國立虎尾科技大學 |

Knowledge-Informed Neural Topic Modeling on Electronic Health Records and Biomedical Knowledge Graph.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	Knowledge-Informed Neural Topic Modeling on Electronic Health Records and Biomedical Knowledge Graph./
Author:	Zou, Yuesong.
Description:	1 online resource (92 pages)
Notes:	Source: Masters Abstracts International, Volume: 85-05.
Contained By:	Masters Abstracts International85-05.
Subject:	Electronic health records. -
Online resource:	click for full text (PQDT)
ISBN:	9798380705264

Knowledge-Informed Neural Topic Modeling on Electronic Health Records and Biomedical Knowledge Graph.
Zou, Yuesong.

Knowledge-Informed Neural Topic Modeling on Electronic Health Records and Biomedical Knowledge Graph. - 1 online resource (92 pages)

Source: Masters Abstracts International, Volume: 85-05.

Thesis (M.C.S.)--McGill University (Canada), 2023.

Includes bibliographical references

The rapid growth of biomedical knowledge and healthcare dataset opens up promising opportunities to understand human diseases in a systematic way. Inferring disease patterns, such as comorbidities, enables better decision-making in clinical practice. In this work, we identify several key challenges in healthcare data mining, then develop methods to deal with these aspects and demonstrate improvement.First, effective electronic health record (EHR) data mining has been hindered by poor interpretability of blackbox models and insufficient occurrence of some rare events. This work presents a neural topic model that effectively leverages exterior information to complement the lack of rare events information. The model is able to be interpreted through learned topics. Specifically, it distills latent disease topics from EHR data by learning the EHR code embedding from a constructed medical knowledge graph. The disease topics present frequently co-occurred and contextually relevant diseases and the drugs used in treatment for them.Second, though rich in amount, biomedical knowledge bases are usually utilized alone due to the diverse modalities. This work develops a multimodal guided graph topic model for a united biomedical knowledge graph including relations between disease, drug, and gene nodes. Under guidance, the learned topics matches provided disease topic anchors and collect information of multiple modalities related to the anchor.Overall, this work tackles different challenges of healthcare data mining by introducing existing knowledge and novel model design. Empirical results demonstrated that our proposed methods are effective and can be beneficial in various application tasks.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798380705264Subjects--Topical Terms:

1467061
Electronic health records.
Index Terms--Genre/Form:

554714
Electronic books.

Knowledge-Informed Neural Topic Modeling on Electronic Health Records and Biomedical Knowledge Graph.
LDR:05055ntm a22003377 4500 001 1147318
005 20240909100755.5
006 m o d
007 cr bn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798380705264
035 $a (MiAaPQ)AAI30718207
035 $a (MiAaPQ)McGill_2f75rf260
035 $a AAI30718207
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Zou, Yuesong. $3 1473018
245 1 0 $a Knowledge-Informed Neural Topic Modeling on Electronic Health Records and Biomedical Knowledge Graph.
264 0 $c 2023
300 $a 1 online resource (92 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Masters Abstracts International, Volume: 85-05.
500 $a Advisor: Li, Yue.
502 $a Thesis (M.C.S.)--McGill University (Canada), 2023.
504 $a Includes bibliographical references
520 $a The rapid growth of biomedical knowledge and healthcare dataset opens up promising opportunities to understand human diseases in a systematic way. Inferring disease patterns, such as comorbidities, enables better decision-making in clinical practice. In this work, we identify several key challenges in healthcare data mining, then develop methods to deal with these aspects and demonstrate improvement.First, effective electronic health record (EHR) data mining has been hindered by poor interpretability of blackbox models and insufficient occurrence of some rare events. This work presents a neural topic model that effectively leverages exterior information to complement the lack of rare events information. The model is able to be interpreted through learned topics. Specifically, it distills latent disease topics from EHR data by learning the EHR code embedding from a constructed medical knowledge graph. The disease topics present frequently co-occurred and contextually relevant diseases and the drugs used in treatment for them.Second, though rich in amount, biomedical knowledge bases are usually utilized alone due to the diverse modalities. This work develops a multimodal guided graph topic model for a united biomedical knowledge graph including relations between disease, drug, and gene nodes. Under guidance, the learned topics matches provided disease topic anchors and collect information of multiple modalities related to the anchor.Overall, this work tackles different challenges of healthcare data mining by introducing existing knowledge and novel model design. Empirical results demonstrated that our proposed methods are effective and can be beneficial in various application tasks.
520 $a La croissance rapide de la connaissance biomedicale et des ensembles de donnees de soins de sante ouvre des opportunites prometteuses pour comprendre les maladies humaines de maniere systematique. L'inference de schemas de maladies, tels que les comorbidites, permet une meilleure prise de decision en pratique clinique. Dans ce travail, nous identifions plusieurs defis cles dans l'exploration de donnees de soins de sante, puis developpons des methodes pour faire face a ces aspects et demontrer des ameliorations.Premierement, l'efficacite de l'exploration de donnees dossiers medicaux electroniques (DME) a ete entravee par l'interpretabilite mediocre des modeles de boite noire et la frequence insuffisante de certains evenements rares. Ce travail presente un modele de sujet neuronal qui tire efficacement parti de l'information exterieure pour completer le manque d'information sur les evenements rares. Le modele peut etre interprete grace aux sujets appris. Plus precisement, il distille des sujets de maladies latentes a partir des donnees de DME en apprenant l'integration de code de DME a partir d'un graphe de connaissances medicales construit. Les sujets de maladies presentent des maladies co-occurentes frequemment et des medicaments utilises dans leur traitement.Deuxiemement, bien que riches en quantite, les bases de connaissances biomedicales sont generalement utilisees seules en raison des modalites diverses. Ce travail developpe un modele de sujet de graphe guide multimodal pour un graphe de connaissances biomedicales unifiees comprenant des relations entre les noeuds de maladie, de medicament et de gene. Sous la direction, les sujets appris correspondent aux ancres de sujets de maladies fournies et collectent des informations de plusieurs modalites liees a l'ancre.Dans l'ensemble, ce travail aborde les differents defis de l'exploration de donnees de soins de sante en introduisant des connaissances existantes et une conception de modele innovante. Les resultats empiriques ont demontre que nos methodes proposees sont efficaces et peuvent etre benefiques dans diverses taches d'application.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Electronic health records. $3 1467061
650 4 $a Privacy. $3 575491
650 4 $a Medical coding. $3 1473019
650 4 $a Information science. $3 561178
655 7 $a Electronic books. $2 local $3 554714
690 $a 0723
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a McGill University (Canada). $3 845629
773 0 $t Masters Abstracts International $g 85-05.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30718207 $z click for full text (PQDT)