Title: New perspectives on deep neural networks in decision support in surgery

Authors: Konstantin Savenkov; Vladimir Gorbachenko; Anatoly Solomakha

Addresses: Faculty of Computer Engineering, Department of Computer Technologies, Penza State University, 440026, Penza, Russia ' Faculty of Computer Engineering, Department of Computer Technologies, Penza State University, 440026, Penza, Russia ' Medical Institute, Department of Surgery, Penza State University, 440026, Penza, Russia

Abstract: The paper considers the development of a neural network system for predicting complications after acute appendicitis operations. A neural network of deep architecture has been developed. As a learning set, a set developed by the authors based on real clinic data was used. To select significant features, a method for selecting features based on the interquartile range of the F1-score is proposed. For preliminary processing of training data, it is proposed to use an overcomplete autoencoder. Overcomplete autoencoder converts the selected features into a space of higher dimension, which, according to Cover's theorem facilitates the classification of features according to complication and not corresponding to complication. To overcome the overfitting of the network, the dropout method of neurons was used. The neural network is implemented using the Keras and TensorFlow libraries. Trained neural network showed high diagnostic metrics on test data set.

Keywords: neural networks; features selection; learning neural networks; overfitting; overcomplete autoencoder; medical diagnostics.

DOI: 10.1504/IJDMMM.2021.119628

International Journal of Data Mining, Modelling and Management, 2021 Vol.13 No.4, pp.317 - 336

Received: 02 Mar 2020
Accepted: 03 Jun 2020

Published online: 13 Dec 2021 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article