Key result
A domain-adversarial neural network model using four minutes of training data estimated blood pressure with an average root mean square error of 4.64 mmHg for diastolic pressure, meeting ISO standards.
Why the study?
Building cuffless continuous ambulatory blood pressure regression models from bioimpedance sensors is limited by the prohibitive amount of data required to train them.
Does a domain-adversarial neural network model improve cuffless blood pressure estimation accuracy with minimal training data compared to standard training methods?
Does a domain-adversarial neural network model improve cuffless blood pressure estimation accuracy with minimal training data compared to standard training methods?
Mean Difference: -0.19
Absolute Event Rate: 4.64% vs 4.83%
A domain-adversarial neural network can accurately estimate continuous blood pressure from wearable bioimpedance sensors using only four minutes of personalized calibration data.
Cuffless ambulatory BP monitoring may enable frequent remote capture; leaves open accuracy, validation, and clinical adoption.
Blood pressure monitoring is an essential component of hypertension and in the prediction of associated comorbidities. Blood pressure is dynamic vital sign with frequent changes throughout a given day. Capturing pressure remotely and frequently (also known as ambulatory blood pressure) has traditionally been achieved by measuring blood pressure at intervals using an inflatable cuff. However, there is growing interest developing a cuffless ambulatory blood pressure monitoring system to measure pressure continuously. One such approach is by utilizing bioimpedance to build regression models. A practical problem with this approach is the amount of data required to confidently train such a regression model be prohibitive. In this paper, we propose the application of the-adversarial training neural network (DANN) method on our multitask (MTL) blood pressure estimation model, allowing for knowledge transfer subjects. Our proposed model obtains average root mean square error(RMSE) of 4.80 \± 0.74 mmHg for diastolic blood pressure and 7.34 \± 1.88 for systolic blood pressure when using three minutes of training data,4.64 \± 0.60 mmHg and 7.10 \± 1.79 respectively when using four minutes training data, and 4.48 \± 0.57 mmHg and 6.79 \± 1.70 respectively using five minutes of training data. DANN improves training with minimal in comparison to both directly training and to training with a pretrained from another subject, decreasing RMSE by $0.19$ to $0.26$ mmHg(diastolic) and by $0.46$ to $0.67$ mmHg (systolic) in comparison to the best models. We observe that four minutes of training data is the minimum for our framework to exceed ISO standards within this cohort of.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2020) studied Blood pressure estimation (n=11). Domain-adversarial neural network (DANN) model vs. Pretrained model from another subject was evaluated on Root mean square error (RMSE) for diastolic blood pressure (MD -0.19). A domain-adversarial neural network model using four minutes of training data estimated blood pressure with an average root mean square error of 4.64 mmHg for diastolic pressure, meeting ISO standards.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: