Determination of Cardiac Ejection Fraction by Electrical Impedance Tomography Using an Artificial Neural Network

详细信息    查看全文

• 作者：Rogerio G. N. Santos Filho (22)
  Luciana C. D. Campos (22)
  Rodrigo Weber dos Santos (22)
  Luis Paulo S. Barra (22)
  
• 关键词：Cardiac Mechanics ; Medical Applications ; Cardiac Ejection Fraction ; Electrical Impedance Tomography ; Artificial Neural Networks
• 刊名：Lecture Notes in Computer Science
• 出版年：2013
• 出版时间：2013
• 年：2013
• 卷：8266
• 期：1
• 页码：139-149
• 全文大小：255KB
• 作者单位：Rogerio G. N. Santos Filho (22)
  Luciana C. D. Campos (22)
  Rodrigo Weber dos Santos (22)
  Luis Paulo S. Barra (22)
  
  22. Federal University of Juiz de Fora, Juiz de Fora, Brazil
  
• ISSN：1611-3349

文摘

The cardiac ejection fraction (EF) is a clinical parameter that determines the amount of blood pumped by the heart in each cardiac cycle. An EF outside the normal range indicates the heart is contracting abnormally. Diverse non invasive methods can be applied to measure EF, like Computer Tomography, Magnetic Resonance. Nevertheless, these techniques cannot be used for the continuous monitoring of EF. On the other hand, Electrical Impedance Tomography (EIT) may be applied to obtain continuous estimations of cardiac EF. Low cost and high portability are also EITs features that justify its use. EIT consists in fixing a finite number of electrodes on the boundary of the tomography body, injecting low amplitude currents and recording the resulting potential differences. The problem we are interested is how to estimate the blood volume inside the ventricles by using the electric potentials obtained via the EIT technique. This problem is normally classified as a non-linear inverse problem. However, in this work we propose to face it as a classification problem. Because artificial neural networks (ANN) are nonlinear models simple to understand and to implement it was decided to use them in the context of EF estimation. The use of ANNs requires less computational resources than other methods. In addition, our ANN-based solution only requires as input the measurements of the electrical potentials obtained by the electrodes; and has as output only the scalar value that defines cardiac EF. In this work, ANNs were trained and tested with data from electrical potentials simulated computationally. Two-dimensional magnetic resonance images were used for the generation of synthetic EIT data set with various types of heart configurations, spanning from normal to pathological conditions. Our preliminary results indicate that the ANN-based method was very fast and was able to provide reliable estimations of cardiac EF. Therefore, we conclude that ANN is a promising technique that may support the continuous monitoring of patient’s heart condition via EIT.