An automated diagnostic workflow integrating machine learning with radial pulse wave analysis has demonstrated high exploratory accuracy in identifying menstrual disorders, according to a cross-sectional study published in Frontiers in Digital Health. Conducted by researchers Ying Chen, Shuang Cheng, bo huang Lin, and yang shao Men at Guangzhou University of Chinese Medicine, the investigation evaluated whether objective physiological pulse measurements could address the diagnostic subjectivity historically associated with traditional pulse assessment. Among six prespecified computational models tested on a cohort of 246 participants, a Linear Discriminant Analysis model with shrinkage achieved an area under the receiver operating characteristic curve of 0.977 and an overall accuracy of 0.986 in distinguishing between affected individuals and healthy controls.
The research team established their investigation against the background of diagnostic challenges in women’s reproductive health. Menstrual disorders represent a common health condition among young women and correlate clinically with endocrine abnormalities as well as elevated cardiovascular risks. Despite the prevalence and potential systemic consequences of these conditions, conventional clinical detection continues to rely heavily on subjective evaluations due to a lack of objective, non-invasive diagnostic tools. While Traditional Chinese Medicine pulse diagnosis provides holistic information regarding systemic health status, its clinical application has long been constrained by subjectivity and variability among practitioners, underscoring the need for standardized quantitative measurement techniques.
To address these diagnostic limitations, the researchers designed a cross-sectional study involving 246 participants who were divided into a menstrual disorders group and a normal control group. The study protocol involved collecting radial pulse wave data alongside Traditional Chinese Medicine constitution information from every enrolled participant across both cohorts. By gathering both physical arterial pulse recordings and constitutional profiles under standardized conditions, the investigation sought to capture physiological variations that characterize menstrual pathology compared to normal health.
The analytical framework implemented strict methodological controls to ensure the validity and generalizability of the machine learning algorithms. All data preprocessing steps, feature selection procedures, and hyperparameter tuning operations were executed exclusively within the training set using nested cross-validation, preventing data leakage into the evaluation phase. The investigators prespecified and systematically compared six distinct machine learning algorithms: Logistic Regression with ElasticNet (LR-ElasticNet), Linear Discriminant Analysis with Shrinkage (LDA-Shrinkage), standard Logistic Regression (LR), Support Vector Machines (SVM), Decision Trees (DT), and Extreme Gradient Boosting (XGBoost). Model evaluation across the cohorts was comprehensively assessed using accuracy, recall, specificity, precision, F1 score, and the area under the receiver operating characteristic curve (AUC), with 95 percent confidence intervals calculated for each metric.
Statistical analysis revealed significant baseline differences in both arterial pulse wave parameters and constitutional patterns between the menstrual disorders group and the normal cohort. Between-group comparisons were executed using the Mann–Whitney U test, with the Benjamini–Hochberg procedure applied to control the false discovery rate across multiple comparisons (P < 0.05, FDR-adjusted q < 0.05). Participants diagnosed with menstrual disorders predominantly displayed a Yin deficiency constitution in comparison to the healthy participants. Furthermore, significant differences emerged across multiple pulse wave parameters: the menstrual disorders group exhibited increased bilateral w, increased bilateral w/t, increased bilateral h3/h1 ratios, and increased left-side h5 values, accompanied by decreased bilateral h4/h1 ratios, decreased left-side h3 values, and decreased ad parameters.
Among the six prespecified machine learning algorithms, the LDA-Shrinkage model was selected as the optimal classifier based on its superior predictive accuracy, calibration, and interpretability. The LDA-Shrinkage model attained an AUC of 0.977 (95% CI 0.919–1.000), an overall accuracy of 0.986, an F1 score of 0.989, a precision of 0.978, a recall of 1.000, and a specificity of 0.966. In addition to achieving the highest discrimination metrics, LDA-Shrinkage demonstrated the most favorable calibration curve and explanatory interpretability among the tested algorithms. To interpret the model’s decision-making process, the investigators performed SHAP (Shapley Additive Explanations) analysis, which identified the right-side w/t ratio (Rw/t) and the left-side h4/h1 ratio (Lh4/h1) as the most influential candidate features, proposing them as core observational parameters for future diagnostic research.
The authors concluded that combining radial pulse wave analysis with machine learning algorithms demonstrates promising exploratory performance for non-invasive menstrual disorder screening in clinical practice. The observed Traditional Chinese Medicine constitutional variations, particularly the predominance of Yin deficiency, provide valuable mechanistic insights into disease heterogeneity that could support personalized treatment strategies. By pairing machine learning classification with physiological pulse wave profiling and constitutional analysis, the researchers stated that these findings contribute toward the objective standardization of Traditional Chinese Medicine pulse diagnosis while offering a non-invasive screening tool for clinical practice.
Subject of Research: Machine learning screening of menstrual disorders using radial pulse wave analysis and Traditional Chinese Medicine constitution
Article Title: Machine learning-assisted radial pulse wave analysis as a novel diagnostic tool for menstrual disorders: a cross-sectional study
Article References: Chen, Y., Cheng, S., Lin, H., & Men, S. (2026). Machine learning-assisted radial pulse wave analysis as a novel diagnostic tool for menstrual disorders: a cross-sectional study. Frontiers in Digital Health, 8, 1964508. https://doi.org/10.3389/fdgth.2026.1964508
Image Credits: AI Generated
DOI: 10.3389/fdgth.2026.1964508
Keywords: menstrual disorders, radial pulse wave, Traditional Chinese Medicine, pulse diagnosis, machine learning, LDA-Shrinkage, Yin deficiency, SHAP analysis, cross-sectional study, nested cross-validation, endocrine abnormalities, cardiovascular risks, diagnostic screening, Rw/t parameter, Lh4/h1 ratio, personalized medicine, Frontiers in Digital Health
News Source: Ophelia Keating. (November 16, 2026). Machine Learning Analyzes Radial Pulse Waves to Identify Menstrual Disorders in Study. Scienmag.



