[Computer-aided diagnosis for pancreatic function test (author's transl)].

The use of multivariate nonlinear discriminant analysis raised the rate of correct classification for 585 pancreatic function tests from 84 percent by the doctor to 93 percent by computer analysis. In addition to the "normal" and "pancreatic" disease groups, a group of 388 patients was found in whom secretion levels were neither normal nor typical of pancreatic disease. For this group, nonpancreatic gastroenterologic disease was established with a diagnostic accuracy of 98 percent. Representation of secretion data by Andrews' method for the differentiation of pancreatitis from carcinoma allows moderately sensitive but highly specific testing.