摘要
人工神经网络是一门新兴的边缘学科,与传统医学统计方法相比,人工神经网络不需要精确的数学模型,没有任何对变量的假设(如正态性、独立性等)要求,因此可以弥补传统统计方法的不足,解决一些用传统统计方法不能解决的问题。
人工神经网络的应用已经逐渐进入医学领域。对医学工作者而言,如何构建网络模型,使之具有较好的推广能力,解决实际问题是他们所关心的重点。本文针对在医学领域应用最广的BP神经网络展开研究,介绍了BP网络的基本原理,并从统计学的角度对学习过程进行描述。结合数据模拟通过对BP网络的构建、训练、优化和评价进行讨论,提出一些简易可行的网络优化方法,如如何确定隐单元数、通过修剪对网络进行简化以提高网络性能、提前终止训练等,并探讨它们的统计学应用和联系。
我们通过构造无隐含层的单层BP网络,对网络构建的一般原则、训练过程、评价指标进行总结,这些原则对多层BP网络也同样适用,对实际工作者具有指导意义。由于单层BP网络在医学危险因素筛选方面具有特殊的意义,我们对单层网络训练的样本大小要求进行模拟研究,而在此以前,国外只针对多层网络的训练样本提出过要求。模拟结果显示,训练样本大小与连接权数的比为10:1时,即可获得具有一定推广能力的网络模型。
我们对多层BP网络中隐单元的功能进行描述,并引入信息论的观点,提出一种应用信息熵估计隐单元数的方法。通过数据模拟与现有几种隐单元数确定方法的效果进行比较,熵法构造的模型的判别效果优于其它隐单元数模型。
探讨修剪算法对网络结构优化的作用以及它的实际意义。我们将修剪算法引入医学多因素资料的分析;首次探讨了单层网络中的修剪算法在变量筛选方面的作用,提出修剪后的单层BP网络模型的连接权与回归系数具有相同的含义,为
医学研究中危险因素的筛选提供了一个新的途径。
将前面提出的网络构建和优化的原理,包括隐单元数目的估计,修剪算法、
提前终止训练等进一步应用于预后资料的实例分析中,并与传统医学统计方法
比较。
本研究创新之处在于:
l)将信息论中的嫡引入网络模型构建,提出一种新的隐单元计算方法:根
据信息嫡估计隐单元数,在数据模拟和实际应用中取得较好的效果。
2)首次将网络修剪算法应用到医学多元资料的分析中,并对修剪算法与传
统统计学的联系进行探讨,实现了在网络训练中进行变量筛选。
3)通过模拟,对单层神经网络训练所需的样本含量进行初步探讨,提出最
低样本含量的要求。
由于时间和搜集的资料的限制,实例应用中对生存资料的分析仅限于单一
时间点模型,未对其他模型进行深入探讨,对删失数据在神经网络中的处理也
没有进一步研究。此外,对信息嫡估计隐单元数的方法还要进行更大规模的模
拟,以进一步证实。
Artificial neural network (ANN) is a rising borderline science. Compared to the mathematical statistics, it doesn't need exact mathematical model and dose not have any consumption (such as distribution and independence) demanding the variables to meet. It can make up the deficiency of mathematical statistical methods and solve some problems that traditional statistical methods failed to resolve.
In medical study fields, the application of ANN is more and more popular. Particularly attention had been given to how to build a suitable ANN model to resolve the actual problems. This study focuses on the back propagation network (BP network) which is the most popular model used in medical filed. The fundamental of BP network was introduced and the description from the statistical point of view was given. Several simulative BP network architectures had been set up to discuss the designing, learning, optimizing and evaluation of the BP networks. The details of some simple optimizing methods, such as how to design the number of hidden units, prune the network to improve its generalization, and stop learning properly were studied. The performance and relation to statistics of these methods had also been discussed.
There were three parts of the main study.
