模糊神经网络在肺癌诊断中的应用
|
摘要
肺癌是当今世界各国常见的恶性肿瘤,已成为大多数国家癌症死亡的主要原因。在我国,肺癌的发病率已由原来的第六位升至第一位,病死率在城市人口恶性肿瘤中居首位。在农村,升幅明显大于城市,尤以农村女性的增长趋势较为突出。由于肺癌早期无症状或症状轻微或与其它疾病症状类似,不易发现,待到发现时已有转移。所以,早期诊断具有重要的价值,对治疗计划有很重要的影响。因此,提高早期肺癌的检出率是提高患者生存率和降低死亡率的必要手段。
本文回顾了肺癌诊断的历史现状,在此基础上,介绍并讨论了人工神经网络和模糊理论的结合——模糊神经网络。由于人工神经网络是基于人脑的计算机模型,具有良好的自适应性、自组织和很强的自学习能力,是数据分类和模式识别的有力工具。而模糊理论能更直接更自然地表达人类习惯用的逻辑含义,适用于直接的或高层的知识表达。将二者结合用于早期肺癌诊断,为肺癌诊断开辟一条新途径。本课题以从胸部CT片中提取的21项放射学特征和5个临床参数为基础,旨在提高早期肺癌的检出率,使诊断结果更加准确。
方法：选用隶属度函数为高斯型的模糊神经网络用于肺癌诊断,将26个特征参数中的13个非二值变量进行模糊化处理,每个参数分为3个模糊子空间,用大(L)、中(M)、小(S)3个语言变量表示,每一个输入变量就有3个模糊化神经元与其在3个子空间对应的隶属度函数对应,然后和其它13个二值参数一起作为BP神经网络的输入。将所获得的117例病例样本随机分为训练集和测试集,训练模糊神经网络,选择合适的隐节点数。用测试集测试该网络区分肺癌和非肺癌的能力,并将结果与三角形隶属度函数模糊神经网络的测试结果进行比较。
结果：对于早期肺癌的预测,高斯型隶属度函数模糊神经网络的虚警和漏检率较低,比作为对照的三角形隶属度函数模糊神经网络诊断正确率有所提高。高斯型隶属度函数模糊神经网络4例错误(良性5、6例,肺癌36、38例),而三角形隶属度函数模糊神经网络有5例错误,除上述4例外,肺癌中又增加第28例。高斯型隶属度函数模糊神经网络的总诊断正确率为91%,比三角形隶属度函数模糊神经网络高出3个百分点,而且对病例样本分组变化不敏感。因此,高斯型隶属度函数模糊神经网络更适用于肺癌诊断。
Lung cancer is a common malignant tumor in the world today, which has become the main reason of cancer patients' death. In China, the incidence of lung cancer has risen from the sixth to the first, and the mortality is in the top of urban population malignant tumor. In rural areas, the increase is significantly larger than that in urban, especially the trend of the rural women's mortality rise is more prominent. Because there are no or few specific symptoms in the early period of lung cancer, it is difficult to be detected. It has usually metastasized when it is detected. Early diagnosis has an important prognostic value and has a huge impact on treatment planning. So the early diagnosis and treatment is a necessary method to improve the survival rate and reduce the mortality of the patients with lung cancer.
The recent progress of lung cancer diagnosis was reviewed in the paper. Fuzzy neural network was introduced and discussed. Fuzzy neural network was the combination of artificial neural network and fuzzy theory. Artificial neural network is a computational model based on the brain; it is a powerful tool for data classification and pattern identification because it has good adaptability, self-organization and self-learning ability. The prominent feature of fuzzy logic is to describe logical meaning to more naturally and directly that human are accustomed, which is suitable to express direct or high level of knowledge. And fuzzy neural network was used to diagnose lung cancer, developing a new approach for lung cancer diagnosis. The network is based on 21 features extracted from chest CT and 5 clinical parameters. The aim of the study was to improve the lung cancer diagnosis accuracy.
