A machine learning approach for identification of vascular access patency in hemodialysis patients using photoplethysmography: A pilot study.

Introduction: Vascular access (VA) is essential for patients with hemodialysis, and its dysfunction is a major complication that can reduce quality of life or even threaten life. VA patency is not only difficult to predict on an individual basis, but also challenging to predict in real-time. To overcome this challenge, this study aimed to develop a machine learning approach to predict 6-month primary patency (PP) using photoplethysmography (PPG) signals acquired from the tips of both index fingers.

Materials and methods: PPG signals were obtained from hemodialysis patients who received an arteriovenous fistula or an arteriovenous graft as primary VA in a single center from April 2023 to December 2023. With PPG wearables, we propose a method that can efficiently and quickly generate the morphological features of the PPG signal to recognize different groups of patients. For the generated features, an independent sample t-test was used to evaluate their effectiveness for machine learning. Then, two supervised learning algorithms, k-nearest neighbors (kNN) and support vector machine (SVM), are used further to identify VA patency in advance.

Results: The study involved 31 patients, of whom 14 had 6-month PP, while 17 did not. Using the kNN algorithm, machine learning classified patients into two groups with 82% precision based on PPG signals, while the SVM algorithm showed a precision of 82%.

Conclusions: Our approach can provide reliable classifications for VA patency. It is effective to use the proposed PPG signal features to predict 6-month PP of VA.