The BP model without hidden layer (also can be taken as single layer model) was constructed and its principles of architecture designing and learning process and the methods of evaluation were summarized, which also can be used in multi-layer BP model and can be used as a conduct in application. Residual analysis and AUC of
ROC were introduced as methods of evaluation. For the particular meaning of single layer BP model using in risk factor screening in medicine, a simulating study to derive the approximation bounds of its learning sample was conducted. Only the multi-layer BP model' bound had been studied before by other scholars. The outcome of simulation indicated that a stable model with proper ability of generalization had been derived when the relationship between sample size and links number is 10:1.
For multi-layer BP model, we focused on the study of hidden layer. Description of the function of hidden layer was given. Information theory was inducted into the study of the number of hidden units. A novel method using entropy to estimate the number of hidden units was proposed. The simulation outcome showed that the effect of BP model with the number of hidden units generated by entropy performed better than other model.
To simplifying the structure, the effect of pruning algorithms application in BP network model and its relationship with medical statistics were studied. In our study, one of the pruning algorithms - optimal brain damage method was applied in multi-variable data analysis. It is the first time that the application in variable selection with this method was discussed. We found that pruning in single layer BP network model can act as variable selection, and the weight of BP network model after pruning had the same meaning as regression coefficient. Thus a new method for the hazard factor screening in medical study has been proposed.
Those principles of BP network designing and optimization were also applied on actual medical prognosis data and survival data. The result of that was also compared with the performance of logistic regression to test and verify.
The innovation of this study lied in the following:
Inducting the entropy in information theory into the BP network architecture constructing and bring a novel method of hidden units number designing using entropy.
The pruning algorithm being used in medical data as a method of variable selection for the first time.
The approximation bounds of single layer BP model being studied by
simulated data.
For the limit of time and data collection, the study on actual survival data was only conducted by one architecture model. The further study should be conducted on the analysis of other models and the censored data treatment should be considered. The entropy method of
引文
1 Wu C, Shivakumar S. back-propagation and counter-propagation neural networksd for phylogenetic classification of ribosomal RNA sequences. Nucleic Acids Res, 1994, 22(20):4291-9
2 Alvager T, Graham G, Hutchison D. Neural network method to analyze data compression in DNA and RNA sequences. J Chem Inf Comput Sci, 1997, 37(2):335-7
3 耿新,陈光等.人工神经网络解析紫外光度法同时测定复方制剂.理化检验:化学分,2001,37(3):99-101
4 李睿,相秉仁.人工神经网络在药动学/药效学研究中的应用.药学进展,2000,24(2):97-100