Methods:The usefulness of a fuzzy neural network with Gaussian membership function for distinguishing between lung cancer and benign cases was studied to improve lung cancer diagnosis.13 non-binary parameters of 26 characteristic parameters were fuzzed with Gaussian membership function. Every input variable was divided into 3 fuzzy subspaces, using large(L), medium(M) and small(S) 3 linguistic variables to express. Each input variable had 3 fuzzy neurons, corresponding to subject function values in 3 fuzzy subspaces. Output after fuzziness was 39 elements. Then, the fuzzed outputs added with the other 13 binary parameters were used as inputs of the BP neural network.117 cases, including lung cancer and benign cases, were divided into training set and test set randomly. And these cases were used to train fuzzy neural network and select appropriate hidden nodes. The test set was also used to test the performance of the trained fuzzy neural network in differentiation of benign from malignant pulmonary nodules. The performances of Gaussian membership function fuzzy neural network were compared with that of fuzzy neural network with triangle membership function.
Conclusions:The performance of the fuzzy neural network with Gaussian membership function was better than that of the fuzzy neural network with triangle membership function in prediction of probability of malignancy in pulmonary nodules. There were 2 false-positive and 2 false-negative in Gaussian membership function fuzzy neural network. But in triangle membership function fuzzy neural network, there was 1 false-negative more than that of in GMF FNN. The diagnostic accuracy rate of fuzzy neural network with Gaussian membership function was 91%, which was 3 percentage points higher than that of fuzzy neural network with triangular membership function. And the sensitivity of Gaussian membership function fuzzy neural network was better than that of triangle membership function fuzzy neural network. So, the fuzzy neural network with Gaussian membership function has the potential to improve the diagnostic accuracy of distinction between the benign and malignant cases.
本文回顾了肺癌诊断的历史现状,在此基础上,介绍并讨论了人工神经网络和模糊理论的结合——模糊神经网络。由于人工神经网络是基于人脑的计算机模型,具有良好的自适应性、自组织和很强的自学习能力,是数据分类和模式识别的有力工具。而模糊理论能更直接更自然地表达人类习惯用的逻辑含义,适用于直接的或高层的知识表达。将二者结合用于早期肺癌诊断,为肺癌诊断开辟一条新途径。本课题以从胸部CT片中提取的21项放射学特征和5个临床参数为基础,旨在提高早期肺癌的检出率,使诊断结果更加准确。
方法：选用隶属度函数为高斯型的模糊神经网络用于肺癌诊断,将26个特征参数中的13个非二值变量进行模糊化处理,每个参数分为3个模糊子空间,用大(L)、中(M)、小(S)3个语言变量表示,每一个输入变量就有3个模糊化神经元与其在3个子空间对应的隶属度函数对应,然后和其它13个二值参数一起作为BP神经网络的输入。将所获得的117例病例样本随机分为训练集和测试集,训练模糊神经网络,选择合适的隐节点数。用测试集测试该网络区分肺癌和非肺癌的能力,并将结果与三角形隶属度函数模糊神经网络的测试结果进行比较。
结果：对于早期肺癌的预测,高斯型隶属度函数模糊神经网络的虚警和漏检率较低,比作为对照的三角形隶属度函数模糊神经网络诊断正确率有所提高。高斯型隶属度函数模糊神经网络4例错误(良性5、6例,肺癌36、38例),而三角形隶属度函数模糊神经网络有5例错误,除上述4例外,肺癌中又增加第28例。高斯型隶属度函数模糊神经网络的总诊断正确率为91%,比三角形隶属度函数模糊神经网络高出3个百分点,而且对病例样本分组变化不敏感。因此,高斯型隶属度函数模糊神经网络更适用于肺癌诊断。
Lung cancer is a common malignant tumor in the world today, which has become the main reason of cancer patients' death. In China, the incidence of lung cancer has risen from the sixth to the first, and the mortality is in the top of urban population malignant tumor. In rural areas, the increase is significantly larger than that in urban, especially the trend of the rural women's mortality rise is more prominent. Because there are no or few specific symptoms in the early period of lung cancer, it is difficult to be detected. It has usually metastasized when it is detected. Early diagnosis has an important prognostic value and has a huge impact on treatment planning. So the early diagnosis and treatment is a necessary method to improve the survival rate and reduce the mortality of the patients with lung cancer.
The recent progress of lung cancer diagnosis was reviewed in the paper. Fuzzy neural network was introduced and discussed. Fuzzy neural network was the combination of artificial neural network and fuzzy theory. Artificial neural network is a computational model based on the brain; it is a powerful tool for data classification and pattern identification because it has good adaptability, self-organization and self-learning ability. The prominent feature of fuzzy logic is to describe logical meaning to more naturally and directly that human are accustomed, which is suitable to express direct or high level of knowledge. And fuzzy neural network was used to diagnose lung cancer, developing a new approach for lung cancer diagnosis. The network is based on 21 features extracted from chest CT and 5 clinical parameters. The aim of the study was to improve the lung cancer diagnosis accuracy.
Methods:The usefulness of a fuzzy neural network with Gaussian membership function for distinguishing between lung cancer and benign cases was studied to improve lung cancer diagnosis.13 non-binary parameters of 26 characteristic parameters were fuzzed with Gaussian membership function. Every input variable was divided into 3 fuzzy subspaces, using large(L), medium(M) and small(S) 3 linguistic variables to express. Each input variable had 3 fuzzy neurons, corresponding to subject function values in 3 fuzzy subspaces. Output after fuzziness was 39 elements. Then, the fuzzed outputs added with the other 13 binary parameters were used as inputs of the BP neural network.117 cases, including lung cancer and benign cases, were divided into training set and test set randomly. And these cases were used to train fuzzy neural network and select appropriate hidden nodes. The test set was also used to test the performance of the trained fuzzy neural network in differentiation of benign from malignant pulmonary nodules. The performances of Gaussian membership function fuzzy neural network were compared with that of fuzzy neural network with triangle membership function.
Conclusions:The performance of the fuzzy neural network with Gaussian membership function was better than that of the fuzzy neural network with triangle membership function in prediction of probability of malignancy in pulmonary nodules. There were 2 false-positive and 2 false-negative in Gaussian membership function fuzzy neural network. But in triangle membership function fuzzy neural network, there was 1 false-negative more than that of in GMF FNN. The diagnostic accuracy rate of fuzzy neural network with Gaussian membership function was 91%, which was 3 percentage points higher than that of fuzzy neural network with triangular membership function. And the sensitivity of Gaussian membership function fuzzy neural network was better than that of triangle membership function fuzzy neural network. So, the fuzzy neural network with Gaussian membership function has the potential to improve the diagnostic accuracy of distinction between the benign and malignant cases.
引文
[1]Qiang Li. Recent Progress in Computer-Aided Diagnosis of Lung Cancer on Thin-Section CT. Comput Med Imaging Graph,2007; 31(4-5):248-257.
