This web site use private and third party cookies to optimize your navigation, suit your preferences and perform analytical tasks. By continuing browsing accept our Cookies Policy.

Accept
13 September 2022Bilbao

The eVIDA research group from the University of Deusto presents a research work at the European Society of Cardiology Congress 2022

 

The eVIDA research group from the University of Deusto participated in the European Society of Cardiology Congress 2022 between  26 and  29 of  August 2022. The researcher Mario Fernando Jojoa Acosta (eVIDA) presented the research paper entitled  "Novel Complex-Valued Deep Learning Applied to Automatic Classification of Heart Sounds" where  the use of novel artificial intelligence algorithms based on complex numbers is proposed for the automatic detection of cardiac anomalies. 


This work develops a novel approach in which the input data are heartbeat recordings previously pre-processed for their representation in two dimensions through the use of the Wavelet transform. 


The core mission of the eVIDA team is to provide solutions which have an impact on the quality of life of the community. Thus, the results obtained from the research work contribute to the development of medical support tools for early detection of heart disease. 



About the research 


The applied algorithms are the result of the doctoral thesis of the researcher Mario Fernando Jojoa Acosta, under the direction of Professor María Begoña García-Zapirain from the University of Deusto and Professor Winston Spencer Percybrooks from the University of the North. The proposal opens up new possibilities to mitigate some Real-Value Deep learning issues while providing a complementary tool for the use of the inputs from the information phase to improve the predictive model.


A detailed description of the structures used could be found in the paper by Jojoa et al.  "A Fair Performance Comparison between Complex-Valued and Real-Valued Neural Networks for Disease Detection". 


Moreover, a general category within the theory of artificial intelligence is proposed, since the algorithms based on real numbers could be understood as a particular case of complex-valued algorithms, when their imaginary part is zero. However, this simple concept maintains a high mathematical complexity, since most of the learning algorithms for deep learning structures are based on the calculation of the gradient vector which leads to the need to use complete derivable mathematical functions where they are defined. The latter is a challenge for the eVIDA team, since the next step is to find an optimisation algorithm that works entirely in Hilbert space, and thus improving the performance metrics obtained so far. 

For more information. eVIDA Research Group