Cluster-oriented discriminant analysis; taxonomic classification of the thyroid function.

A computer assisted procedure for the diagnosis of thyroid diseases, based on seven clinical chemical parameters, is proposed. The population studied consisted of 592 consecutive outpatients with a tentative diagnosis of thyroid disease. Thyroxine, triiodothyronine, T3 uptake test, TSH before and after TRH application, its difference and thyroxine binding globulin have been determined. The patients were clinically examined and in each case a Tc-scintigram was obtained. As a first step, 20 biochemical patterns were defined by cluster analysis (pattern cognition). T check how good in grouping process was, linear discriminant analysis was applied after which the reclassification rates were very satisfactory. The clusters found corresponded well to the pathophysiological situations with some exceptions. As a second step, patients were assigned to these patterns by use of the derived discriminant functions (pattern recognition). The proposed method seems to have some advantages over other diagnostic models published hitherto.

Performance of various mathematical methods for computer-aided processing of radioimmunoassay results.

Interpolation and regression methods are available for computer aided determination of radioimmunological end results. We compared the performance of 6 algorithms (weighted and unweighted linear logit log regression; quadratic logit log regression, smoothing spline interpolation with a large and small smoothing factor, respectively, and polygonal interpolation and the manual curve fitting on the basis of three radioimmunoassays with different reference curve characteristics (digoxin, estriol, human chorionic somatomammotrophin (HCS)). Great store was set by the accuracy of the approximation at the intermediate points on the curve, i.e. those points that lie midway between two standard concentrations. These concentrations were obtained by weighing and inserted as unknown samples. In the case of digoxin and estriol the polygonal interpolation provided the best results, while the weighted logit log regression proved superior in the case of HCS.