[2]Kwang Gi Kim, Jin Mo Goo, Jong Hyo Kim, et al. omputed-aided Diagnosis of Localized Ground-Glass Opacity in the Lung at CT:Initial Experience1. Radiology,2005,237:657-661
[3]Katsumi Nakamura, Hiroyuki Yoshida, Roger Engelmann, et al. Computerized analysis of the likelihood of malignancy in solitary pulmonary nodules with use of artificial neural networks. Radiology 2000,214:823-830
[4]Asada N, Doi K, MacMahon H, et al. Potential usefulness of an artificial neural network for differential diagnosis of interstitial lung disease:pilot study. Radiology 1990,177:857-860
[5]Hubert Vesselle, Eric Turcotte, Linda Wiens, David Haynor. Application of a Neural Network to Improve Nodal Staging Accuracy with 18F-FDG PET in Non-Small Cell Lung Cancer. The JOURNAL OF NUCLEAR MEDICINE,2003,44(12):1918-1926
[6]万雪峰,徐力平,吴拥军,吴逸明.模糊神经网络在肺癌诊断中的应用[J].微计算机信息,2009,25(4)：290-291
[7]董志伟,乔友林,李连弟等.中国癌症控制策略研究报告.中国肿瘤,2002,11(5)：250-260
[8]全国肿瘤防治办公室,卫生部卫生统计信息中心,中国试点市、县恶性肿瘤的发病与死亡第二卷(1993-1997)[M].北京：中国医药科技出版社,2002
[9]杨玲,李连弟,陈育德.中国肺癌死亡趋势分析及发病死亡的估计与预测.中国肺癌杂志,2005.8(4)：1-5
[10]Sone S, Takashima S, Li F, et al. Mass screening for lung cancer with mobile spiral computed tomography scanner[J]. Lancet,1998,351(9111):1242-1245
[11]Diederich S, Semik M, Lentscig MG et al. Helical CT of pulmonary nodules in patients with extrathoracic malignancy:CT surgical correction. Lancet,1999,345(9173):99-105
[12]Holiday DB, Mclarty JW, Farley ML, et al. Sputum cytology within and across laboratories. A reliability study [J]. Acta Cytol,1995,39(2):195-206
[13]Vermylen P, Pierard P, Roufosse C, et al. Detection of bronchial preneoplastic lesions and early lung cancer with fluoresence bronchoscopy:a study about its ambulatory feasibility under local anaesthesia[J]. Lung Cancer,1999,25(3):161-168.
[14]Sueoka E, Goto Y, et al. Heterogenous nuclear ribonucleoprotein B1 as a new marker of early detection for human lung cancers [J]. Cancer Res,1999,59(7):1404-1407
[15]Satoru Isaka. On Neural Approximation of Fuzzy System [A].In:Proceeding of INCNN' 92[C].New York:IEEE,1992,1263-1268.
[16]Jokinen Petri A. On the Relations Between Radial Basis Function Networks and Fuzzy System [A]. In:Proceeding of INCNN'92[C]. New York:IEEE,1992,1220-1225
[17]Wang zhenni, Tham Ming T, Morris A. Multilayer Feedforward Neural Networks:A Canonical form Approximation of Nonlinearity, International Journal of Control,1992,56(3): 655-672
[18]Werbos P J. Back-propagation:Past and Future[C]. ICNN,1988:343-353
[19]R. Hechi-Nielsen. Theory of the Back-Propagation Neural Network[C], Pro. of IJCNN,1989: 593-603
[20]Rumelhart D E,Hinton G E,Willians R J.Learning Representations by Back-PropagationErrors.Nature.1986,323：533-536
[21]McCulloch W S and Pitts W H. A Logical Calculus of the Ideas Immanent in Nervous Activity. Bulletin of Mathematical Biophysics,1943,5(1):115-133
[22]Ito Y. Representation of function by superposition of step or sigmoidal function and their application to neural network theory. Neural Network,1991(4):385-394
