Using recurrent neural networks to predict colorectal cancer among patients

Ryan Amirkhan, Mark Hoogendoorn, Mattijs E. Numans, Leon Moons

Research output: Chapter in Book / Report / Conference proceedingConference contributionAcademicpeer-review

Abstract

Development of predictive models from Electronic Medical Records (EMRs) is a far from trivial task. Especially the temporal nature of health records is an aspect that is often ignored yet of utmost importance. Additionally, data is extremely sparse. Previous research has shown that the identification of temporal patterns from EMR data can be highly beneficial in the prediction of colorectal cancer (CRC). In this paper, we try to apply recurrent neural networks, and more specifically Long Short Term Memory (LSTM) networks and Gated Recurrent Units (GRUs) to see whether these networks could learn such valuable temporal patterns themselves and generate accurate predictive models for CRC. Results show that we attain performance on par with state-of-the-art algorithms (while being outperformed by one). The eventual Area under the ROC Curve (AUC) obtained is 0.811.

Original languageEnglish
Title of host publication2017 IEEE Symposium Series on Computational Intelligence (SSCI)
Subtitle of host publication[Proceedings]
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1-8
Number of pages8
ISBN (Electronic)9781538627259
ISBN (Print)9781538627273
DOIs
Publication statusPublished - 2018
Event2017 IEEE Symposium Series on Computational Intelligence, SSCI 2017 - Honolulu, United States
Duration: 27 Nov 20171 Dec 2017

Conference

Conference2017 IEEE Symposium Series on Computational Intelligence, SSCI 2017
Country/TerritoryUnited States
CityHonolulu
Period27/11/171/12/17

Funding

We want to thank the Julius Center, Department of General Practice of the Utrecht University Medical Center for their support in composing the dataset.

Fingerprint

Dive into the research topics of 'Using recurrent neural networks to predict colorectal cancer among patients'. Together they form a unique fingerprint.

Cite this