5 耿利娜,李静.神经网络法在使用裂解气相色谱鉴别中草药中的应用.分析化学,2000,28(5):549-553
6 高守国,屈凌波.人工神经网络用于特征提取及其在牛黄鉴别中的应用.计算机与应用化学,2000,17(1):105-106
7 Takayama K, Fujikawa M, Nagai T. Artificial neural network as a novel method to optimoze pharmaceutical formulations. Pharm Res, 1999,Jan; 16(1): 1-6
8 袁景淇,沈云岐.生物制药生产中的在线产量预测及经济效益优化.中国抗生素杂志,1999,24(3):161-168
9 黄德生,周宝森.BP人工神经网络用于肺鳞癌预后预测.中国卫生统计,2000,17(6):337-340
10 宋红,林家瑞.用于医学辅助诊断的神经网络方法的应用研究.生物医学工程学杂志,1996,13(2):141-144
11 孙高,杨静宇,基于人工神经网络的肺癌诊断研究中国卫生统计.1999,16(3):142-144
12 邓小元,李坤成.人工神经网络在Alzheimer病的MRI诊断研究中的应用.中华放射学杂志,1998,32(12):-812-816
13 David Faraggi, Richard Simon. A neural network model for survival data.Statistic in Medicine, 1995, 14:73-82
14 Knut Liestol,Per Kragh, Andersen, Ulrich Andersen. Survival analysis and neural net. Statistics in Medicine, 1994,13:1189-1200
15 王衍文,王海滨.心音信号的识别与分类生物医学工程学杂志.1999,16(3):
382-387
16 乔友林,侯俊等.我国太行山高发区食管癌流行趋势及防治策略.中国医学科学院学报,2001,23(1):10-14
17 蔡煜东,朱海鸾.流行性脑脊髓膜炎预报的人工神经网络模型.生物医学工程学杂志,1994,11(4):314-318
18 Janet Fricker. Artificial neural networks improve diagnosis of acute myocardial infarction. The Lancet, 1997, 350 (9082): 935
19 Ali A El-Solh, Chiu-Bin Hsiao, Susan Goodnough. Predicting active pulmonary tuberculosis using an artificial neural network. Chest, 1999, 116 (4): 968
20 Joannis G Vlachonikolis, Dimitris A Karras, Manolis J Hatzakis. Improved statistical classification methods in computerized psychiatric diagnosis. Medical Decision Making, 2000, Jan-Mar. (201): 95
21 Frize M, Frasson C. An artificial neural network that uses eye-tracking performance to identify patients with Schizophrenia. Clinical & Investigative Medicine, 2000 Aug, (234): 266-9
22 Jack V Tu, Milton C Weinstein, Barbara J McNeil. Predicting mortality after coronary artery bypass surgery: What do artificial neural networks learn? Medical Decision Making. 1998 Apr-Jun, 18 (2):229-235
23 Toddy Rowland Lucila Ohno Machado Aleksander Ohrn Comparison of multiple prediction model for ambulation following spinal cord injury. Proc AMIA Symp,1998
24 周利锋.人工神经网络在医学多元统计中的应用研究.上海医科大学研究生论文,1998.5
25 贺宪民.医学多元资料分析中的人工神经网络方法研究.第二军医大学研究生论文,2001,4
26 张乃尧,阎平凡.神经网络与模糊控制.北京:清华大学出版社,1998:5-6
27 袁曾任.人工神经网络及其应用.北京:清华大学出版社,1999.10
28 Martin Schumacher, Werner Vach, Reinhard Robnr. Neural networks and logostoc regression:Part Ⅰ. Comp Stat Data Analy, 1996,21:661-682
29 王科俊,王克成.神经网络建模、预报与控制.哈尔滨:哈尔滨工程大学出版社,1996.12
30 张立明.人工神经网络的模型及其应用.复旦大学出版社.1993.7
31 S. Ergezinger, E. Thomsen. An accelerated learning algorithm for multiplayer
perceptrons: optimization layer by layer. IEEE Trans NN 1995, 6(1):31
32 罗四维.人工神经网络建造.1998,5:96
33 孟建,屈梁生.基于信息优化和前馈神经网络及其应用.中国机械工程,1997,8(2):54-57
34 王宗炎,洪振华.BP网络学习算法的改进及其在模式识别中的应用.南京航空航天大学学报,1994,26 Suppl.:216-222
35 Martin Schumacher, Werner Vach, Reinhard Robnr. Neural networks and logostoc regression:Part Ⅱ. Comp Stat Data Analy, 1996,21:663-701
36 张小蒂.应用回归分析.浙江:浙江大学出版社,1991.8
37 盛骤,谢式千,潘承武.概率论与数理统计.北京:高等教育出版社,1979.3
38 E. B. Baum, D. Haussler. "What size net gives valid generalization?". Neural Computation. 1988,1:151-160
39 Kishan G. Mehrotra, Chilukuri K, Mohan, Sanjay Ranka. Bounds on the nember of sample needs for neural learning. IEEE Trans. NN, 1991, 2(6):548-558
40 周利锋,高尔生,金丕焕.BP神经网络与logistic回归对比初探.中国卫生统计,1998,15(1):1-4.
