환자 간 고차원적 표현을 반영한 그래프 어텐션 신경망 기반 심장 질환 예측 모델
A Graph Attention Network-based Heart Disease Prediction Model for Reflecting High-order Representations Between Patients
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 연구는 환자의 임상적, 병리학적 데이터를 고려하여 심장 질환 여부를 예측하기 위해 새로운 예측 모델로 심장 관계 그래프 신경망(HRGNN)을 제안한다. 구체적으로, 인구통계, 임상적 습관, 기저질환 총 3가지 유형의 임상 정보로 GAT(Graph Attention Network) 기반 심장병 예측 모델을 설계하고, 각 유형과 속성의 고차적 표현을 학습하기 위한 이중 어텐션 메커니즘을 제안한다. 실험에 따르면 HRGNN은 노드 분류 작업에서 F1과 AUC 기준 기존 그래프 신경망(GNN) 및 기계 학습 모델보다 성능이 좋게 나옴을 확인할 수 있으며, 실험에 사용한 데이터 리샘플링 방법과 모델 구성요소 별 비교 실험을 통해 본 연구에서 제안하는 모델의 성능을 확인하였다.
Abstract
Heart disease is one of the leading causes of death in modern society. Over 50% of adults have at least one of three risk factors, high blood pressure, high cholesterol and smoking, for heart disease. Also, there are other major causes including diabetes, obesity, lack of exercise and heavy drinking. Diagnosing heart disease is a complex medical problem that requires an analysis of a combination of clinical and pathological data. Meeting a need for a high level of clinical expertise for the analysis gives rise to excessive cost. Reducing the cost has become an important issue. To resolve the issue, we propose Heart Relation Graph Neural Network(HRGNN) model with a novel-designed architecture. Specifically, we design Graph Attention Network(GAT)-based heart disease prediction model with three-type clinical information and propose dual-level attention mechanism to learn high-order representation of each type and attribute. HRGNN is can learn the hidden patterns in the high-order graph structure, which is a framework considering the complex type-specific patient data correlations. Health survey data of 401,958 adults provided by the Centers for Disease Control and Prevention are used in this study. Our experiments show that HRGNN outperforms existing Graph Neural Networks(GNN) and traditional machine learning models at node classification tasks. Also, it is experimentally demonstrated that three-type branch, dual-level attention modules and data imbalance resolving methods applied to classification models improve the performance.
참고문헌 (41)
- "World Health Organization. Cardiovascular diseases (CVDs)"[URL]
- "The cause of death statistics in Koreans"[URL]
- H. S. Shon, "Bayesian classification method for diagnosing heart disease", Proceedings of the 25th Spring Conference of The Korean Information Processing Society, pp.39~42 (2006)
- S. Mohan, "Effective heart disease prediction using hybrid machine learning techniques", IEEE Access, 7, pp.81542~81554 (2019)
- A. Gavhane, "Prediction of heart disease using machine learning", 2018 Second International Conference on Electronics, Communication and Aerospace Technology (ICECA), pp.1275~1278 (2018)
- M. Durairaj, "Prediction of heart disease using back propagation MLP algorithm", International Journal of Scientific & Technology Research, 4(8), pp.235~239 (2015)
- P. H. Guzzi, "Using graphs to relate patient’s clinical and geographical data", Procedia Computer Science, 110, pp.448~452 (2017)
- D. Jarrett, "Applications and limitations of machine learning in radiation oncology", The British journal of radiology, 92(1100) (2019)
- Z. H. Zhou, "Training cost-sensitive neural networks with methods addressing the class imbalance problem", IEEE Trans. Knowl. Data Eng, 18(1), pp.63~77 (2006)
- 강경호, "A Prediction Algorithm Using a Combined Heart Rate and Movement Index to Estimate the Energy Expenditure of Anaerobic Exercises", 한국지식정보기술학회 논문지, 10(2), pp.261~269 (2015), 한국지식정보기술학회
- M. H. Oh, "Comparison of the SVM-based regression analysis for prediction of recurrence risk factors of heart disease and diabetes", The Journal of Information Technology and Marketing, pp.65~67 (2014)
- P. W. Park, "Classification of heart disease using K-nearest neighbor imputation", Proceedings of the Korea Information Processing Society Conference, pp.742~745 (2017)
- 김철진, "딥러닝 기반의 다층 퍼셉트론을 이용한 심장병 예측 연구", 한국지식정보기술학회 논문지, 13(4), pp.411~419 (2018), 한국지식정보기술학회[DOI]
