Despite using electronic medical records, free narrative text is still widely used for medical records. Such text cannot be analyzed by statistical tools and be proceed by decision support systems. To make data from texts available for such tasks a supervised machine learning algorithms might be successfully applied. In this work, we develop and compare a prototype of a medical data extraction system based on different artificial neuron networks architectures to process free medical texts in Russian language. The best F-score (0.9763) achieved on a combination of CNN prediction model and large pre-trained word2vec model. The very close result (0.9741) has shown by the MLP model with the same embedding.