Aug 2026· Frontiers in Cardiovascular Medicine· Vol 13· 0 citations· 39 references
Medicine
TL;DR
A two-fold framework integrating deep learning and large language models to predict MACE risk and generate personalized, guideline-based recommendations is proposed, which effectively combines accurate MACE risk prediction with interpretable, evidence-based preventive recommendations delivered via a web interface.
Abstract
Background Major adverse cardiovascular events (MACE) remain a leading cause of global morbidity and mortality, necessitating accurate risk prediction and actionable prevention strategies. This study proposes a two-fold framework integrating deep learning and large language models (LLMs) to predict MACE risk and generate personalized, guideline-based recommendations. Methods Emergency medical record (EMR) data from Chungbuk National University (CBNU) Hospital were preprocessed and split into training and test sets, with class imbalance addressed using SMOTETomek. A one-dimensional convolutional neural network (1D-CNN) was developed to predict individual MACE risk from clinical features. To translate predictions into practice, a retrieval-augmented generation (RAG) pipeline with prompt engineering was implemented using LLM Model (MISTRAL AI), grounded in WHO and AHA cardiovascular prevention guidelines. Predictive performance was evaluated using accuracy, precision, recall, and F1-score. The recommendation system was assessed using a rubric-based LLM judge evaluating guideline fidelity, clinical safety, dietary plans, lifestyle/exercise guidance, and OTC medication caution. Results The proposed1D-CNN achieved strong performance, with overall accuracy of 0.97. For the MACE class, precision and recall reached 1.00, with an F1-score of 0.98. The recommendation system demonstrated high guideline fidelity (5/5 for both WHO and AHA), strong dietary and lifestyle guidance (5/5), good clinical safety (4/5), and moderate OTC medication caution (3/5). Conclusion The proposed framework effectively combines accurate MACE risk prediction with interpretable, evidence-based preventive recommendations delivered via a web interface. This integrated approach supports clinical decision-making and personalized cardiovascular risk management.
Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide, accounting for approximately 17.9 million deaths annually according to the World Health Organization. Early and accurate risk assessment is critical for timely intervention and improved patient outcomes. However, existing machine learning and deep learning approaches face significant limitations including poor modeling of nonlinear relationships, inadequate feature interaction learning, lack of uncertainty quantification, class imbalance challenges, overfitting tendencies, limited interpretability, insufficient multimodal fusion capabilities, poor cross-dataset generalization, absence of attention-based learning mechanisms, and inadequate clinical explainability. To address these critical gaps, we propose CardioAttentionNet, a novel hybrid deep learning framework that integrates residual convolutional neural networks, transformer encoders with multi-head self-attention, bidirectional long short-term memory networks, and cross-attention modules for comprehensive cardiovascular risk prediction. The framework incorporates intelligent data preprocessing, hybrid feature engineering with attention-based feature selection, adaptive feature fusion with learnable weights, channel attention mechanisms, probability calibration, and explainable AI components including SHAP, LIME, and Integrated Gradients. Evaluated on a large-scale multi-source dataset comprising 75,000 samples from UCI Cleveland, Framingham Heart Study, Kaggle, and Cardiovascular Disease datasets, CardioAttentionNet achieved exceptional performance with 98.12% accuracy, 97.84% precision, 98.37% recall, 97.93% specificity, 98.10% F1-score, 0.983 AUC, 0.962 MCC, and 0.961 Cohen’s Kappa, significantly outperforming 13 baseline methods. Comprehensive ablation studies, statistical validation, and explainability analysis demonstrate the framework’s clinical utility for early heart disease risk stratification and decision support in real-world healthcare settings.
Swapnil Hiralal Chaudhari, Anish Kumar Choudhary· International journal of com...· 0 citations
It is argued that LASSO, not the highest-discriminating model, is the model best suited to direct clinical deployment, and lessons for the machine learning and healthcare community regarding data infrastructure, model selection, and value of calibration and interpretability in high-stakes decision support are presented.
