On the misuses of arti"cial neural networks for prognostic and diagnostic classi"cation in oncology

The application of arti"cial neural networks (ANNs) for prognostic and diagnostic classi"cation in clinical medicine has become very popular. In particular, feed-forward neural networks have been used extensively, often accompanied by exaggerated statements of their potential. In this paper, the essentials of feed-forward neural networks and their statistical counterparts (that is, logistic regression models) are reviewed. We point out that the uncritical use of ANNs may lead to serious problems, such as the "tting of implausible functions to describe the probability of class membership and the underestimation of misclassi"cation probabilities. In applications of ANNs to survival data, further di$culties arise. Finally, the results of a search in the medical literature from 1991 to 1995 on applications of ANNs in oncology and some important common mistakes are reported. It is concluded that there is no evidence so far that application of ANNs represents real progress in the "eld of diagnosis and prognosis in oncology. Copyright ( 2000 John Wiley & Sons, Ltd.

[1]  E Biganzoli,et al.  Feed forward neural networks for the analysis of censored survival data: a partial logistic regression approach. , 1998, Statistics in medicine.

[2]  Niels Keiding,et al.  Statistical Models Based on Counting Processes , 1993 .

[3]  R L Becker,et al.  Computer-assisted image classification: use of neural networks in anatomic pathology. , 1994, Cancer letters.

[4]  D. J. Finney,et al.  The estimation from individual records of the relationship between dose and quantal response. , 1947, Biometrika.

[5]  P. S. Maclin,et al.  How to improve a neural network for early detection of hepatic cancer. , 1994, Cancer letters.

[6]  W. Vach,et al.  Neural networks and logistic regression: Part I , 1996 .

[7]  Mann A. Shoffner,et al.  Application of backpropagation neural networks to diagnosis of breast and ovarian cancer. , 1994, Cancer letters.

[8]  John Scott Bridle,et al.  Probabilistic Interpretation of Feedforward Classification Network Outputs, with Relationships to Statistical Pattern Recognition , 1989, NATO Neurocomputing.

[9]  C. Floyd,et al.  Breast cancer: prediction with artificial neural network based on BI-RADS standardized lexicon. , 1995, Radiology.

[10]  H. Akaike.  Fitting autoregressive models for prediction , 1969 .

[11]  L. Bottaci,et al.  Artificial neural networks applied to outcome prediction for colorectal cancer patients in separate institutions , 1997, The Lancet.

[12]  J. Gurney,et al.  Solitary pulmonary nodules: determining the likelihood of malignancy with neural network analysis. , 1995, Radiology.

[13]  S K Rogers,et al.  Artificial neural networks for early detection and diagnosis of cancer. , 1994, Cancer letters.

[14]  Patrick van der Smagt,et al.  Introduction to neural networks , 1995, The Lancet.

[15]  Richard D. De Veaux,et al.  [Neural Networks in Applied Statistics]: Discussion , 1996 .

[16]  David R. Cox,et al.  Regression models and life tables (with discussion , 1972 .

[17]  P. Royston,et al.  Regression using fractional polynomials of continuous covariates: parsimonious parametric modelling. , 1994 .

[18]  J S Ostrem,et al.  Application of neural nets to ultrasound tissue characterization. , 1991, Ultrasonic imaging.

[19]  Ken-ichi Funahashi,et al.  On the approximate realization of continuous mappings by neural networks , 1989, Neural Networks.

[20]  A. Walts,et al.  Neural networks as an aid in the diagnosis of lymphocyte-rich effusions. , 1995, Analytical and quantitative cytology and histology.

[21]  D V Cicchetti,et al.  Neural networks and diagnosis in the clinical laboratory: state of the art. , 1992, Clinical chemistry.

[22]  W. Baxt.  Application of artificial neural networks to clinical medicine , 1995, The Lancet.

[23]  P M Ravdin,et al.  Treatment decisions in axillary node-negative breast cancer patients. , 1992, Journal of the National Cancer Institute. Monographs.

[24]  D Faraggi,et al.  A neural network model for survival data. , 1995, Statistics in medicine.

[25]  F. Mandelli,et al.  Classification of patients affected by multiple myeloma using a neural network software , 1994, European journal of haematology.

[26]  H Kolles,et al.  Automated grading of astrocytomas based on histomorphometric analysis of Ki-67 and Feulgen stained paraffin sections. Classification results of neuronal networks and discriminant analysis. , 1995, Analytical cellular pathology : the journal of the European Society for Analytical Cellular Pathology.

[27]  Ian Diamond,et al.  Analysis of Binary Data. 2nd Edn. , 1990 .

[28]  M. Giger,et al.  Computerized characterization of mammographic masses: analysis of spiculation. , 1994, Cancer letters.

[29]  R. Prentice,et al.  Regression analysis of grouped survival data with application to breast cancer data. , 1978, Biometrics.

[30]  Frans Van de Werf,et al.  An international randomized trial comparing four thrombolytic strategies for acute myocardial infarction. , 1993, The New England journal of medicine.

[31]  C. Floyd,et al.  Prediction of breast cancer malignancy using an artificial neural network , 1994, Cancer.

[32]  D B Fogel,et al.  Evolving neural networks for detecting breast cancer. , 1995, Cancer letters.

[33]  D. Manallack,et al.  Statistics using neural networks: chance effects. , 1993, Journal of medicinal chemistry.

