A Pattern Recognition' Algorithm for Solving the Problem of Disease Prognosis

It's considered disease prognosis' problem which is close connected with the task of Byes clasifier' synthesis under condition that the set of offered to recognition patients is divided on two parts. An integrated characteristic of a patient i. e. 'class index' is proposed, for which a recurrent algorithm of distribution density functions' approached calculation on the basis of convolution integral is developed. The task of the early (first day) prognosis of pneumonia by severe poisonings with psichotropic drugs is solved. The parameters essential to the prognosis are found. It's shown that the hypothesis of their statistical independence well coordinated with real clinical data. The approximations of density functions of an 'class index' are calculated. The results are received on the base of information about 129 patients who were treated at Poisonings' Center of N.V.Sklifosovsky Emergency Medical Aid Scientific Research Institute.