Key result
Transformer model applied to PPG signals detects hypertension with ~96% accuracy.
Why the study?
Most blood pressure monitors use cuff inflation and deflation, which may cause discomfort and anxiety that impact blood pressure measurements, motivating a non-invasive, cuffless method for continuous blood pressure monitoring.
Does a custom Transformer deep learning model with Recursive Feature Elimination improve the accuracy of non-invasive, cuffless hypertension detection from photoplethysmography signals?
Does a custom Transformer deep learning model with Recursive Feature Elimination improve the accuracy of non-invasive, cuffless hypertension detection from photoplethysmography signals?
Answers are grounded in this paper and cite their sources.
A Transformer-based deep learning model combined with Recursive Feature Elimination achieved 96% accuracy, precision, recall, and F1-score in detecting hypertension from non-invasive, cuffless photoplethysmography signals.
Hypertension is a primary cause of mortality in humans, imposing a significant burden on society, families, and individuals. Real-time monitoring and prediction of HBP is essential for prompt intervention and treatment which invariably can lower death rates. When it comes to the identification of illnesses, having access to an automated disease detection system is extremely beneficial to medical personnel. This is because it helps them to give reliable, efficient, and prompt responses, while also reducing the mortality rate. The majority of blood pressure monitors use cuff inflation and deflation to calculate the systolic and diastolic arterial blood pressure. The user may experience discomfort and anxiety as a result of this approach, which will impact the measurement of blood pressure. This study focuses on the continuous measurement of the BP using a non-invasive, cuffless method. The approach of this paper is based on Transformer model applied to photoplethysmography signals from the wellknown time series dataset from Kaggle. We devised a data preprocessing technique that incorporates physiological signal transformations and also utilized Recursive Feature Elimination (RFE) for the feature selection process. The model’s performance is evaluated based on a number of criteria such as accuracy, precision, F1 Score, and recall. The result showed classification accuracy of 96% and a remarkable value for precision and recall of 96% and 96% respectively.
No takes yet. Share an insight, caveat, or question.
Olulajiki et al. (2025) studied Hypertension. Transformer-based deep learning model with Recursive Feature Elimination (RFE) on PPG signals vs. Conventional machine learning model was evaluated on Hypertension classification accuracy. A custom Transformer model applied to photoplethysmography signals detected hypertension with 96% accuracy, 96% precision, and 96% recall.
Synapse has enriched 10 related papers. Consider them for comparative context: