| Title | Modeling Techniques for Medical Diagnosis: I. Heuristics and Learning Programs in Selected Neonatal Hepatic Disease |
| Author(s) | W. M. Lively; Stephen A. Szygenda; Charles E. Mize |
| Source | Computers and Biomedical Research, Vol. 6, Pages 393-410 |
| Publication Date | 1973 |
| Abstract | Four methods have been employed in attempting to solve the differential diagnosis problem of neonatal hepatitis and extrahepatic biliary atresia. The primary method utilized is a heuristic and learning program approach. Three other approaches are considered: Fletcher-Powell function optimization, cluster analysis, and discriminant analysis. In the heuristic and learning program approach, a heuristic model of the bilirubin flow is initially formulated and programmed in FORTRAN-V to run on a UNIVAC-1108 computer. Blockage of this flow occurs in the two disease entities under consideration, and proper interpretation of the resultant flow may be of aid in differential diagnosis. Learning program techniques utilizing learning patients' data are employed to cause convergence to the best possible model. The final model obtained presented a disease profile that allowed the correct diagnosis for all of the learning patients. This disease profile is felt to be useful for differential diagnosis. |