SNP Position Filtering and Multimodal Approach in Heart Failure Risk Prediction
- 문서 유형
- 연구출판물 · 학술대회 발표논문
- 등재
- Scopus
- 저널
- Proceedings of the IEEE International Conference on Big Data and Smart Computing, BIGCOMPpp.337 · 2025
- 피인용
- Scopus 1
- FWCI (Scopus)
- 1.04
- ISSN
- 2375-933X
저자
- 신승윤제1저자Pusan National University, Department of Information Convergence Engineering, Busan, Republic of Korea
- 이수빈Pusan National University, Department of Information Convergence Engineering, Busan, Republic of Korea
- 김기범Pusan National University, Department of Information Convergence Engineering, Busan, Republic of Korea
- 이혜원Pusan National University, College of Medicine, Busan, Republic of Korea
- 송길태교신저자Pusan National University, College of Medicine, Busan, Republic of Korea
초록
Heart failure (HF) is a leading cause of mortality worldwide, with an 80% death rate within a year of diagnosis. Early identification of high-risk patients is crucial for timely interventions. Conventional HF prediction models based on electronic health records (EHR) face limitations due to irregular data intervals and missing information, while models based on genetic data like single nucleotide polymorphisms (SNPs) are difficult to learn and suffer from the curse of dimensionality. In this study, we propose a multimodal model combining EHR and SNP data to improve mortality prediction for HF patients. The model applies feature selection and filtering to SNP data, reducing its dimensionality to 1.7% to enhance trainability. We use Transformer architecture to process SNP sequences and a multilayer perceptron (MLP) to handle EHR data. Experimental results show that models trained on SNP data significantly outperform EHR-only models. However, our multimodal approach integrating both data types outperforms single modality models, demonstrating improved prediction accuracy. This work suggests that combining genomic and clinical data enhances the prediction of heart failure mortality, overcoming the limitations of traditional methods.
키워드
deep learningelectronic health recordsgenomicsHeart failuremortality predictionTransformer
식별자
- Scopus EID
- 2-s2.0-105006632171



로그인
연구물 목록으로