Pathology Data Prioritisation: A Study of Using Multi-variate Time Series

Qi, Jing ORCID: 0000-0003-2476-7620, Burnside, Girvan ORCID: 0000-0001-7398-1346 and Coenen, Frans ORCID: 0000-0003-1026-6649 (2023) Pathology Data Prioritisation: A Study of Using Multi-variate Time Series. .

PDF Pathology Data Prioritisation_A Study of Using Multi-variate Time Series Without a Ground Truth.pdf - Author Accepted Manuscript Download (473kB) | Preview

Abstract

In any hospital, pathology results play an important role for decision making. However, it is not unusual for clinicians to have hundreds of pathology results to review on a single shift; this “information overload” presents a particular challenge. Some form of pathology result prioritisation is therefore a necessity. One idea to deal with this problem is to adopt the tools and techniques of machine learning to identify prioritisation patterns within pathology results and use these patterns to label new pathology data according to a prioritisation classification protocol. However, in most clinical situations there is an absence of any pathology prioritisation ground truth. The usage of supervised learning therefore becomes a challenge. Unsupervised learning methods are available, but are not considered to be as effective as supervised learning methods. This paper considers two mechanisms for pathology data prioritisation in the absence of a ground truth: (i) Proxy Ground Truth Pathology Data Prioritisation (PGR-PDP), and (ii) Future Result Forecast Pathology Data Prioritisation (FRF-PDP). The first uses the outcome event, what happened to a patient, as a proxy for a ground truth, and the second forecasted future pathology results compared with the known normal clinical reference range. Two variation of each are considered: kNN-based and LSTM-based PGR-PDP, and LSTM-based and Facebook Profit-based FRF-PDP. The reported evaluation indicated that the PGR-PDP mechanism produced the best results with little distinction between the two variations.

Item Type:	Conference or Workshop Item (Unspecified)
Uncontrolled Keywords:	Patient Safety
Divisions:	Faculty of Health and Life Sciences Faculty of Science and Engineering > School of Electrical Engineering, Electronics and Computer Science Faculty of Health and Life Sciences > Institute of Population Health
Depositing User:	Symplectic Admin
Date Deposited:	28 Sep 2023 09:38
Last Modified:	14 Mar 2024 21:44
DOI:	10.1007/978-3-031-35924-8_1
Related URLs:	Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3173126