Knowledge Extraction and Representation Learning from Medical Text Data: Application to Resuscitation Data Analysis

16 Pages Posted: 11 Oct 2022

See all articles by Mohanad Abukmeil

Mohanad Abukmeil

University of Stavanger

Øyvind Meinich-Bache

University of Stavanger

Trygve Christian Eftestøl

University of Stavanger

Hege Ersdal

University of Stavanger - Stavanger University Hospital

Siren Rettedal

University of Stavanger - Stavanger University Hospital

Helge Myklebust

affiliation not provided to SSRN

Estomih Mduma

Haydom Lutheran Hospital

thomas bailey

University of Stavanger - Stavanger University Hospital

Kjersti Engan

University of Stavanger

Abstract

Deprivation of oxygen in an infant during and after birth leads to birth asphyxia, which is considered one of the leading causes of newborn death. Adequate resuscitation activities are performed immediately after birth to save the majority of the newborns. The main resuscitation activities include ventilation, stimulation, drying, suction, and chest compression. Resuscitation guidelines exist, but little research has been conducted on measured resuscitation episodes. Modeling the performed resuscitation activities to generate temporal data and extract knowledge can give a unique insight into dominant resuscitation activities, and it also assists in building a resuscitation timeline to visualize and describe the activities performed on a newborn. In this paper, we propose a novel method to generate and encode the temporal resuscitation data to describe and visualize the resuscitation timeline. Besides the timeline visualization, we use neighborhood component analysis (NCA) to cluster the generated data based on the presence of ventilation, and the outcome of the newborn. Additionally, we use an autoencoder (AE) model to improve the clustering performance by visualizing its latent space. Our proposed method shows high-quality visual clustering results on two different datasets, where it can draw an insight of the complex structure of the generated resuscitation data through grouping the similar unlabeled resuscitation episodes to joint clusters.

Note:

Funding Information: This work is part of the NewbornTime project funded by the Norwegian research council (NRC) project number 320968 and Fondation Idella. Some of the data is collected during the Safer Births project which has received funding from Laerdal Foundation, Laerdal Global Health, Skattefunn, Norwegian Ministry of Education and USAID. The Safer Births project was partly supported by NRC through the Global Health and Vaccination program (GLOBVAC) project number 228203.

Declaration of Interests: Consulting fees via Stavenger university hospital, Participation on a Data Safety Monitoring Board or Advisory Board: Regional Committee for Medical and Health Research Ethics (REK) in Norway. no other competing interests.

Ethics Approval Statement: The Safer Births study was ethically approved prior to implementation by the National Institute for Medi cal Research (NIMR) in Tanzania : NIMR/HQ/R.8a/Vol. IX/3852 and the Regional Committee for Medical and Health Research Ethics (REK) in Norway, REK West number: 172126. The NeoBeat is approved by REK West, number:2018/338. The NewbornTime is approved by REK West, number: 222455.

Keywords: Newborn, Resuscitation Activities, Visualization, Clustering, Autoencoder

Suggested Citation

Abukmeil, Mohanad and Meinich-Bache, Øyvind and Eftestøl, Trygve Christian and Ersdal, Hege and Rettedal, Siren and Myklebust, Helge and Mduma, Estomih and bailey, thomas and Engan, Kjersti, Knowledge Extraction and Representation Learning from Medical Text Data: Application to Resuscitation Data Analysis. Available at SSRN: https://ssrn.com/abstract=4224297 or http://dx.doi.org/10.2139/ssrn.4224297

Mohanad Abukmeil (Contact Author)

University of Stavanger ( email )

Øyvind Meinich-Bache

University of Stavanger ( email )

PB 8002
Stavanger, 4036
Norway

Trygve Christian Eftestøl

University of Stavanger ( email )

PB 8002
Stavanger, 4036
Norway

Hege Ersdal

University of Stavanger - Stavanger University Hospital ( email )

Siren Rettedal

University of Stavanger - Stavanger University Hospital ( email )

Helge Myklebust

affiliation not provided to SSRN ( email )

No Address Available

Estomih Mduma

Haydom Lutheran Hospital ( email )

P.O Box 9000
Haydom
Tanzania

Thomas Bailey

University of Stavanger - Stavanger University Hospital ( email )

Armauer Hansens vei 20
Stavanger, 4011
Norway

Kjersti Engan

University of Stavanger ( email )

PB 8002
Stavanger, 4036
Norway

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
61
Abstract Views
206
Rank
629,677
PlumX Metrics