Design of medical decision support system

Astafyev Andrey Nikolaevich, Gerashchenko Sergey Ivanovich, Sharapov Sergey Ivanovich

An important aspect of assessing the reliability of medical results obtained using the methods and means of a clinical laboratory is the use of decision support systems to confirm the diagnosis. To improve the effectiveness of treatment of patients with hepatitis, the present work presents a modified hybrid model of a deep trust network. Assessing the effectiveness of treatment in this study is tracking the changes in the patient’s condition over time and can be used to assess the condition over certain periods. To train a modified model of a deep trust network, reference state data obtained by analyzing medical publications are used. Learning a modified deep trust network involves two steps, the first of which adopts a contrast divergence algorithm for optimizing hidden parameters in pre-training mode, while the second determines the output weight vector by the least squares method. Experimental results show that the proposed hybrid model based on a deep confidence network has good performance when applied in a decision support system. The general structure of a decision support system model for the medical task of assessing the severity of hepatitis is proposed.

Keywords:neural network, data classification, decision support system, prognosis of treatment