Cancer Research and Treatment

Original Articles

Multimodal Knowledge Graph–Guided RAG-LLM for Clinical Decision Support in Pediatric Leukemia: Jong Keon Song, Dong Bin Youk, Hyery Kim, Sang-Hyun Hwang; Received January 15, 2026 Accepted April 5, 2026 Published online April 21, 2026; DOI: https://doi.org/10.4143/crt.2026.0047 [Accepted]

Abstract PDF Supplementary Material: Purpose
This study aims to develop and evaluate a multimodal, knowledge graph–guided retrieval-augmented generation (RAG) framework for clinical decision support in pediatric acute leukemia.
Materials and Methods
Authoritative pediatric hematology-oncology textbooks were decomposed into text, tables, and figures. Visual and tabular elements were converted into structured textual descriptions using a multimodal large language model (LLM). A biomedical knowledge graph was constructed using LightRAG with gpt-oss-20b and Qwen3 embeddings. System performance was evaluated using 10 clinical questions, with responses generated by the RAG system and GPT-4.5. Nine medical experts (4 pediatric hematology-oncology specialists, 3 nurse specialists, and 2 medical students) conducted blind evaluations, complemented by two LLM evaluators (Claude Sonnet 4.5 and Gemini 3).
Results
The knowledge graph comprised 10,062 nodes and 15,876 edges. In expert evaluation, RAG was preferred in 47.8% of 90 paired comparisons versus 35.6% for GPT-4.5, with higher completeness scores (3.84 vs 3.51, p = 0.016). RAG showed significant advantage for ETP-ALL immunophenotype definition (p = 0.016). LLM-based evaluation consistently favored RAG: Claude Sonnet 4.5 preferred RAG in 6 of 10 questions, and Gemini 3 in 9 of 10 (Fast mode) and 7 of 10 (Thinking mode).
Conclusion
Multimodal graph-based RAG is feasible for clinical decision support in pediatric leukemia. RAG showed complementary strengths to foundation model LLMs, providing added value for questions requiring evidence-dependent information. Unlike LLMs with static training knowledge, RAG can incorporate updated guidelines and protocols without model retraining, particularly relevant in rapidly evolving fields. Further validation regarding privacy and regulatory issues is required before clinical deployment.

563 View
38 Download

Hematologic malignancy

Literature-Guided 6-Gene Signature for the Stratification of High-Risk Acute Myeloid Leukemia: Jong Keon Song, Dong Hyeok Lee, Hyery Kim, Sang-Hyun Hwang; Cancer Res Treat. 2025;57(4):1207-1217. Published online January 24, 2025; DOI: https://doi.org/10.4143/crt.2024.1114

Abstract PDF Supplementary Material PubReader ePub

Purpose
Acute myeloid leukemia (AML) shows significant heterogeneity in therapeutic responses. We aimed to develop a gene signature for the stratification of high-risk pediatric AML using publicly available AML datasets, with a focus on literature-based prognostic gene sets.
Materials and Methods
We identified 300 genes from 12 well-validated studies on AML-related gene signatures. Clinical and gene expression data were obtained from three datasets: TCGA-LAML, TARGET-AML, and BeatAML. Least absolute shrinkage and selection operator–Cox regression analysis was used to perform the initial gene selection and to construct a prognostic model using the The Cancer Genome Atlas (TCGA) database (n=132). The final gene signature was validated with two independent cohorts: BeatAML (n=411) and TARGET-AML (n=187).
Results
We identified a six-gene signature (ETFB, ARL6IP5, PTP4A3, CSK, HS3ST3B1, PLA2G4A), referred to as the literature-based signature 6 (LBS6), that was significantly associated with lower overall survival rates across the TCGA (high-risk [HR], 4.2; 95% confidence interval [CI], 2.59 to 6.81; p < 0.001), BeatAML (HR, 1.52; 95% CI, 1.17 to 1.96; p=0.001), and TARGET (HR, 2.05; 95% CI, 1.36 to 3.08; p < 0.001) datasets. The high-LBS6 score group exhibited significantly poorer five-year event-free survival compared to the low-LBS6 score group (HR, 2.09; 95% CI, 1.38 to 3.15; p < 0.001). After adjusting for key risk factors, including gene mutations (WT1, FLT3, and NPM1), protocol-based risk group, white blood cell count, and age, the LBS6 score was independently associated with worse survival rates in validation cohorts.
Conclusion
Our literature-driven approach identified a robust gene signature that stratifies AML patients into distinct risk groups. The LBS6 score shows promise in redefining initial risk stratification and identifying high-risk AML patients.

Citations

Citations to this article as recorded by

Integrated network propagation identifies prognostic metabolic signatures in acute myeloid leukemia
Jong Keon Song, Hyery Kim, Sang-Hyun Hwang
Journal of Translational Medicine.2025;[Epub] CrossRef

3,804 View
176 Download
1 Crossref

First
Prev
Page of 1
Next
Last

Search