[23]Cybenko G. Approximation by superposition of a sigmoidal function. Mathematic of Control, Signals, Systems,1989,2(4):303-314
[24]韩力群,人工神经网络教程(第一版).北京：北京邮电大学出版社,2006[25] Partridge D. Network generalization differences quantified. Neural Networks,1996,9(2): 263-271
[26]张鸿彬训练多层网络的样本数问题.自动化学报,1993,19(1)：71-77
[27]Niyogo P and Girosi F. On the Relationship between Generalization Error, Hypothesis Complexity and Sample Complexity for Radial Basis Function. Neural Computation,1996(8): 819-842
[28]Kolen J F and Pollack B J. Back Propagation is Sensitive to Initial Condition. NIPS 3,1991: 860-867
[29]Schmidt W F, Raudys S, Kraaijveld M A, Skurikhina M and Duin R P W. Initialization, Back-Propagation, and Generalization of Feed-Forward Classifier. In Proc. IEEE Int. Conf. Neural Networks,1993:598-604
[30]Atiya A and Ji C. How Initial Condition Affect Generalization Performance in Large
Networks. IEEE Trans. Neural Networks,1997(8):448-451
[31]Polycarpou M and Ioannou P. Learning and Convergence Analysis of Neural-type Structured
Networks. IEEE Trans. Neural Networks 1992(3):39-50
[32]Baum E and Haussler D. What Size Net Gives Valid Generalization? NIPS 1,1989:81-90
[33]Bartlett P L. For Valid Generalization, the Size of the Weight Is More Important Than the Size of the Network. NIPS 9,1997:134-140
[34]Sollich P and Krogh A. Learning with ensembles:How over-fitting Can be Useful. NIPS 8, Cambridge,1996:190-196
[35]K. Funahashi. On the approximate realization of continuous mappings by neural network. Neural Networks, vol.2(3), pp.183-192,1989
[36]Jacques de Villiers and Etienne Barnard. Back-propagation Neural Nets with One and Two Hidden Layers. IEEE Transactions on Neural Network, vol.4, NO.1:136-141.Jaunary 1992
[37]Zadeh L A. Fuzzy Sets. Information and Control.1965,8:338-353
[38]Qingqing zhou, Martin Purvis and Nikola Kasabov, A Membership Function Selection Method for Fuzzy Neural Networks. International Conference on Neural Information Processing and Intelligence Information System,1997, vol,2(1):1-5
[39]曾光奇,胡均安,王东等。模糊控制理论与工程应用[M].武汉：华中科技大学出版社,2006,14-15,23,56-57
[40]赵宇,基于变隶属函数的倒立摆模糊控制算法的研究.电子科技大学硕士学位论文,2007
[41]Horikawa.A, Furuhashi.T and Uchikawa.Y,(1992) On Fuzzy Modeling Using Fuzzy Neural Network with the Back-Propagation Algorithm. IEEE Transactions On Neural Networks, Vol.3, No.5,801-806
[42]Horikawa.S,Furuhashi.T,Okuma.S and Uchikawa.Y, Composition Methods Of Fuzzy Neural Networks.16th Annual Conference of IEEE, vol.2,1253-1258,1990
[43]Dunyak, Wunsch Jamens, Donald. Fuzzy regression by fuzzy number neural network[J]. Fuzzy Sets and Systems,2000,112(3):371-380
[44]伍昌鸿,马晓茜,廖艳芬.基于模糊神经网络的电站燃烧锅炉结渣预测[J].燃烧科学与技术,2006,12(2)：175-179
[45]丛爽.几种模糊神经网络关系的对比研究[J].信息与控制,2001,30(6)：486-492
[46]魏海坤.神经网络结构设计的理论与方法.北京：国防工业出版社,2005.2
[47]诸静,等,著.模糊控制原理与应用.北京：机械工业出版社,1999：1-1 出生年月日：1983、10、9获得学士学位的学校、时间：解放军信息工程大学、2007年7月获得硕士学位的学校、时间：郑州大学、2010年6月攻读学位期间发表论文：
[1]徐力平,张华杰.平台罗经故障检测的BP神经网络.郑州大学(理学版).2009,41(3)：30-33
[2]徐力平,张华杰,吴逸明.高斯隶属度函数模糊神经网络在肺癌中的应用.郑州大学(理学版)(已录用)
