Pulse-line intersection method with unboxed artificial intelligence for hesitant pulse wave classification. Issue 2 (March 2022)

• Record Type: Journal Article
• Authors: Chen, Hsing-Chung; Damarjati, Cahya; Putra, Karisma Trinanda; Chen, Han-MI; Hsieh, Ching-Liang; Lin, Hung-Jen; Wu, Mei-Yao; Chen, Chin-Sheng
• Is Part Of: Information processing & management. Volume 59:Issue 2(2022)
• Publication Date: 2022-03
• Subjects: CDSS Clinical decision support systems -- PLI Pulse-Line Intersection -- XAI eXplainability AI; Clinical decision support systems -- Pulse-Line Intersection (PLI) -- eXplainability AI (XAI); Information storage and retrieval systems -- Periodicals; Information science -- Periodicals; Systèmes d'information -- Périodiques; Sciences de l'information -- Périodiques; Information science; Information storage and retrieval systems; Periodicals; 658.4038
• DOI: 10.1016/j.ipm.2021.102855
• Languages: English
• ISSNs: 0306-4573

Abstract

Highlights: In this paper, the pulse feature extraction namely pulse-line intersection method could generate statistically highly significant 11 features for distinguishing types of pulse. Moreover, SHAP (SHapley additive exPlanations) adopted in this paper as part of predictive intelligence component become a proof of the importance of each feature to predict hesitant pulse wave. The performance of long short-term memory (LSTM) overcome other prediction models i.e., LR, SVM, XGBoost, random Forest, and M in hesitant pulse wave prediction. This manuscript provides image preprocessing to convert pulse image into pulse wave data. Abstract: State-of-the-art artificial intelligence (AI) methods are progressively strengthened in Traditional Chinese Medicine (TCM) pulse palpation, aiding physicians to make comprehensive preliminary clinical decisions through non-invasive diagnostics. One of the well-known proven examinations i.e., hesitant pulse wave diagnosis, is a sign that the blood circulation of a person is sluggish. This examination provides a preliminary diagnosis for physiological problems. Modern AI methods such as artificial neural networks achieve better performance than traditional methods; however, the final decision of such examination lacks of interpretability. In clinical situations, patients need an easy-to-understand diagnosis to be provided for selecting appropriate clinical treatment. Therefore, this study presents feature extraction and clinical decision support systems based on Pulse-Line Intersection (PLI) and eXplainability AI (XAI) methods. The pulses were recorded from 46 patients in six different measurement points for six seconds. In addition, a comparison of several AI methods was provided to classify hesitant and normal pulse. The contribution of each feature in the classification process was analyzed by unboxing each predictive intelligence model. The results revealed that all models performed comparably, evaluated using performance matric on the testing data with average F1-score of Logistic Regression, Support Vector Machine, Random Forest, XGBoost, Multi-Layer Perceptron, and Long Short-Term Memory were 0.74, 0.74, 0.74, 0.78, 0.73, and 0.80, respectively. This work suggests that modern AI methods can provide more comprehensive explainability and higher accuracy than traditional method rankings.