Unified Clinical Vocabulary Embeddings for Advancing Precision Medicine DOI Creative Commons
Ruth Johnson,

Uri Gottlieb,

Galit Shaham

и другие.

medRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Дек. 5, 2024

Integrating clinical knowledge into AI remains challenging despite numerous medical guidelines and vocabularies. Medical codes, central to healthcare systems, often reflect operational patterns shaped by geographic factors, national policies, insurance frameworks, physician practices rather than the precise representation of knowledge. This disconnect hampers in representing relationships, raising concerns about bias, transparency, generalizability. Here, we developed a resource 67,124 vocabulary embeddings derived from graph tailored electronic health record vocabularies, spanning over 1.3 million edges. Using transformer neural networks, generated that provide new unifying seven These were validated through phenotype risk score analysis involving 4.57 patients Clalit Healthcare Services, effectively stratifying individuals based on survival outcomes. Inter-institutional panels clinicians evaluated for alignment with across 90 diseases 3,000 confirming their robustness transferability. addresses gaps integrating vocabularies models training datasets, paving way knowledge-grounded population patient-level models.

Язык: Английский

LLM-KGMQA: large language model-augmented multi-hop question-answering system based on knowledge graph in medical field DOI

FeiLong Wang,

Donghui Shi, José Aguilar

и другие.

Knowledge and Information Systems, Год журнала: 2025, Номер unknown

Опубликована: Апрель 21, 2025

Язык: Английский

Процитировано

0
LLM-KGMQA: Large Language Model-Augmented Multi-Hop Question-Answering System based on Knowledge Graph in Medical Field DOI Creative Commons

FeiLong Wang,

Donghui Shi, José Aguilar

и другие.

Research Square (Research Square), Год журнала: 2024, Номер unknown

Опубликована: Авг. 5, 2024

Abstract In response to the problems of poor performance large language models in specific domains, limited research on knowledge graphs and question-answering systems incorporating models, this paper proposed a multi-hop system framework based graph medical field, which was fully augmented by (LLM-KGMQA). The method primarily addressed entity linking path reasoning. To address problem, an fast-linking algorithm proposed, categorized entities multiple attributes. Then, it used user mentions obtain target attribute set attributes further narrowed search scope through intersection operations. Finally, for that remained too numerous after intersection, suggested using pre-trained model similarity calculation ranking, determine final construction instructions. Regarding reasoning, three-step reasoning included n-hop subgraph algorithm, fusion semantics-based pruning algorithm. experiments, maximum computational complexity reduced 99.9% Additionally, new evaluation metric, top@n, introduced. When Roberta calculations, top@n score reached 96.4, accuracy 96.6%. first validated need constructing three different forms Subsequently, experiments were conducted with several concluded GLM4 showed best Chinese semantic rates 99.9%, 83.3%, 86.6% 1-hop, 2-hop, 3-hop, respectively, compared 95.0%, 6.6%, 5.0% before pruning. average time 1.36s, 6.21s 27.07s

Язык: Английский

Процитировано

1
The Application of GCN Algorithm in Building Construction Knowledge Graph Updating under the Combination of Artificial Intelligence and Knowledge Management DOI Creative Commons
Lü He,

Hu Xu

International Journal of Cognitive Computing in Engineering, Год журнала: 2024, Номер unknown

Опубликована: Ноя. 1, 2024

Язык: Английский

Процитировано

1
Unified Clinical Vocabulary Embeddings for Advancing Precision Medicine DOI Creative Commons
Ruth Johnson,

Uri Gottlieb,

Galit Shaham

и другие.

medRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Дек. 5, 2024

Integrating clinical knowledge into AI remains challenging despite numerous medical guidelines and vocabularies. Medical codes, central to healthcare systems, often reflect operational patterns shaped by geographic factors, national policies, insurance frameworks, physician practices rather than the precise representation of knowledge. This disconnect hampers in representing relationships, raising concerns about bias, transparency, generalizability. Here, we developed a resource 67,124 vocabulary embeddings derived from graph tailored electronic health record vocabularies, spanning over 1.3 million edges. Using transformer neural networks, generated that provide new unifying seven These were validated through phenotype risk score analysis involving 4.57 patients Clalit Healthcare Services, effectively stratifying individuals based on survival outcomes. Inter-institutional panels clinicians evaluated for alignment with across 90 diseases 3,000 confirming their robustness transferability. addresses gaps integrating vocabularies models training datasets, paving way knowledge-grounded population patient-level models.

Язык: Английский

Процитировано

1