Scholarlink

<p>Intradialytic Hypotension is one of the potential complications of Kidney dialysis. Medical professionals use all possible means to avoid such complications. This study presents a machine learning-based method, the Intradialytic Hypotension Detection model using Multi-Scale Temporal Features (IDH-MSTF), which can be used continuously during the dialysis session to predict the occurrence of Intradialytic Hypotension. The proposed method utilizes 32 readily available features, including systolic and diastolic blood pressure, ultrafiltration, and blood flow rates. One of the key contributions of this study is that these features are calculated at different temporal scales and fed to the model, rather than using the typical baseline values. The proposed method was validated using both balanced and unbalanced data extracted from a public dataset. The results showed that XGBoost outperformed other algorithms, achieving 93.4% accuracy and 98.1% AUC. The method was tested against multiple definitions of Intradialytic Hypotension, outperforming all recent literature. The study also presents a dive into the developed model AI explainability and provides insight justifying how the proposed technique achieves such high-performance metrics and what key features are crucial for its results.</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

Intradialytic hypotension detection using explainable machine learning techniques aided with multi-scale temporal features

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Amjed Al-Mousa,
  • Rajaa Alqudah,
  • Tala Ibraheem

摘要

Intradialytic Hypotension is one of the potential complications of Kidney dialysis. Medical professionals use all possible means to avoid such complications. This study presents a machine learning-based method, the Intradialytic Hypotension Detection model using Multi-Scale Temporal Features (IDH-MSTF), which can be used continuously during the dialysis session to predict the occurrence of Intradialytic Hypotension. The proposed method utilizes 32 readily available features, including systolic and diastolic blood pressure, ultrafiltration, and blood flow rates. One of the key contributions of this study is that these features are calculated at different temporal scales and fed to the model, rather than using the typical baseline values. The proposed method was validated using both balanced and unbalanced data extracted from a public dataset. The results showed that XGBoost outperformed other algorithms, achieving 93.4% accuracy and 98.1% AUC. The method was tested against multiple definitions of Intradialytic Hypotension, outperforming all recent literature. The study also presents a dive into the developed model AI explainability and provides insight justifying how the proposed technique achieves such high-performance metrics and what key features are crucial for its results.