Building Clinical Classifiers Using Incomplete Observations – A Neural Network Ensemble for Hepatoma Detection in Patients with Cirrhosis

Abstract:

One objective of liver transplant evaluation is to identify patients that harbor a hepatoma, but standard screening techniques are not sensitive enough. We trained neural network ensembles to predict the presence of hepatoma in patients with cirrhosis, based on information collected at the time of transplant evaluation. Network architecture and training were modified to handle missing observations. Three ensembles were trained: ensemble A using the subset with no missing observations (528 patients); ensemble B using the complete set, which included missing observations (853 patients); and ensemble C using the smaller subset, originally with complete data, but after a fixed number of observations were deleted (i. e., made “missing”). Ensemble performance on testing sets was very good. The areas under the ROC curves were 0.91, 0.89, and 0.90, for ensembles A. B, and C, respectively. Neural networks can successfully perform this classification task, and strategies can be developed that allow use of incomplete observations.

Keywords:

Neural Networks - Missing Data - Diagnosis - Hepatocellular Carcinoma

• References

• 1 Kew MC. Tumors of The Liver. Zakim D, Boyer TD. eds. Hepatology. Second Edition.. Philadelphia: WB Saunders; 1990. Vol 2: 1206-40.
• 2 Iwatsuki S, Starzl TE, Sheahan DG. et al. Hepatic resection versus transplantation for hepatocellular carcinoma. Ann Surg 1991; 214: 221-8.
• 3 Penn I. Hepatic transplantation for primary and metastatic cancers of the liver. Surgery 1991; 110: 726-34.
• 4 Doyle HR, Parmanto B, Munro PW. et al. Using artificial neural networks to detect the presence of hepatocellular carcinoma during liver transplant evaluation. In: Proceedings of the First World Congress on Computational Medicine. Austin, Texas: (in press).
• 5 Alpert E. Human alpha-l fetoprotein. In: Okuda K, Peters RL. eds. Hepatocellular Carcinoma. New York: John Wiley and Sons; 1976: 353-67.
• 6 Liebman HA, Furie BC, Tong MJ. et al. Des-gamma-carboxy (abnormal) prothrombin as a serum marker of primary hepatocellular carcinoma. N Engl J Med 1984; 310: 1427-31.
• 7 Kato I, Tominaga S, Ikari A. The risk and predictive factors for developing liver cancer among patients with decompensated liver cirrhosis. Japanese J Clin Oncol 1992; 22: 278-85.
• 8 Baxt WG. Use of an artificial neural network for the diagnosis of myocardial infarction. Ann Intern Med 1991; 115: 843-8.
• 9 Harrison FR, Marshall SJ, Kennedy RL. The early diagnosis of heart attacks: a neurocomputational approach. In. Proceedings of the International Joint Conference on Neural Networks. Seattle: 1991. 1: 1-5.
• 10 Wu Y, Giger ML, Doi K. et al. Artificial neural networks in mammography: application to decision making in the diagnosis of breast cancer. Radiology 1993; 187: 81-7.
• 11 Veselis RA, Reinsel R, Sommer A, Carlon G. Use of neural network analysis to classify electroencephalographic patterns against depth of midazolam sedation in intensive care unit patients. J Clin Monit 1991; 7: 259-67.
• 12 Lamb DJ, Niederberger CS. Artificial intelligence in medicine and male infertility. World J Urol 1993; 11: 129-36.
• 13 Doyle HR, Dvorchik I, Mitchell S. et al. Predicting outcome after liver transplantation: a connectionist approach. Ann Surg 1994; 219: 408-15.
• 14 Parmanto B, Munro PW, Doyle HR. et al. Neural network classifier for hepatoma detection. In: Proceedings of the Fourth World Congress in Neural Networks. San Diego, CA: 1994: 1-285-1-290.
• 15 Hertz J, Krogh A, Palmer RG. Introduction to the Theory of Neural Computation. (Santa Fe Institute Studies in the Sciences of Complexity. Lecture Notes: v. 1). Reading MA: Addison-Wesley; 1991
• 16 Munro P, Cosic C, Tabasko M. A network for encoding, decoding and translating locative prepositions. Connection Science 1991; 3: 225-40.
• 17 Rumelhart DE, Hinton GE, Williams RJ. Learning representations by back-propagating errors. Nature 1986; 323: 533-6.
• 18 Hinton GE. Connectionist Learning Procedures. Pittsburgh PA: Carnegie-Mellon University; Tech. Rep. CMU-CS-88-152 (Version 2) 1987
• 19 Kolen JF, Pollack JB. Backpropagation is sensitive to initial conditions. NIPS 3 1991; 860-7.
• 20 Hansen LK, Salamon P. Neural network ensembles. IEEE Trans Pattern Anal Mach Intell 1990; 12: 993-1001.
• 21 Perrone M. Improving Regression Estimation: Averaging Methods for Variance Reduction with General Extensions to General Convex Measure Optimization. PhD Thesis, Department of Physics. Providence RI: Brown University; 1993
• 22 Hanley JA. Receiver operating characteristic (ROC) methodology: the state of the art. Crit Rev Diagn Imaging 1989; 29: 307-35.
• 23 Patil S, Henry JW, Rubenfire M, Stein PD. Neural network in the clinical diagnosis of acute pulmonary embolism. Chest 1993; 104: 1685-9.
• 24 Buchman TG, Kubos KL, Seidler AJ, Siegforth MJ. A comparison of statistical and connectionist models for the prediction of chronicity in a surgical intensive care unit. Crit Care Med 1994; 22: 750-62.
• 25 Baxt WG. Improving the accuracy of an artificial neural network using multiple differently trained networks. Neural Computation 1992; 4: 772-80.
• 26 Little RJA, Rubin DB. Statistical analysis with missing data. New York: John Wiley & Sons; 1987