[2]Kwang Gi Kim, Jin Mo Goo, Jong Hyo Kim, et al. omputed-aided Diagnosis of Localized Ground-Glass Opacity in the Lung at CT:Initial Experience1. Radiology,2005,237:657-661
[3]Katsumi Nakamura, Hiroyuki Yoshida, Roger Engelmann, et al. Computerized analysis of the likelihood of malignancy in solitary pulmonary nodules with use of artificial neural networks. Radiology 2000,214:823-830
[4]Asada N, Doi K, MacMahon H, et al. Potential usefulness of an artificial neural network for differential diagnosis of interstitial lung disease:pilot study. Radiology 1990,177:857-860
[5]Hubert Vesselle, Eric Turcotte, Linda Wiens, David Haynor. Application of a Neural Network to Improve Nodal Staging Accuracy with 18F-FDG PET in Non-Small Cell Lung Cancer. The JOURNAL OF NUCLEAR MEDICINE,2003,44(12):1918-1926
[6]万雪峰,徐力平,吴拥军,吴逸明.模糊神经网络在肺癌诊断中的应用[J].微计算机信息,2009,25(4)：290-291
[7]董志伟,乔友林,李连弟等.中国癌症控制策略研究报告.中国肿瘤,2002,11(5)：250-260
[8]全国肿瘤防治办公室,卫生部卫生统计信息中心,中国试点市、县恶性肿瘤的发病与死亡第二卷(1993-1997)[M].北京：中国医药科技出版社,2002
[9]杨玲,李连弟,陈育德.中国肺癌死亡趋势分析及发病死亡的估计与预测.中国肺癌杂志,2005.8(4)：1-5
[10]Sone S, Takashima S, Li F, et al. Mass screening for lung cancer with mobile spiral computed tomography scanner[J]. Lancet,1998,351(9111):1242-1245
[11]Diederich S, Semik M, Lentscig MG et al. Helical CT of pulmonary nodules in patients with extrathoracic malignancy:CT surgical correction. Lancet,1999,345(9173):99-105
[12]Holiday DB, Mclarty JW, Farley ML, et al. Sputum cytology within and across laboratories. A reliability study [J]. Acta Cytol,1995,39(2):195-206
[13]Vermylen P, Pierard P, Roufosse C, et al. Detection of bronchial preneoplastic lesions and early lung cancer with fluoresence bronchoscopy:a study about its ambulatory feasibility under local anaesthesia[J]. Lung Cancer,1999,25(3):161-168.
[14]Sueoka E, Goto Y, et al. Heterogenous nuclear ribonucleoprotein B1 as a new marker of early detection for human lung cancers [J]. Cancer Res,1999,59(7):1404-1407
[15]Satoru Isaka. On Neural Approximation of Fuzzy System [A].In:Proceeding of INCNN' 92[C].New York:IEEE,1992,1263-1268.
[16]Jokinen Petri A. On the Relations Between Radial Basis Function Networks and Fuzzy System [A]. In:Proceeding of INCNN'92[C]. New York:IEEE,1992,1220-1225
[17]Wang zhenni, Tham Ming T, Morris A. Multilayer Feedforward Neural Networks:A Canonical form Approximation of Nonlinearity, International Journal of Control,1992,56(3): 655-672
[18]Werbos P J. Back-propagation:Past and Future[C]. ICNN,1988:343-353
[19]R. Hechi-Nielsen. Theory of the Back-Propagation Neural Network[C], Pro. of IJCNN,1989: 593-603
[20]Rumelhart D E,Hinton G E,Willians R J.Learning Representations by Back-PropagationErrors.Nature.1986,323：533-536
[21]McCulloch W S and Pitts W H. A Logical Calculus of the Ideas Immanent in Nervous Activity. Bulletin of Mathematical Biophysics,1943,5(1):115-133
[22]Ito Y. Representation of function by superposition of step or sigmoidal function and their application to neural network theory. Neural Network,1991(4):385-394
