Scholarlink

Background <p>Intradialytic hypotension (IDH) occurs in 20–40% of hemodialysis patients and is associated with adverse cardiovascular events and a poor prognosis. Although early prediction of and prompt intervention for IDH are essential, early detection can be challenging because it often relies on subjective assessments, such as changes in blood pressure trends, facial complexion, facial expressions, and patient-reported symptoms. In this study, we constructed two machine learning models to detect and predict IDH based on facial images captured during dialysis sessions, and identified key facial features associated with IDH.</p> Methods <p>Eight dialysis patients who frequently experienced IDH were included in this study. Facial images and blood pressure data were collected at 15&#xa0;min intervals from the start to the end of dialysis. In total, 14 facial features related to the eyes and mouth were extracted, including the eyebrow slope, inter-eyebrow area, eye aspect ratio, and the areas of the inner and outer lips. For each patient, a linear discriminant analysis model was used to construct IDH detection and prediction models. We evaluated the prediction accuracy and identified key facial regions that showed significant expression changes associated with IDH. This study was conducted with the approval of the Jichi Medical University Ethics Committee (approval no. 23-107).</p> Results <p>The accuracy rates of the detection and prediction models for IDH were 57.6–75.4% and 58.8–76.9%, respectively, with the precision rates ranging from 56.0–73.3% to 59.0–73.2%, recall rates ranging from 60.7–79.4% to 56.8–84.5%, and F-values ranging from 61.9–76.2% to 57.9–78.5%, respectively, demonstrating consistent accuracy. Notable facial expression changes for detection and prediction of IDH were observed: eye region changes were prominent in subjects A and F; mouth region changes were prominent in subjects B, C, and G; and both eye and mouth region changes were observed in subject D. In subjects E and H, different facial areas showed changes depending on the task—mouth region for prediction and eye region for detection.</p> Conclusions <p>In this study, we demonstrated the usefulness of detection and prediction models for IDH based on facial expression changes observed during hemodialysis. In particular, expressions involving the eyes and mouth were found to be important for predicting IDH, suggesting the potential usefulness of facial expression analysis as an effective supplementary tool to blood pressure monitoring in patients undergoing hemodialysis.</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

Analysis of facial expression changes for the prediction of dialysis-related hypotension using facial monitoring: a fundamental study on prediction accuracy and feature extraction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月20日

查看全文
  • Takehiro Okama,
  • Kousuke Oiwa,
  • Satoshi Suzuki,
  • Yoshitaka Maeda,
  • Akio Nozawa,
  • Koji Sakuma,
  • Akiko Uchida,
  • Toshiharu Ishizuka,
  • Sumihiko Sato

摘要

Background

Intradialytic hypotension (IDH) occurs in 20–40% of hemodialysis patients and is associated with adverse cardiovascular events and a poor prognosis. Although early prediction of and prompt intervention for IDH are essential, early detection can be challenging because it often relies on subjective assessments, such as changes in blood pressure trends, facial complexion, facial expressions, and patient-reported symptoms. In this study, we constructed two machine learning models to detect and predict IDH based on facial images captured during dialysis sessions, and identified key facial features associated with IDH.

Methods

Eight dialysis patients who frequently experienced IDH were included in this study. Facial images and blood pressure data were collected at 15 min intervals from the start to the end of dialysis. In total, 14 facial features related to the eyes and mouth were extracted, including the eyebrow slope, inter-eyebrow area, eye aspect ratio, and the areas of the inner and outer lips. For each patient, a linear discriminant analysis model was used to construct IDH detection and prediction models. We evaluated the prediction accuracy and identified key facial regions that showed significant expression changes associated with IDH. This study was conducted with the approval of the Jichi Medical University Ethics Committee (approval no. 23-107).

Results

The accuracy rates of the detection and prediction models for IDH were 57.6–75.4% and 58.8–76.9%, respectively, with the precision rates ranging from 56.0–73.3% to 59.0–73.2%, recall rates ranging from 60.7–79.4% to 56.8–84.5%, and F-values ranging from 61.9–76.2% to 57.9–78.5%, respectively, demonstrating consistent accuracy. Notable facial expression changes for detection and prediction of IDH were observed: eye region changes were prominent in subjects A and F; mouth region changes were prominent in subjects B, C, and G; and both eye and mouth region changes were observed in subject D. In subjects E and H, different facial areas showed changes depending on the task—mouth region for prediction and eye region for detection.

Conclusions

In this study, we demonstrated the usefulness of detection and prediction models for IDH based on facial expression changes observed during hemodialysis. In particular, expressions involving the eyes and mouth were found to be important for predicting IDH, suggesting the potential usefulness of facial expression analysis as an effective supplementary tool to blood pressure monitoring in patients undergoing hemodialysis.