探索心脏疾病智能预测：人工智能体与文本嵌入模型的综合应用

Exploring heart disease smart prediction: integration of AI agents and text embedding models

摘要

目的 探讨人工智能体(AI智能体)如何整合多种分类算法和文本嵌入模型,以优化心脏疾病诊断中的预测准确性和上下文相关性。方法 基于智能体的心脏疾病预测模型(agent-based heart disease prediction, ABHDP)整合了分类建模、文档检索和用户交互组件。研究利用心脏疾病临床数据集以及补充数据集,训练了包括随机森林和梯度提升在内的多种分类算法,文本嵌入模型以及本地化部署的 DeepSeek。采用准确率、精确率、召回率、F1值和受试者工作特征(receiver operating characteristic, ROC)曲线分析量化模型性能和效果。结果 对分类和连续变量的分析揭示了与心脏疾病患病率相关的重要模式。CatBoost和随机森林等预测模型显示出较高的性能,召回率分别为0.980和0.979。在文本嵌入模型方面,FastText的F1评分为46%,而SBERT 和 OpenAI嵌入模型的表现优于FastText,分别达到了95%和96%的F1评分。结论 ABHDP模型可提高心脏疾病管理的准确性和个性化水平,推动精准医疗的发展。

Abstract

Objective To develop an agent-based heart disease prediction (ABHDP) model that combines multiple machine learning classifiers with text embedding model to improve prediction performance and contextual interpretation in heart disease risk assessment. Methods The proposed ABHDP model integrates three functional modules: disease classification, medical knowledge retrieval, and interactive interpretation. A heart disease dataset was used together with medical knowledge dataset. Eleven classification algorithms and three text embedding approaches were evaluated. Model performance was evaluated using accuracy, precision, recall, F1-score, and receiver operating characteristic (ROC). Results Analysis of categorical and continuous variables revealed important patterns associated with the prevalence of heart disease. Among the classification algorithms, CatBoost and random forest achieved the highest recall values (0.980 and 0.979, respectively). For text embedding methods, FastText produced an F1-score of 46%, while SBERT and OpenAI achieved substantially higher scores of 95% and 96%. Conclusion The proposed ABHDP model demonstrates the potential to support interpretable and intelligent clinical decision-making for heart disease risk prediction, thus promoting the development of precision healthcare.

关键词

大型语言模型 / SBERT / OpenAI / 人工智能体(AI智能体) / 基于智能体的心脏疾病预测模型(ABHDP)

Key words

large language model / SBERT / OpenAI / AI agent / agent-based heart disease prediction (ABHDP) model

参考文献

[1] HUANG J, DANG F L. Analysis of inducing factors of chronic pulmonary heart disease caused by chronic obstructive pulmonary disease at high altitude through epidemiological investigation under intelligent medicine and big data[J]. J Healthc Eng, 2022, 2022: 2612074.

[2] MOHAMMAD U, SAEED F. Robustness of ML—based seizure prediction using noisy EEG data from limited channels[C]//2024 20th International Conference on Distributed Computing in Smart Systems and the Internet of Things (DCOSS—IoT). Abu Dhabi, United Arab Emirates: IEEE, 2024: 620-626.

[3] LIN X, SU Z Q, LIU Y L, et al. SuperMetal: a generative AI framework for rapid and precise metal ion location prediction in proteins[J/OL].bioRxiv(2025—03—21)[2026—03—20].https://pubmed.ncbi.nlm.nih.gov/40196543/.

[4] NEUMANN A U, BORM K J, TRAIDL HOFFMANN C. Comprehensive risk stratification to guide an optimal preventive strategy for breast radiation dermatitis—reply[J]. JAMA Oncol, 2024, 10(8): 1137-1138.

[5] SUN X Y. EcoCurrentNet an integrated DNN—CatBoost model for predicting optoelectronic material performance under varying conditions[J]. Sci Rep, 2025, 15(1): 41485.

[6] HOSSAIN M R. Optimizing heart disease prediction: a comparative study of machine learning techniques[J]. Asian J Cardiol Res, 2024, 7(1): 348-359.

[7] DISSANAYAKE K, MD J M. Comparative study on heart disease prediction using feature selection techniques on classification algorithms[J]. Applied Computational Intelligence and Soft Computing, 2021, 2021(1): 5581806.

[8] KAUSHAL P, SINGH S, VIJAYVERGIYA R. A kernel attention—based transformer model for survival prediction of heart disease patients[J]. J Cardiovasc Transl Res, 2024, 17(6): 1295-1306.

[9] GISKE C G, BRESSAN M, FIECHTER F, et al. GPT—4—based AI agents—the new expert system for detection of antimicrobial resistance mechanisms?[J]. J Clin Microbiol, 2024, 62(11): e0068924.

[10] TARAMASCO C, PINEIRO M, ORMEÑO—ARRIAGADA P, et al. Multimodal dataset for sensor fusion in fall detection[J]. PeerJ, 2025, 13: e19004.