[23]Cybenko G. Approximation by superposition of a sigmoidal function. Mathematic of Control, Signals, Systems,1989,2(4):303-314
[24]韩力群,人工神经网络教程(第一版).北京：北京邮电大学出版社,2006[25] Partridge D. Network generalization differences quantified. Neural Networks,1996,9(2): 263-271
[26]张鸿彬训练多层网络的样本数问题.自动化学报,1993,19(1)：71-77
[27]Niyogo P and Girosi F. On the Relationship between Generalization Error, Hypothesis Complexity and Sample Complexity for Radial Basis Function. Neural Computation,1996(8): 819-842
[28]Kolen J F and Pollack B J. Back Propagation is Sensitive to Initial Condition. NIPS 3,1991: 860-867
[29]Schmidt W F, Raudys S, Kraaijveld M A, Skurikhina M and Duin R P W. Initialization, Back-Propagation, and Generalization of Feed-Forward Classifier. In Proc. IEEE Int. Conf. Neural Networks,1993:598-604
[30]Atiya A and Ji C. How Initial Condition Affect Generalization Performance in Large
Networks. IEEE Trans. Neural Networks,1997(8):448-451
[31]Polycarpou M and Ioannou P. Learning and Convergence Analysis of Neural-type Structured
Networks. IEEE Trans. Neural Networks 1992(3):39-50
[32]Baum E and Haussler D. What Size Net Gives Valid Generalization? NIPS 1,1989:81-90
[33]Bartlett P L. For Valid Generalization, the Size of the Weight Is More Important Than the Size of the Network. NIPS 9,1997:134-140
[34]Sollich P and Krogh A. Learning with ensembles:How over-fitting Can be Useful. NIPS 8, Cambridge,1996:190-196
[35]K. Funahashi. On the approximate realization of continuous mappings by neural network. Neural Networks, vol.2(3), pp.183-192,1989
[36]Jacques de Villiers and Etienne Barnard. Back-propagation Neural Nets with One and Two Hidden Layers. IEEE Transactions on Neural Network, vol.4, NO.1:136-141.Jaunary 1992
[37]Zadeh L A. Fuzzy Sets. Information and Control.1965,8:338-353
[38]Qingqing zhou, Martin Purvis and Nikola Kasabov, A Membership Function Selection Method for Fuzzy Neural Networks. International Conference on Neural Information Processing and Intelligence Information System,1997, vol,2(1):1-5
[39]曾光奇,胡均安,王东等。模糊控制理论与工程应用[M].武汉：华中科技大学出版社,2006,14-15,23,56-57
[40]赵宇,基于变隶属函数的倒立摆模糊控制算法的研究.电子科技大学硕士学位论文,2007
[41]Horikawa.A, Furuhashi.T and Uchikawa.Y,(1992) On Fuzzy Modeling Using Fuzzy Neural Network with the Back-Propagation Algorithm. IEEE Transactions On Neural Networks, Vol.3, No.5,801-806
[42]Horikawa.S,Furuhashi.T,Okuma.S and Uchikawa.Y, Composition Methods Of Fuzzy Neural Networks.16th Annual Conference of IEEE, vol.2,1253-1258,1990
[43]Dunyak, Wunsch Jamens, Donald. Fuzzy regression by fuzzy number neural network[J]. Fuzzy Sets and Systems,2000,112(3):371-380
[44]伍昌鸿,马晓茜,廖艳芬.基于模糊神经网络的电站燃烧锅炉结渣预测[J].燃烧科学与技术,2006,12(2)：175-179
[45]丛爽.几种模糊神经网络关系的对比研究[J].信息与控制,2001,30(6)：486-492
[46]魏海坤.神经网络结构设计的理论与方法.北京：国防工业出版社,2005.2
[47]诸静,等,著.模糊控制原理与应用.北京：机械工业出版社,1999：1-1 出生年月日：1983、10、9获得学士学位的学校、时间：解放军信息工程大学、2007年7月获得硕士学位的学校、时间：郑州大学、2010年6月攻读学位期间发表论文：
[1]徐力平,张华杰.平台罗经故障检测的BP神经网络.郑州大学(理学版).2009,41(3)：30-33
[2]徐力平,张华杰,吴逸明.高斯隶属度函数模糊神经网络在肺癌中的应用.郑州大学(理学版)(已录用)