Publication

DEEP LEARNING APPROACHES FOR SMALL DIMENSIONAL BIOMEDICAL DATA

Publication, 2017

Outline

K. Pröll - DEEP LEARNING APPROACHES FOR SMALL DIMENSIONAL BIOMEDICAL DATA - Proceedings of the 29th European Modeling and Simulation Symposium EMSS 2017, Barcelona, Spain, 2017, pp. 176-180

Abstract

In this paper we apply convolutional neuronal networks in different configurations to solve prediction tasks on medical data: Given 27 blood parameters obtained by labor blood examination the classes of tumor markers C153 and PSA should be predicted. Based on former work the results of trained Multi-Layer-Perceptrons (MLP) were moderate. Our major interest was now focused on the question if the prediction quality of CNN models outperforms MLPs. We had to transform the vector of input data into a two-dimensional pseudo image and augment it with different correlation values for increasing spatial structure. Various experiments with CNNs show that the prediction quality slightly increases compared to MLPs.