Diagnostic Options for Postoperative Complications in Surgery

Currently, scientific research is usually carried out in accordance with the postulates of evidence-based medicine Russia and foreign countries. However, implementation these principles requires deep knowledge surgery mathematical modeling. Authors: a surgeon programmer developed models involved diagnosis postoperative complications surgery. In this paper, we investigated deep, fully connected neural network for on clinical example acute appendicitis. As training set parameters, used by authors basis real data, which has state registration number form database, includes base. The interquantile range F1 measure proposed selection significant features. An approach to coding composite categorical features, characterized compact representation, proposed. For pre-processing it use step-up autoencoder. autoencoder converts selected functions into higher-dimensional space, which, according Kover's theorem, facilitates classification implemented using Keras TensorFlow libraries. To train network, Adam algorithm adaptive learning speed used. reduce effect overfitting, modern regularization method dropout-was analysis classifier quality metrics are out. evaluate characteristics k-block cross-validation was trained showed high diagnostic performance test data set.