Systematic review Artificial neural networks in pancreatic disease A. Bartosch-H ¨ arlid 1 , B. Andersson 2 , U. Aho 2 , J. Nilsson 3 and R. Andersson 2 1 Department of Cell and Organism Biology, Lund University, Departments of 2 Surgery and 3 Cardiothoracic Surgery, Heart and Lung Centre, Lund University Hospital, Lund, Sweden Correspondence to: Dr R. Andersson, Department of Surgery, Clinical Sciences Lund, Lund University Hospital, SE-221 85 Lund, Sweden (e-mail: roland.andersson@med.lu.se) Background: An artificial neural network (ANNs) is a non-linear pattern recognition technique that is rapidly gaining in popularity in medical decision-making. This study investigated the use of ANNs for diagnostic and prognostic purposes in pancreatic disease, especially acute pancreatitis and pancreatic cancer. Methods: PubMed was searched for articles on the use of ANNs in pancreatic diseases using the MeSH terms ‘neural networks (computer)’, ‘pancreatic neoplasms’, ‘pancreatitis’ and ‘pancreatic diseases’. A systematic review of the articles was performed. Results: Eleven articles were identified, published between 1993 and 2007. The situations that lend themselves best to analysis by ANNs are complex multifactorial relationships, medical decisions when a second opinion is needed and when automated interpretation is required, for example in a situation of an inadequate number of experts. Conclusion: Conventional linear models have limitations in terms of diagnosis and prediction of outcome in acute pancreatitis and pancreatic cancer. Management of these disorders can be improved by applying ANNs to existing clinical parameters and newly established gene expression profiles. Paper accepted 25 March 2008 Published online in Wiley InterScience (www.bjs.co.uk). DOI: 10.1002/bjs.6239 Introduction Acute pancreatitis is a rapidly developing inflammation of the pancreas initiated by premature activation of trypsin within the pancreatic acinar cells 1 . With an annual incidence of 20 to 40 cases per 100 000 population, it is a relatively common disease 2,3 . The condition occurs predominantly in a mild, self-limiting form, although about 20 per cent of patients develop severe acute pancreatitis with an associated mortality rate of up to 20 per cent 4 . Early identification of patients at risk of developing the severe form of disease is of great potential therapeutic value, but all current predictive systems are limited by weaknesses in this area 5–9 . Cancer is another serious disorder of the pancreas. Ductal adenocarcinoma is one of the most devastating forms of malignancy. It is the fourth leading cause of death from cancer in the Western world and has one of the lowest 5-year survival rates, with an overall survival rate of less than 1 per cent 10 . As most patients present with disease that has spread locally or with metastases, more efficient diagnostic methods are required to improve survival. Acute pancreatitis and pancreatic cancer are two poten- tially life-threatening disorders. Both have proven difficul- ties in terms of diagnosis and prognosis. Their treat- ment would benefit greatly from improved predictive systems. Artificial neural networks (ANNs) have been used successfully in making diagnostic and prognostic decisions in several clinical situations considered gen- uinely difficult, for example in predicting the outcome of terminal liver disease 11 and automated electrocardio- graphic (ECG) interpretation in the diagnosis of acute myocardial infarction 12 . Other applications include image analyses in malignant melanoma and breast cancer 13,14 , diagnosis of colonic tumours 15 and treatment of breast cancer 16 . An ANN consists of a set of processing units that simu- late neurones, analogous to the human brain 17,18 . To date, there are only a handful of reports of ANNs used as pre- dictive systems for acute pancreatitis 19–24 and pancreatic cancer 25–29 . Copyright  2008 British Journal of Surgery Society Ltd British Journal of Surgery 2008; 95: 817–826 Published by John Wiley & Sons Ltd Downloaded from https://academic.oup.com/bjs/article/95/7/817/6142865 by guest on 13 June 2022