Scholarlink

Background and objective <p>The accurate prediction of symptomatic intracranial heamorrhage (sICH) risk after intravenous thrombolysis (IVT) is essential for clinical decision-making in patients with acute ischaemic stroke(AIS).We aimed to develop and validate machine learning-based models for sICH risk prediction and compare their performance against that of established clinical scores.</p> Methods <p>For this retrospective study,611 AIS patients(2017–2024)were included for model development and internal validation (70/30 split),along with an independent external cohort (<i>n</i> = 100) for testing.Using artificial neural network (ANN) and decision tree (DT) algorithms,we constructed predictive models incorporating 14 demographic,clinical,and neuroimaging variables.A simplified bedside “ANN score” was derived.The performance was compared against that of seven conventional scales (ASPECTS, GRASPS, MSS, HAT, SEDAN, THRIVE, and DRAGON) using the area under the receiver operating characteristic curve (AUC).</p> Results <p>The incidence of sICH was 9.3% in the primary cohort and 11% in the external cohort.The ANN model demonstrated superior predictive performance,with AUCs of 0.988 (95% CI: 0.977–1.000) in the training,0.996 (95% CI: 0.988–1.000) internal validation,and 0.997 (0.990–1.000) external validation cohorts.The derived ANN-score also maintained high accuracy (external AUCs: 0.932).The core infarction volume (ICV) was identified as the strongest predictor. Among conventional scores,the SEDAN performed best (external AUC:0.919),yet the superiority of the ANN model remained significant (<i>p</i> &lt; 0.001).</p> Conclusion <p>The ANN-based model and its derived simplified score provide a reliable approach for sICH risk stratification.Compared with traditional scoring systems,this model offers enhanced precision,potentially aiding clinicians in identifying high-risk patients more effectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From prediction to practice: an artificial neural network-derived risk score superior to conventional scales for predicting symptomatic intracranial haemorrhage risk after thrombolysis

  • 期刊论文

  • 发表日期：

    2026年4月18日

查看全文
  • Huihui Dong,
  • Zhongge Dong,
  • Lin Fu

摘要

Background and objective

The accurate prediction of symptomatic intracranial heamorrhage (sICH) risk after intravenous thrombolysis (IVT) is essential for clinical decision-making in patients with acute ischaemic stroke(AIS).We aimed to develop and validate machine learning-based models for sICH risk prediction and compare their performance against that of established clinical scores.

Methods

For this retrospective study,611 AIS patients(2017–2024)were included for model development and internal validation (70/30 split),along with an independent external cohort (n = 100) for testing.Using artificial neural network (ANN) and decision tree (DT) algorithms,we constructed predictive models incorporating 14 demographic,clinical,and neuroimaging variables.A simplified bedside “ANN score” was derived.The performance was compared against that of seven conventional scales (ASPECTS, GRASPS, MSS, HAT, SEDAN, THRIVE, and DRAGON) using the area under the receiver operating characteristic curve (AUC).

Results

The incidence of sICH was 9.3% in the primary cohort and 11% in the external cohort.The ANN model demonstrated superior predictive performance,with AUCs of 0.988 (95% CI: 0.977–1.000) in the training,0.996 (95% CI: 0.988–1.000) internal validation,and 0.997 (0.990–1.000) external validation cohorts.The derived ANN-score also maintained high accuracy (external AUCs: 0.932).The core infarction volume (ICV) was identified as the strongest predictor. Among conventional scores,the SEDAN performed best (external AUC:0.919),yet the superiority of the ANN model remained significant (p < 0.001).

Conclusion

The ANN-based model and its derived simplified score provide a reliable approach for sICH risk stratification.Compared with traditional scoring systems,this model offers enhanced precision,potentially aiding clinicians in identifying high-risk patients more effectively.