- A. K. Gárate-Escamila, "Classification models for heart disease prediction using feature selection and PCA", Informatics in Medicine Unlocked, 19, pp.100330~ (2020)
- F. Scarselli, "The graph neural network model", IEEE transactions on neural networks, 20(1), pp.61~80 (2009)
- Z. Wu, "A comprehensive survey on graph neural networks", IEEE transactions on neural networks and learning systems, 32(1), pp.4~24 (2020)
- M. Welling, "Semi-supervised classification with graph convolutional networks", J. International Conference on Learning Representations(ICLR) (2016)
- P. Veličković, "Graph attention networks", ~ (2017)
- S. Parisot, "Disease prediction using graph convolutional networks: application to autism spectrum disorder and Alzheimer’s disease", Medical image analysis, 48, pp.117~130 (2018)
- T. B. Mudiyanselage, "Predicting CircRNA disease associations using novel node classification and link prediction models on Graph Convolutional Networks", Methods, 198, pp.32~44 (2022)
- L. Wang, "gGATLDA:lncRNA-disease association prediction based on graph-level graph attention network", BMC bioinformatics, 23(1), pp.1~24 (2022)
- S. Mohan, "Effective heart disease prediction using hybrid machine learning techniques", IEEE access, 7, pp.81542~81554 (2019)
- D. Shah, "Heart disease prediction using machine learning techniques", SN Computer Science, 1(6), pp.1~6 (2020)
- S. Palaniappan, "Intelligent heart disease prediction system using data mining techniques", 2008 IEEE, 2008, 108~115
- C. Kalaiselvi, "Prediction of heart diseases and cancer in diabetic patients using data mining techniques", Indian Journal of Science and Technology, 1 (2015)
- J. C. Gower, "A general coefficient of similarity and some of its properties", pp.857~871 (1971), Biometrics
- J. Gilmer, "Neural message passing for quantum chemistry", International conference on machine learning, pp.1263~1272 (2017)
- A. L. Maas, "Rectifier nonlinearities improve neural network acoustic models", The 30th International Conference on Machine Learning(ICML), 30(1), pp.3~ (2013)
- S. Brody, "How attentive are graph attention networks?", ~ (2021)
- J. Han, "The influence of the sigmoid function parameters on the speed of backpropagation learning", International workshop on artificial neural networks, pp.195~201 (1995)
- D. Mesquita, "Rethinking pooling in graph neural networks", Advances in Neural Information Processing Systems, 33, pp.2220~2231 (2020)
- I. S. Dhillon, "Weighted graph cuts without eigenvectors a multilevel approach", IEEE transactions on pattern analysis and machine intelligence, 29(11), pp.1944~1957 (2007)
- V. Nair, "Rectified linear units improve restricted boltzmann machines", Icml (2010)
- M. Zeng, "Effective prediction of three common diseases by combining SMOTE with Tomek links technique for imbalanced medical data", 2016 IEEE International Conference of Online Analysis and Computing Science, pp.225~228 (2016)
- T. Y. Lin, "Focal loss for dense object detection", Proceedings of the IEEE international conference on computer vision, pp.2980~2988 (2017)
- X. Glorot, "Understanding the difficulty of training deep feedforward neural networks", Proceedings of the thirteenth international conference on artificial intelligence and statistics, JMLR Workshop and Conference Proceedings, pp.249~256 (2010)
- F. Pedregosa, "Scikit-learn: Machine learning in Python", the Journal of machine Learning research, 12, pp.2825~2830 (2011)
- V. Sharma, "Prediction of Heart disease using DNN", Second International Conference on Inventive Research in Computing Applications, pp.554~562 (2020)
- D. W. Hosmer Jr, "Applied logistic regression" (2013), John Wiley & Sons
- T. M. Padmaja, "Unbalanced data classification using extreme outlier elimination and sampling techniques for fraud detection", 15th International Conference on Advanced Computing and Communications, pp.511~516 (2007), IEEE
- B. Bengfort, "Yellowbrick:visualizing the scikit-learn model selection process", Journal of Open Source Software, 4(35), pp.1075~ (2019)