Deep learning ensemble with asymptotic techniques for oscillometric blood pressure estimation

Area of Science:

• Biomedical Engineering
• Machine Learning
• Medical Devices

Background:

• Oscillometric blood pressure (BP) measurements can have inherent uncertainties.
• Traditional deep learning models like Deep Belief Networks (DBN) and Deep Neural Networks (DNN) face challenges with small sample sizes and estimator selection, potentially leading to overfitting or underfitting.
• Accurate BP estimation is crucial for clinical decision-making.

Purpose of the Study:

• To develop a deep learning-based ensemble regression estimator to decrease uncertainty in oscillometric BP measurements.
• To utilize bootstrap and Monte-Carlo methods for estimating systolic blood pressure (SBP) and diastolic blood pressure (DBP), and their confidence intervals (CIs).
• To improve the accuracy and reduce the uncertainty of BP estimation.

Main Methods:

• Employed an ensemble approach combining bootstrap aggregation with DBN-DNN techniques to generate pseudo-features for BP estimation.
• Utilized AdaBoost for the second-stage SBP and DBP estimation.
• Applied bootstrap and Monte-Carlo techniques to determine confidence intervals for the estimated BP values.

Main Results:

• The proposed DBN-DNN ensemble regression estimator significantly mitigated estimation uncertainty.
• Standard deviations of error (SDE) for SBP and DBP were reduced by 0.58 mmHg and 0.57 mmHg, respectively, compared to single DBN-DNN estimators.
• Performance enhancement of 9.18% for SBP and 10.88% for DBP was observed.

Conclusions:

• The proposed methodology effectively improves the accuracy of BP estimation.
• The ensemble approach successfully reduces uncertainty associated with oscillometric BP measurements.
• This technique offers a more reliable method for BP assessment.