41 Maura E. Stokes, Charles S. Davis, Gary G.Koch. Categorical data analysis using the SAS system. SAS Institute Inc., USA, 1995
42 Peter Armitage, Theodore Colton. ENCYCLOPEDIA of BIOSTATISTICS (Volume Ⅰ). John Wiley &Sons Ltd, England, 1998.
43 David B.Fogel. An information criterion for optimal neural network selection. IEEE Trans NN. 1991,2(5):490-497
44 Julian Faraway, Chris Chatfieldt. Time series forcasting with neural networks: a comparative study using the airline data. Appl. Statist.,199847(2): 231-250
45 曹素华.实用医学多因素统计方法.上海:上海医科大学出版社,1997.11:190
46 Douglas M.Bates,Donald G.Watts著,韦博成等译.非线性回归分析及其应用.北京:中国统计出版社,1997.11
47 Jack V. Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes. Clin. Epidemiol. 1996, 49(11):1225-1231
48 Melanie T.Young, Susan M. Blanchard, Mark W. White, et. al. Using artificial neural network to detect activations during ventiricular fibrillation. Computers and Biomedical Research 2000,33:43-58
49 Peter Armitage, Theodore Colton. ENCYCLOPEDIA of BIOSTATISTICS (Volume V). John Wiley &Sons Ltd, England, 1998.
50 王家良.临床流行病学.中华医学杂志,1998,78(12):941-943
51 李康,林一帆.评价判别模型诊断效果的ROC分析.中国卫生统计,1996,13(3):9-12
52 Birol Emir, Sam Wieand, Sin-Ho Jung, et al. Comparison of diagnostic markers with repeated measurement a non-parametric ROC curve approach. Statist. Med., 2000, 19:511-523
53 宇传华,徐勇勇.非参数法估计ROC曲线下面积.中国卫生统计,1999,16(4):241-244
54 R. P. Lippmann. An Introduction to Computing with Neural Nets. IEEE ASSP Magazine, 1987 April: 4-22
55 Hornik K. Approximation capabilities of multiplayer feedward network. Neural Networks, 1994,4(2): 251-257
56 Hornik K. Some new results on neural network approximation. Neural Networks,1993,6:1069-1072
57 White H. Learning in artificial neural networks: A statistical perspective. Neural Comput, 1989,1: 425-464
58 王伟.人工神经网络原理:入门与应用.北京:北京航空航天大学出版社,1995.
59 Tianping Chen, Hong Chen, Ruet-wen Liu. Approximation capability in C(Rn) by multilayer feedforward networks and related problems. IEEE Trans NN,1995,6(1):25
60 Shih-Chi Huang, Yih-Fang Huang. Bounds on the number of hidden neurons in multiplayer perceptrons. IEEE Trans. NN, 1991,2(1):47-55
61 鲁晨光.广义信息论.安徽合肥:中国科技大学出版社,1983.10
62 陈力行.概率与信息.济南:山东科学技术出版社,1984.11
63 钟义信.信息科学原理.北京:北京邮电大学出版社,1996.2
64 [日]有本卓 著,杨逢春译.近代信息论.北京:人民邮电出版社,1985.11
65 王雨田.控制论、信息论、系统科学与哲学.北京:中国人民大学出版社,1986
66 姜丹.信息论.安徽合肥:中国科技大学出版社,1987.9
67 侯炳辉,来珠,曹慈惠,等.计算机原理与系统结构.北京:清华大学出版
社,1992
68 Reed R. Pruning algorithms a survey. IEEE Trans.NN,1993,4:740-747
69 Randal A.Koene, Yoshio Takane. Discrimnant component pruning: regularization and interpretation of multilayered Backpropagation networks. Neural Computation, 1999, 11:783-802
70 徐秉铮,张百灵,韦岗.华南理工大学科学丛书-神经网络理论与应用.广州:华南理工大学出版社,1994.12
71 J.Gorodkin, L.K.Hansen, B.Lautrup, et.al. Universal distribution of saliencies for pruning in layed neural networks. Internation J. Neural Sys, 1997, 8(5):489-498
72 Andreas Zell,Gunter mamier, Michael Vogt, et al. Stuttgart Neural Networks Simulator(SNNA), Use Manual, Version 4.1.University of Suttgart, 1995,192-198.
73 刘勤.COX模型拟合过程中存在的主要问题及其解决方法的研究和应用.上海医科大学研究生论文,1998