Scholarlink

Background <p>This study verified whether the dynamic changes in finger plethysmograms (FPGs) can be trained into a machine learning model to discriminate status changes associated with dialysis therapy.</p> Methods <p>FPGs were recorded in 18 participants undergoing dialysis therapy at the start of dialysis and during dialysis for 3&#xa0;min. The FPGs were converted to acceleration pulse waves, the indices obtained from the acceleration pulse waves were trained as features into an elastic net model (a type of explainable machine learning), and the performance of classification for distinguishing between the state change at the start of and during dialysis was verified.</p> Results <p>Throughout the dialysis therapy, significant changes were observed in d/a (an index of peripheral vascular resistance), PR, low frequency, low frequency/high frequency, Lyapunov index, entropy, systolic, and diastolic blood pressures. The classification performance of the elastic net model was as follows: precision, 88%; recall, 64%; F1 score, 74%; accuracy, 77%; and area under curve, 83%.</p> Conclusions <p>These results suggest that machine learning models trained on FPG-derived features offer a promising new technique for noninvasive and objective observation of hemodynamic changes during dialysis. However, given the small sample size (<i>n</i> = 18) and participant variability, these preliminary findings require validation in larger studies to establish their generalizability.</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

Investigating the usefulness of finger plethysmography for monitoring hemodynamic changes during dialysis therapy: a pre- and post-dialysis comparative study using machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月19日

查看全文
  • Yoshiaki Ito,
  • Takahashi Michiaki

摘要

Background

This study verified whether the dynamic changes in finger plethysmograms (FPGs) can be trained into a machine learning model to discriminate status changes associated with dialysis therapy.

Methods

FPGs were recorded in 18 participants undergoing dialysis therapy at the start of dialysis and during dialysis for 3 min. The FPGs were converted to acceleration pulse waves, the indices obtained from the acceleration pulse waves were trained as features into an elastic net model (a type of explainable machine learning), and the performance of classification for distinguishing between the state change at the start of and during dialysis was verified.

Results

Throughout the dialysis therapy, significant changes were observed in d/a (an index of peripheral vascular resistance), PR, low frequency, low frequency/high frequency, Lyapunov index, entropy, systolic, and diastolic blood pressures. The classification performance of the elastic net model was as follows: precision, 88%; recall, 64%; F1 score, 74%; accuracy, 77%; and area under curve, 83%.

Conclusions

These results suggest that machine learning models trained on FPG-derived features offer a promising new technique for noninvasive and objective observation of hemodynamic changes during dialysis. However, given the small sample size (n = 18) and participant variability, these preliminary findings require validation in larger studies to establish their generalizability.