北京生物医学工程

基于多粒度级联森林的骨质疏松性骨折预测研究

Prediction of osteoporotic fracture based on multi-grained cascade forest

作者：徐辉煌张海宇林勇

单位：上海理工大学医疗器械与食品学院(上海 200093)

关键词：机器学习；骨质疏松性骨折； t分布邻域嵌入；随机森林；多粒度级联森林

分类号：R318；TP181

出版年·卷·期（页码）：2019·38·4（384-391）

摘要：

目的骨质疏松性骨折(osteoporotic fracture, OF)的预测对于骨折防范具有重要的临床指导意义。针对传统logistic回归预测模型存在的精度不高和未考虑遗传因子问题，本文引入多粒度级联森林(multi-grained cascade forest, gcForest)并结合遗传因子来预测OF。方法首先基于t分布邻域嵌入(t-distributed stochastic neighbor embedding, t-SNE)算法对OF关联基因位点进行非线性降维，降维后的基因位点与临床因素构成特征组。然后构建gcForest模型对OF进行预测。最后通过10次十折分层交叉验证与logistic、梯度提升决策树、随机森林进行对比。结果基于gcForest的模型分类精度为0.892 7，AUC值为0.92±0.05，泛化性能最优。结论在考虑遗传因素的条件下，gcForest分类效果优于其他模型，验证了本文方法的高效性和实用性。

Objective The prediction of osteoporotic fracture (OF) has important clinical guiding significance for fracture prevention. In view of the low precision of traditional logistic regression model and the lack of consideration of genetic factors, this paper introduces multi-grained cascade forest (gcForest) and combines genetic factors to predict the OF. Methods Firstly, based on the t-distributed stochastic neighbor embedding (t-SNE) method, the nonlinear descending dimension of the associated gene loci was carried out, and the gene loci and clinical factors were formed in the feature group after the reduction of the dimension. The gcForest model was then constructed to predict the OF. Finally, 10 times 10-fold stratified cross-validation was compared with logistic, gradient boosting decision tree and random forest. Results The model classification accuracy based on gcForest was 0.8927, AUC value was 0.92±0.05, and the generalization performance was optimal. Conclusions Under the condition of considering genetic factors, the gcForest classification effect is better than other models, which verifies the efficiency and practicability of the method.

参考文献：

[1] 章轶立, 魏戌, 申浩, 等. 骨质疏松性骨折诊断技术与风险预测工具研究进展[J]. 中国骨质疏松杂志, 2018,24(5):676-680.

Zhang YL, Wei X, Shen H, et al. Advances in diagnostic techniques and risk prediction tools for osteoporotic fractures[J]. Chinese Journal of Osteoporosis, 2018,24(5):676-680.

[2] 蔡淑芬, 邢其丹, 丰吉南, 等. 老年类风湿关节炎患者发生骨质疏松的危险因素分析 [J]. 中国骨质疏松杂志, 2018,24(7):922-925, 939.

Cai SF, Xing QD, Feng JN, et al. Analysis of the risk factors of osteoporosis in elderly patients with rheumatoid arthritis[J]. Chinese Journal of Osteoporosis, 2018, 24(7):922-925, 939.

[3] 郑煜晖, 吴世强, 庄华峰. 老年骨质疏松患者骨代谢指标、骨密度与骨质疏松性骨折的相关性[J]. 中国老年学杂志, 2017,37(23):5920-5922.

[4] 章轶立, 魏戌, 聂佩芸, 等. 基于Group Lasso的Logistic回归模型构建绝经后骨质疏松性骨折初发风险评估工具[J]. 中国骨质疏松杂志, 2018,24(8):994-999,1028.

Zhang YL, Wei X, Nie PY, et al. Establishment of risk assessment tool for postmenopausal osteoporotic fractures based on Group Lasso's logistic regression model[J]. Chinese Journal of Osteoporosis, 2018,24(8):994-999,1028.

[5] Taylor KC, Evans DS, Edwards DRV, et al. A genome-wide association study meta-analysis of clinical fracture in 10,012 African American women[J]. Bone Reports, 2016, 5: 233-242.

[6] Kilic N, Hosgormez E. Automatic estimation of osteoporotic fracture cases by using ensemble learning approaches[J]. Journal of Medical Systems, 2016, 40(3):61.

[7] Friedman JH. Stochastic gradient boosting[J]. Computational Statistics & Data Analysis, 2002, 38(4):367-378.

[8] Guo Y, Wang JT, Liu H, et al. Are bone mineral density loci associated with hip osteoporotic fractures? A validation study on previously reported genome-wide association loci in a Chinese population [J]. Genetics & Molecular Research, 2012, 11(1): 202-210.

[9] Guo Y, Yang TL, Dong SS, et al. Genetic analysis identifies DDR2 as a novel gene affecting bone mineral density and osteoporotic fractures in Chinese population [J]. PLoS One, 2015, 10(2): e0117102.

[10] Richards JB, Zheng HF, Spector TD. Genetics of osteoporosis from genome-wide association studies: advances and challenges[J]. Nature Reviews Genetics, 2012, 13(8): 576-588.

[11] 周志华. 机器学习[M]. 北京：清华大学出版社, 2016: 246-267.

Zhou ZH. Machine learning[M]. Beijing: Tsinghua University Press, 2016: 246-267.

[12] van der Maaten L, Hinton G. Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008, 9(Nov):2579-2605.

[13] Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: machine learning in Python[J]. Journal of Machine Learning Research, 2011, 12(10): 2825-2830.

[14] 周文怡,顾徐波,施勇,等.基于机器学习的网页暗链检测方法[J].计算机工程, 2018, 44(10):22-27.

Zhou WY, Gu XB, Shi Y, et al. Detection method for hidden hyperlink based on machine learning[J].Computer Engineering, 2018, 44(10): 22-27.

[15] Quinlan JR. Induction on decision tree[J]. Machine Learning, 1986, 1(1): 81-106.

[16] Breiman L. Random forests [J]. Machine Learning, 2001, 45(1): 5-32.

[17] Zhou ZH, Feng J. Deep Forest: Towards An Alternative to Deep Neural Networks [J/OL]. arXiv:1702.08835v3 [cs.LG] (2018-05-14) [2018-11-11]. https://arxiv.org/abs/1702.08835v3.

[18] Matthews BW. Comparison of the predicted and observed secondary structure of T4 phage lysozyme [J]. BBA - Protein Structure, 1975, 405(2): 442-451.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录