[34]  R Nafe,et al.  Introduction of a neuronal network as a tool for diagnostic analysis and classification based on experimental pathologic data. , 1992, Experimental and toxicologic pathology : official journal of the Gesellschaft fur Toxikologische Pathologie.

[35]  K. Liestøl,et al.  Survival analysis and neural nets. , 1994, Statistics in medicine.

[36]  Brad Warner,et al.  Understanding Neural Networks as Statistical Tools , 1996 .

[37]  R. H. Moss,et al.  Neural network diagnosis of malignant melanoma from color images , 1994, IEEE Transactions on Biomedical Engineering.

[38]  T J O'Leary,et al.  Computer-assisted image interpretation: use of a neural network to differentiate tubular carcinoma from sclerosing adenosis. , 1992, Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc.

[39]  John S. Bridle,et al.  Training Stochastic Model Recognition Algorithms as Networks can Lead to Maximum Mutual Information Estimation of Parameters , 1989, NIPS.

[40]  J. Neijt,et al.  Predictability of the survival of patients with advanced ovarian cancer. , 1989, Journal of clinical oncology : official journal of the American Society of Clinical Oncology.

[41]  Yoshua Bengio,et al.  Pattern Recognition and Neural Networks , 1995 .

[42]  J W Moul,et al.  Neural network analysis of quantitative histological factors to predict pathological stage in clinical stage I nonseminomatous testicular cancer. , 1995, The Journal of urology.

[43]  D. M. Titterington,et al.  Neural Networks: A Review from a Statistical Perspective , 1994 .

[44]  P A Lachenbruch,et al.  Some Misuses of Discriminant Analysis , 1977, Methods of Information in Medicine.

[45]  W. Reinus,et al.  Diagnosis of Focal Bone Lesions Using Neural Networks , 1994, Investigative radiology.

[46]  H. Kappen,et al.  Neural network analysis to predict treatment outcome , 1993 .

[47]  George Cybenko,et al.  Approximation by superpositions of a sigmoidal function , 1989, Math. Control. Signals Syst..

[48]  F. Harrell,et al.  Artificial neural networks improve the accuracy of cancer survival prediction , 1997, Cancer.

[49]  Kjell A. Doksum,et al.  On a Correspondence between Models in Binary Regression Analysis and in Survival Analysis , 1990 .

[50]  R. Dybowski,et al.  Artificial neural networks in pathology and medical laboratories , 1995, The Lancet.

[51]  P. McCullagh,et al.  Generalized Linear Models, 2nd Edn. , 1990 .

[52]  Shun-ichi Amari,et al.  Network information criterion-determining the number of hidden units for an artificial neural network model , 1994, IEEE Trans. Neural Networks.

[53]  J V Tu,et al.  Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes. , 1996, Journal of clinical epidemiology.

[54]  Geoffrey E. Hinton.  Connectionist Learning Procedures , 1989, Artif. Intell..

[55]  H. Burke.  Increasing the power of surrogate endpoint biomarkers: the aggregation of predictive factors. , 1994, Journal of cellular biochemistry. Supplement.

[56]  W Penny,et al.  Neural Networks in Clinical Medicine , 1996, Medical decision making : an international journal of the Society for Medical Decision Making.

[57]  C. Niederberger.  This month in Investigative Urology. Commentary on the use of neural networks in clinical urology. , 1995, The Journal of urology.

[58]  Warren S. Sarle,et al.  Neural Networks and Statistical Models , 1994 .

[59]  H. Burke,et al.  Artificial neural networks for cancer research: outcome prediction. , 1994, Seminars in surgical oncology.

[60]  H. Maeta,et al.  Prediction of the early prognosis of the hepatectomized patient with hepatocellular carcinoma with a neural network. , 1995, Computers in biology and medicine.

[61]  B Parmanto,et al.  Building Clinical Classifiers Using Incomplete Observations – A Neural Network Ensemble for Hepatoma Detection in Patients with Cirrhosis , 1995, Methods of Information in Medicine.

[62]  A M Marchevsky,et al.  Image analysis and diagnostic classification of hepatocellular carcinoma using neural networks and multivariate discriminant functions. , 1994, Laboratory investigation; a journal of technical methods and pathology.

[63]  SchumacherMartin,et al.  Neural networks and logistic regression: Part II , 1996 .

[64]  Geoffrey E. Hinton,et al.  A comparison of statistical learning methods on the Gusto database. , 1998, Statistics in medicine.

[65]  C. R. Schweiger,et al.  Evaluation of laboratory data by conventional statistics and by three types of neural networks. , 1993, Clinical chemistry.

[66]  D. Weinberg,et al.  Nuclear grading of breast carcinoma by image analysis. Classification by multivariate and neural network analysis. , 1991, American journal of clinical pathology.

[67]  D. F. Andrews,et al.  Data : a collection of problems from many fields for the student and research worker , 1985 .

[68]  S. Morgenthaler.  Least-Absolute-Deviations Fits for Generalized Linear Models , 1992 .

[69]  P. Wilding,et al.  Application of neural networks to the interpretation of laboratory data in cancer diagnosis. , 1992, Clinical chemistry.

[70]  Y Attikiouzel,et al.  Applications of neural networks in medicine. , 1995, Australasian physical & engineering sciences in medicine.

[71]  K Griffiths,et al.  A non-invasive test for the pre-cancerous breast. , 1995, European Journal of Cancer.

[72]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[73]  Brian D. Ripley,et al.  Statistical aspects of neural networks , 1993 .