Gabriel Erion

Gabriel Erion

Auto-generated profile, based on publicly available information

Gabriel Erion

Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, USA

Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
17 January 2020 in Nature Machine Intelligence
Verified
Journal Article
Closed access
PDF file
SLScott M. Lundberg
Scott M. Lundberg
GEGabriel Erion
Gabriel Erion
HCHugh Chen
Hugh Chen
ADAlex DeGrave
Alex DeGrave
JPJordan M. Prutkin
Jordan M. Prutkin
SLSu-In Lee
Su-In Lee

8 December 2021 in npj Digital Medicine
Verified
Journal Article
Open access
PDF file
HCHugh Chen
Hugh Chen
SLScott M. Lundberg
Scott M. Lundberg
GEGabriel Erion
Gabriel Erion
JKJerry H. Kim
Jerry H. Kim
SLSu-In Lee
Su-In Lee
Hundreds of millions of surgical procedures take place annually across the world, which generate a prevalent type of electronic health record (EHR) data comprising time series physiological signals. Here, we present a transferable embedding method (i.e., a method to transform time series signals into input features for predictive machine learning models) named PHASE (PHysiologicAl Signal Embeddings) that enables us to more accurately forecast adverse surgical outcomes based on physiological signals. We evaluate PHASE on minute-by-minute EHR data of more than 50,000 surgeries from two operating room (OR) datasets and patient stays in an intensive care unit (ICU) dataset. PHASE outperforms other state-of-the-art approaches, such as long-short term memory networks trained on raw data and gradient boosted trees trained on handcrafted features, in predicting six distinct outcomes: hypoxemia, hypocapnia, hypotension, hypertension, phenylephrine, and epinephrine. In a transfer learning setting where we train embedding models in one dataset then embed signals and predict adverse events in unseen data, PHASE achieves significantly higher prediction accuracy at lower computational cost compared to conventional approaches. Finally, given the importance of understanding models in clinical applications we demonstrate that PHASE is explainable and validate our predictive models using local feature attribution methods.

7 April 2022 in Nature Biomedical Engineering
Verified
Journal Article
Closed access
PDF file
GEGabriel Erion
Gabriel Erion
JJJoseph D. Janizek
Joseph D. Janizek
CHCarly Hudelson
Carly Hudelson
RURichard B. Utarnachitt
Richard B. Utarnachitt
AMAndrew M. McCoy
Andrew M. McCoy
SLSu-In Lee
Su-In Lee

2 April 2020
Verified
Conference Paper
Open access
JJJoseph D. Janizek
Joseph D. Janizek
GEGabriel Erion
Gabriel Erion
ADAlex J. DeGrave
Alex J. DeGrave
SLSu-In Lee
Su-In Lee
While deep learning has shown promise in the domain of disease classification from medical images, models based on state-of-the-art convolutional neural network architectures often exhibit performance loss due to dataset shift. Models trained using data from one hospital system achieve high predictive performance when tested on data from the same hospital, but perform significantly worse when they are tested in different hospital systems. Furthermore, even within a given hospital system, deep learning models have been shown to depend on hospital- and patient-level confounders rather than meaningful pathology to make classifications. In order for these models to be safely deployed, we would like to ensure that they do not use confounding variables to make their classification, and that they will work well even when tested on images from hospitals that were not included in the training data. We attempt to address this problem in the context of pneumonia classification from chest radiographs. We propose an approach based on adversarial optimization, which allows us to learn more robust models that do not depend on confounders. Specifically, we demonstrate improved out-of-hospital generalization performance of a pneumonia classifier by training a model that is invariant to the view position of chest radiographs (anterior-posterior vs. posterior-anterior). Our approach leads to better predictive performance on external hospital data than both a standard baseline and previously proposed methods to handle confounding, and also suggests a method for identifying models that may rely on confounders.

1 November 2012 in Crop Science
Verified
Journal Article
Closed access
PDF file
GEGabriel G. Erion
Gabriel G. Erion
WRWalter E. Riedell
Walter E. Riedell

24 August 2015 in PLOS ONE
Verified
Journal Article
Open access
PDF file
SLShelley H. Liu
Shelley H. Liu
GEGabriel Erion
Gabriel Erion
VNVladimir Novitsky
Vladimir Novitsky
VGVictor De Gruttola
Victor De Gruttola
Analyses of viral genetic linkage can provide insight into HIV transmission dynamics and the impact of prevention interventions. For example, such analyses have the potential to determine whether recently-infected individuals have acquired viruses circulating within or outside a given community. In addition, they have the potential to identify characteristics of chronically infected individuals that make their viruses likely to cluster with others circulating within a community. Such clustering can be related to the potential of such individuals to contribute to the spread of the virus, either directly through transmission to their partners or indirectly through further spread of HIV from those partners. Assessment of the extent to which individual (incident or prevalent) viruses are clustered within a community will be biased if only a subset of subjects are observed, especially if that subset is not representative of the entire HIV infected population. To address this concern, we develop a multiple imputation framework in which missing sequences are imputed based on a model for the diversification of viral genomes. The imputation method decreases the bias in clustering that arises from informative missingness. Data from a household survey conducted in a village in Botswana are used to illustrate these methods. We demonstrate that the multiple imputation approach reduces bias in the overall proportion of clustering due to the presence of missing observations.