[11] ATZBERGER D, CECH T, SCHEIBEL W, et al. A large—scale sensitivity analysis on latent embeddings and dimensionality reductions for text spatializations[J]. IEEE Trans Vis Comput Graph, 2025, 31(1): 305-315.

[12] SMITH J. Heart disease dataset[DS/OL]. [2026—03—20].https://www.kaggle.com/datasets/johnsmith88/heart—disease—dataset.

[13] JANOSI A, STEINBRUNN W, PFISTERER M, et al. Heart disease[DS/OL]. (1988—07—01) [2026—03—20].https://archive.ics.uci.edu/dataset/45/heart+disease.

[14] ALMUSTAFA K M. Prediction of heart disease and classifiers' sensitivity analysis[J]. BMC Bioinformatics, 2020, 21(1): 278.

[15] TOYHOM. Chinese—medical—dialogue—data[DS/OL]. [2026—03—20].https://github.com/Toyhom/Chinese—medical—dialogue—data.

[16] GRETELAI. Symptom_to_diagnosis[DS/OL]. [2026—03—20].https://huggingface.co/datasets/gretelai/symptom_to_diagnosis.

[17] SANTANDER CRUZ Y, SALAZAR COLORES S, PAREDES GARCIA W J, et al. Semantic feature extraction using SBERT for dementia detection[J]. Brain Sci, 2022, 12(2): 270.

[18] DIVEICA V, MURAKI E J, BINNEY R J, et al. Mapping semantic space: exploring the higher—order structure of word meaning[J]. Cognition, 2024, 248: 105794.

[19] TEJADA—CASADO M, PÉREZ M M, DELLA BONA A, et al. Chroma—dependence of CIEDE2000 acceptability thresholds for dentistry[J]. J Esthet Restor Dent, 2024, 36(3): 469-476.

[20] YI F, ZHANG C, ZOU Y, et al. Three—dimensional crystal Vue imaging technology assessment of Sylvian fissures at 20—32+6 weeks' normal gestation[J]. Eur Radiol, 2023, 33(4): 2358-2366.

[21] GAVALI S, ROSS K E, COWART J, et al. iPTMnet RESTful API for post—translational modification network analysis[J]. Methods Mol Biol, 2022, 2499: 187-204.

[22] LI D, GUPTA K, BHADURI M, et al. Comparative diagnostic accuracy of GPT—4o and LLaMA 3—70b: proprietary vs. open—source large language models in radiology[J]. Clin Imaging, 2025, 118: 110382.

[23] HUSSAIN Z S, DELSOZ M, ELAHI M, et al. Performance of DeepSeek, Qwen 2.5 MAX, and ChatGPT assisting in diagnosis of corneal eye diseases, glaucoma, and neuro—ophthalmology diseases based on clinical case reports[J/OL].medRxiv(2025—03—17)[2026—03—20].https://pmc.ncbi.nlm.nih.gov/articles/PMC11957078/.

[24] REVATHI A R, AGASH A A. Pomegranate disease diagnosis with severity estimation and treatment remedies using deep learning and RAG—based LLM[J]. Sci Rep, 2025, 15(1): 41278.

[25] WEI X P, LI M, TU Y B, et al. ROC—guided virtual screening, molecular dynamics simulation, and bioactivity validation assessment Z195914464 as a 3CL Mpro inhibitor[J]. Biophys Chem, 2025, 317: 107357.

[26] GATTI D, RAVELING L, PETRENCO A, et al. Valence without meaning: investigating form and semantic components in pseudowords valence[J]. Psychon Bull Rev, 2024, 31(5): 2357-2369.

[27] SEJUTI Z A, ISLAM M S. A hybrid CNN—KNN approach for identification of COVID—19 with 5—fold cross validation[J]. Sens Int, 2023, 4: 100229.

[28] VORMITTAG P, KLAMP T, HUBBUCH J. Optimization of a soft ensemble vote classifier for the prediction of chimeric virus—like particle solubility and other biophysical properties[J]. Front Bioeng Biotechnol, 2020, 8: 881.

[29] AKBAR S, RAZA A, AWAN H H, et al. pNPs—CapsNet: predicting neuropeptides using protein language models and FastText encoding—based weighted multi—view feature integration with deep capsule neural network[J]. ACS Omega, 2025, 10(12): 12403-12416.

[30] MUTHURAJ R, GOPAL D, AHMED I, et al. Insightful t—SNE guided exploration spotlighting palbociclib and ribociclib analogues as novel WEE1 kinase inhibitory candidates[J]. J Biomol Struct Dyn, 2025, 43(9): 4716-4728.

[31] ACACIO—CLARO P J, ESTUAR M R J E, VILLAMOR D A R, et al. Understanding adoption of electronic medical records (EMRs) during a health emergency: an analysis of EMR usage logs from rural health facilities in the Philippines[J]. Acta Med Philipp, 2024, 58(22): 7-13.

基金资助

国家卫生健康委医院管理研究所医疗人工智能临床应用研究课题(YLXX24AIA020)