基于机器学习的出血性脑卒中后病况预测研究

A Machine Learning Based Study on the Prediction of Disease Status after Haemorrhagic Stroke

摘要: 出血性脑卒中是一种具有高死亡率、高致残率的急性脑血管疾病，建立出血性脑卒中后患者的临床病况预测模型对于临床医学界研究具备十分重要的意义。为此本文首先对100位出血性脑卒中患者医学特征数据进行预处理，构建相似距离矩阵填补缺失值，建立了基于Stacking的多模型融合方法预测患者48 h内发生血肿扩张概率；其次，引入SMOTE算法解决患者mRS评分类别分布不均衡问题，分别选用全部特征与提取特征带入以CatBoost作为分类器的SHAP模型，构建出血性脑卒中患者术后90天mRS评分预测模型，综合探究了影响患者90天mRS评分的医学特征因素。实验仿真结果表明，发病后血肿的灰度特征和形状特征对短期内的血肿扩张影响较大，且发病短期内患者右侧大脑前动脉是否出现水肿、水肿和水肿的形状变化体征对预后是否有症状和明显残疾影响程度较高。

Abstract: Haemorrhagic stroke is an acute cerebrovascular disease with high mortality and disability rates, and the establishment of a prediction model for the clinical condition of patients after haemorrhagic stroke is of great significance to the clinical medical research. To this end, this paper firstly preprocesses the medical feature data of 100 haemorrhagic stroke patients, constructs a similar distance matrix to fill in the missing values, and establishes a multi-model fusion method based on Stacking to predict the probability of haematoma expansion occurring in patients within 48 h. Secondly, the SMOTE algorithm is introduced to solve the problem of unbalanced distribution of patients’ mRS scores, and the whole features and the extracted features are selected and brought into a CatBoost as a classifier to predict the clinical condition of patients after a haemorrhagic stroke. CatBoost as the classifier of SHAP model, to construct a prediction model of 90-day postoperative mRS scores of patients with haemorrhagic stroke, and comprehensively explored the medical characteristic factors affecting the 90-day mRS scores of patients. The experimental simulation results showed that the grey scale features and shape features of the haematoma after the onset of the disease had a greater influence on the haematoma expansion in the short term, and whether or not the patient’s right anterior cerebral artery appeared in the short term of the onset of the disease, and whether or not the shape of the haematoma and the shape of the haematoma changed physically had a higher degree of influence on whether or not there was a symptomatic and obvious disability in the prognosis.