The accuracy of radiological and computer dianoses in small bowel examinations in children.

A prospective study has been made of the accuracy of the diagnoses made by radiologists and by two computer methods of prediction in 120 children having radiological examinations of the small bowel. The predicted diagnoses were calculated using the Bayesian and Relative Likelihood techniques. The results have been compared with the final clinical diagnoses, which have been considered to be "correct". It was found that in 94 (78 per cent) children with no small-bowel disease the correct diagnosis was made by the radiologist in 79 (84 per cent), by the Bayesian method in 86 (91 per cent) and by the Relative Likelihood method in 77 (82 per cent) children. In the 26 (22 per cent) children with small-bowel disease the correct diagnosis was made by the radiologist in 21 (81 per cent), by the Bayesian method in 13 (50 per cent), and by the Relative Likelihood method in 11 (42 per cent) children. The accuracy of the three methods of interpretation of signs has been compared using the diagnostic test rating index, which takes into account the false-positive and false-negative results, as well as the correct and incorrect results. The radiologist rating was 0-76, the Bayesian method 0-57 and the Relative Likelihood method 0-51. The index for the radiologist was significantly greater than that for the Relative Likelihood method (p less than 0-05). Computer predictions are based on observations made by the radiologist and may be affected by the initial selection of signs, the compression of information into a "normal" or "abnormal" format, and the size of the bank of information available. They offer an objective test of the significance of selected radiological signs. Computer diagnosis provides a useful research tool, but in this study was less accurate than the radiologist.