Bayesian optimisation of personalised models for patient vital-sign monitoring
Gaussian process regression (GPR) provides a means to generate flexible personalised models of timeseries of patient vital signs. These models can perform useful clinical inference in ways that population-based models cannot. A challenge for the use of personalised models is that they must be amenab...
| Main Authors: | , , |
|---|---|
| Format: | Journal article |
| Published: |
IEEE
2017
|
| _version_ | 1797070859731468288 |
|---|---|
| author | Colopy, GW Roberts, SJ Clifton, DA |
| author_facet | Colopy, GW Roberts, SJ Clifton, DA |
| author_sort | Colopy, GW |
| collection | OXFORD |
| description | Gaussian process regression (GPR) provides a means to generate flexible personalised models of timeseries of patient vital signs. These models can perform useful clinical inference in ways that population-based models cannot. A challenge for the use of personalised models is that they must be amenable to a wide range of parameterisations, to accommodate the plausible physiology of any patient in the population. Additionally, optimal performance is typically achieved when models are regularised in light of knowledge of the physiology of the individual patient. In this paper, we describe a method to build GP models with varying complexity (via covariance kernels) and regularisation (via fixed priors over hyperparameters) on a patient-specific level, for the purpose of robust vitalsign forecasting. To this end, our results present evidence in support of two main hypotheses: (i) the use of patientspecific models can out-perform population-based models for useful clinical tasks, such as vital-sign forecasting; and (ii) the optimal values of (hyper)parameters of these models are best determined by sophisticated methods of optimisation, due to high correlation between dimensions of the search space. The resulting models are sufficiently robust to inform clinicians of a patient’s vital-sign trajectory, and warn of imminent deterioration. |
| first_indexed | 2024-03-06T22:45:03Z |
| format | Journal article |
| id | oxford-uuid:5ce066f1-175d-4de7-80bd-64b2c0fd6068 |
| institution | University of Oxford |
| last_indexed | 2024-03-06T22:45:03Z |
| publishDate | 2017 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:5ce066f1-175d-4de7-80bd-64b2c0fd60682022-03-26T17:30:59ZBayesian optimisation of personalised models for patient vital-sign monitoringJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:5ce066f1-175d-4de7-80bd-64b2c0fd6068Symplectic Elements at OxfordIEEE2017Colopy, GWRoberts, SJClifton, DAGaussian process regression (GPR) provides a means to generate flexible personalised models of timeseries of patient vital signs. These models can perform useful clinical inference in ways that population-based models cannot. A challenge for the use of personalised models is that they must be amenable to a wide range of parameterisations, to accommodate the plausible physiology of any patient in the population. Additionally, optimal performance is typically achieved when models are regularised in light of knowledge of the physiology of the individual patient. In this paper, we describe a method to build GP models with varying complexity (via covariance kernels) and regularisation (via fixed priors over hyperparameters) on a patient-specific level, for the purpose of robust vitalsign forecasting. To this end, our results present evidence in support of two main hypotheses: (i) the use of patientspecific models can out-perform population-based models for useful clinical tasks, such as vital-sign forecasting; and (ii) the optimal values of (hyper)parameters of these models are best determined by sophisticated methods of optimisation, due to high correlation between dimensions of the search space. The resulting models are sufficiently robust to inform clinicians of a patient’s vital-sign trajectory, and warn of imminent deterioration. |
| spellingShingle | Colopy, GW Roberts, SJ Clifton, DA Bayesian optimisation of personalised models for patient vital-sign monitoring |
| title | Bayesian optimisation of personalised models for patient vital-sign monitoring |
| title_full | Bayesian optimisation of personalised models for patient vital-sign monitoring |
| title_fullStr | Bayesian optimisation of personalised models for patient vital-sign monitoring |
| title_full_unstemmed | Bayesian optimisation of personalised models for patient vital-sign monitoring |
| title_short | Bayesian optimisation of personalised models for patient vital-sign monitoring |
| title_sort | bayesian optimisation of personalised models for patient vital sign monitoring |
| work_keys_str_mv | AT colopygw bayesianoptimisationofpersonalisedmodelsforpatientvitalsignmonitoring AT robertssj bayesianoptimisationofpersonalisedmodelsforpatientvitalsignmonitoring AT cliftonda bayesianoptimisationofpersonalisedmodelsforpatientvitalsignmonitoring |