Asra Aslam, Volodymyr Chapman, M. O'Connell et al.· 0 citations
Cardiac Disease continues to be a foremost universal health issue every year, highlighting the need for early and accurate predictions, which are essential for preventing critical outcomes. Traditional clinical prediction models focus largely on structured clinical features such as Age, cholesterol, and blood pressure. This provides valuable insights for identifying patients at risk. However, structured characteristics largely fail to express the Unpredictable characteristics of their data. Existing approaches, such as predictive analytics models and other deep-learning architectures (like CNN +LSTM), have more limited capabilities because they either learn temporal patterns (LSTM) but lack strong feature extraction or extraction of spatial features, or do not model sequential relationships (CNN). A combined CNN+LSTM model for cardiac risk prediction on a clinical dataset is presented in this work. A well-established clinical dataset, such as the publicly available Clinical dataset, with 14 clinically validated features derived from real records, will be used as the model training point. Preprocessing will be performed on the dataset for the purpose of improving performance. Within this proposed architecture, the CNN layers will extract high-level feature patterns from clinical inputs, and the LSTM layers will learn structural dependencies that provide a relatively complete indicator of patient health to allow better decision-making and easier patient stratification. The experimental results indicated that the hybrid CNN+LSTM outperformed in prediction with 98.05% accuracy, 96.29% precision, and 95.56% recall, thereby supporting the conclusion that the integration of CNN and LSTM improves feature learning and better sequence learning improves Cardiac disease prediction.
P. Yellamma, Thadiboina Sai Teja· International Journal of Ele...· 0 citations
The reviewed studies demonstrated that Deep learning approaches, particularly convolutional neural networks, transformer-based models, and multimodal frameworks, showed improved predictive accuracy when large and diverse datasets were available.
Tehreem Khan, Tayyaba Usman, Ifrah Khalid et al.· Discover Artificial Intellig...· 0 citations
Cardiovascular disease is a main cause of mortality worldwide that need the development of reliable and data-driven prediction models for timely diagnosis and intervention. Conventional risk assessment methods depend on statistical scoring and handcrafted clinical features fail to capture complex nonlinear features in several cardiovascular existing data. This systematic literature review investigates recent advances in artificial intelligence-based CVD prediction that include machine learning, deep learning, hybrid, and ensemble learning models. The PRISMA-2020 strategy and Kitchenham guidelines, 110 studies published between 2013 and 2025 were systematically analyzed across diverse data modalities such as electronic health records, electrocardiograms, medical images, wearable sensors, and multimodal datasets. The findings indicate that tree-based ML models such as Random Forest, XGBoost, and Gradient Boosting shows the strong performance on structured clinical datasets. The DL-based models such as convolutional neural networks, long short-term memory networks, and transformer-based models effectively capture the high-dimensional physiological and imaging features. Hybrid model combined deep representations with clinical parameters to achieve predictive performance up to 98.00%, and ensemble learning models shows the superior performance and generalization capabilities across several distinct datasets. The challenges in cardiovascular disease detection such as limited multimodal datasets, very few external validation, insufficient explainability, and the absence of standardized benchmark evaluation parameters continue to show down the clinical translation. This review synthesizes current algorithmic developments, identifies critical research gaps, and highlights the potential of multimodal learning, explainable AI, federated learning, and clinically validated model for advancing next-generation cardiovascular risk prediction systems.
A. Gaikwad, N. Kulkarnir· Journal of Intelligent Decis...· 0 citations
Cardiovascular disease is the leading cause of death worldwide, and early, data-driven risk identification can support timely clinical intervention. This paper presents a study of deep learning models for binary heart disease prediction on a large structured clinical dataset of 50,000 records described by 20 demographic, lifestyle, medical-history, and physiological attributes. After a preprocessing pipeline that recovers a mislabelled "no-alcohol" category, one-hot encodes categorical variables, and standardizes numerical features, we design and compare three deep neural architectures, a regularized deep neural network (DNN), a bidirectional long short-term memory network (BiLSTM), and a hybrid CNN–BiLSTM, against a logistic-regression baseline. All models are evaluated on a held-out test set using accuracy, precision, recall, F1-score, and the area under the ROC curve. The proposed DNN achieves the highest performance, with 99.68% accuracy, a 99.65% F1-score, and an ROC area of approximately 1.0, misclassifying only 24 of 7,500 test samples and outperforming the BiLSTM (98.95%), the CNN–BiLSTM (98.15%), and the linear baseline (92.11%). A feature-correlation and ablation analysis shows that the outcome is governed by a small set of non-linearly interacting risk factors, hypertension, age, total cholesterol, diabetes, and prior myocardial infarction, which explains why the deep models capture a decision boundary that the linear model cannot. Because the dataset is a synthetic benchmark, the results are reported transparently, and the need for external validation on real-world clinical cohorts is discussed.
Rakib Hasan, A. Maimun, Khandaker Fahima Ahmad et al.· 2026 International Conferenc...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
