A feasibility study of computer aided diagnosis in appendicitis

Abstract

In this study, the hypothesis that computer aided diagnosis could enable a more accurate differentiation between patients with acute appendicitis and those with abdominal pain but normal appendixes was examined. A data base was established by analyzing the records of 476 patients having an emergency measure appendectomy during a five year period. There were 360 or 76 per cent with acute appendicitis, 98 or 20 per cent with normal appendixes and 18 or 4 per cent with other diseases requiring operation. The records were analyzed with regard to history, physical examination and laboratory findings. The data base was then divided randomly into two parts. Part 1 was subjected to univariate discriminant analysis, using the chi-square test. The only quantities which were significantly different between appendicitis and a normal appendix were sex, duration of symptoms, anorexia and vomiting. Multivariate discriminant analysis was used to derive an abdominal pain index which discriminated between appendicitis and a normal appendix with a sensitivity of 0.82 and a specificity of 0.39. Using the abdominal pain index to evaluate the patients in part 2 of the data base, 23 or 40 per cent of the 58 patients with a normal appendix would have avoided operation. However, 31 or 18 per cent of the 169 patients with appendicitis would have not been operated upon; three of those 31 had perforated appendixes. Computer aided diagnosis was no more effective than unaided clinical diagnosis in appendicitis.