Heart disease

Acute myocardial infarction (AMI) is the main cause of death in developed and developing countries. AMI is a serious medical problem that necessitates hospitalization and sometimes results in death. Patients hospitalized in the emergency department (ED) should therefore receive an immediate diagnosis and treatment. Many studies have been conducted on the prognosis of AMI with hemogram parameters. However, no study has investigated potential hemogram parameters for the diagnosis of AMI using an interpretable artificial intelligence-based clinical approach. The purpose of this research is to implement the principles of explainable artificial intelligence (XAI) in the analysis of hematological predictors for AMI. In this retrospective analysis, 477 (48.6%) patients with AMI and 504 (51.4%) healthy individuals were enrolled and assessed in predicting AMI. Of the patients with AMI, 182 (38%) had an ST-segment elevation MI (STEMI), and 295 (62%) had a non-ST-segment elevation MI (NSTEMI). Demographic and hematological information of the patients was analyzed to determine AMI. The XAI approach combined with machine learning approaches (Extreme Gradient Boosting, XGB; Adaptive Boosting, AB; Light Gradient Boosting Machine, LGBM) was applied for the estimation of AMI and distinguishing subgroups of AMI (STEMI and NSTEMI). The SHAP approach was used to explain the predictions intuitively. After selecting the 10 most important hematological parameters for AMI, the LGBM model achieved 83% and 74% accuracy for prediction of AMI, and distinguishing subgroups of AMI (STEMI and NSTEMI), respectively. SHAP results showed that neutrophil (NEU), white blood cell (WBC), platelet width of distribution (PDW), and basophil (BA) were the most important for AMI prediction. Mean corpuscular volume (MCV), BA, monocytes (MO), and lymphocytes (LY) were the most important hematological parameters that distinguish STEMI from NSTEMI. The proposed model serves as a valuable tool for physicians, facilitating the diagnosis, treatment, and follow-up of patients with AMI and distinguishing subgroups of AMI (STEMI and NSTEMI). Analyzing readily accessible hemogram parameters empowers medical professionals to make informed decisions and provide enhanced care to a wide range of individuals.急性心肌梗死（Acute Myocardial Infarction, AMI）是发达国家与发展中国家的主要死亡原因。AMI是一类需住院干预的严重临床病症，有时可导致患者死亡。因此，急诊科（Emergency Department, ED）收治的患者需立即接受诊断与治疗。目前已有诸多研究围绕血常规参数对AMI的预后展开，但尚无研究采用基于可解释人工智能（Explainable Artificial Intelligence, XAI）的临床方法，探究可用于AMI诊断的潜在血常规参数。本研究旨在将可解释人工智能的理念应用于AMI血液学预测因子的分析工作中。本次回顾性分析共纳入477例（占比48.6%）AMI患者与504例（占比51.4%）健康个体，用于构建并评估AMI预测模型。在纳入的AMI患者中，182例（38%）为ST段抬高型心肌梗死（ST-segment Elevation Myocardial Infarction, STEMI），295例（62%）为非ST段抬高型心肌梗死（Non-ST-segment Elevation Myocardial Infarction, NSTEMI）。研究通过分析患者的人口学特征与血液学指标，实现AMI的预测与亚型鉴别。本研究将可解释人工智能方法与机器学习模型（极端梯度提升树（Extreme Gradient Boosting, XGB）、自适应提升（Adaptive Boosting, AB）、轻量梯度提升机（Light Gradient Boosting Machine, LGBM））相结合，用于AMI的预测以及AMI亚组（STEMI与NSTEMI）的区分，并采用SHAP（SHapley Additive exPlanations）方法对模型预测结果进行直观解释。在筛选出用于AMI预测的10项最重要血液学参数后，LGBM模型在AMI预测与AMI亚组区分任务中分别达到了83%与74%的准确率。SHAP分析结果显示，中性粒细胞（Neutrophil, NEU）、白细胞（White Blood Cell, WBC）、血小板分布宽度（Platelet Distribution Width, PDW）与嗜碱性粒细胞（Basophil, BA）是AMI预测中最为关键的血液学指标。而区分STEMI与NSTEMI的核心血液学参数则为平均红细胞体积（Mean Corpuscular Volume, MCV）、BA、单核细胞（Monocyte, MO）与淋巴细胞（Lymphocyte, LY）。本研究所提出的模型可为临床医师提供极具价值的辅助工具，助力AMI患者的诊断、治疗与随访，并实现AMI亚组的精准区分。对易于获取的血常规参数进行分析，可帮助医疗从业者做出更科学的临床决策，为更广泛的人群提供更优质的诊疗服务。