.73%, sensitivity = 99.75%, specificity = 99.66%). The high AUC and other metrics indicate strong diagnostic ability. Based on the reported results, the CNN-GRU model using lead II was the best model.

Conclusions: The findings suggest that the proposed model can support clinical decision-making and guide future research in cardiovascular medicine.