Deep learning scheme PSPNet for electrical impedance tomography

2021 ◽  
Author(s):  
Peng Wang ◽  
Haofeng Chen ◽  
Gang Ma ◽  
Rui Li ◽  
Xiaojie Wang
Keyword(s):  
Deep Learning ◽  
Electrical Impedance Tomography ◽  
Electrical Impedance ◽  
Impedance Tomography ◽  
Learning Scheme

scholarly journals Electrical Impedance Tomography-Based Abdominal Subcutaneous Fat Estimation Method Using Deep Learning

2020 ◽  
Vol 2020 ◽  
pp. 1-14
Author(s):  
Kyounghun Lee ◽  
Minha Yoo ◽  
Ariungerel Jargal ◽  
Hyeuknam Kwon
Keyword(s):  
Deep Learning ◽  
Electrical Impedance Tomography ◽  
Electrical Impedance ◽  
Subcutaneous Fat ◽  
Estimation Method ◽  
Training Dataset ◽  
Strong Nonlinearity ◽  
Learning Method ◽  
Impedance Tomography ◽  
Reconstruction Methods

This paper proposes a deep learning method based on electrical impedance tomography (EIT) to estimate the thickness of abdominal subcutaneous fat. EIT for evaluating the thickness of abdominal subcutaneous fat is an absolute imaging problem that aims at reconstructing conductivity distributions from current-to-voltage data. Existing reconstruction methods based on EIT have difficulty handling the inherent drawbacks of strong nonlinearity and severe ill-posedness of EIT; hence, absolute imaging may not be possible using linearized methods. To handle nonlinearity and ill-posedness, we propose a deep learning method that finds useful solutions within a restricted admissible set by accounting for prior information regarding abdominal anatomy. We determined that a specially designed training dataset used during the deep learning process significantly reduces ill-posedness in the absolute EIT problem. In the preprocessing stage, we normalize current-voltage data to alleviate the effects of electrodeposition and body geometry by exploiting knowledge regarding electrode positions and body geometry. The performance of the proposed method is demonstrated through numerical simulations and phantom experiments using a 10 channel EIT system and a human-like domain.

Abstract 295: Deep Learning and Electrical Impedance Tomography in the Prediction of Interventions Needed to Resuscitate From Hypoxic Pseudo-Pulseless Electrical Activity

Circulation ◽  
2020 ◽  
Vol 142 (Suppl_4) ◽  
Author(s):  
Alexander L Lindqwister ◽  
Weiyi Wu ◽  
Samuel Klein ◽  
Ethan K Murphy ◽  
Karen L Moodie ◽  
...  
Keyword(s):  
Deep Learning ◽  
Electrical Activity ◽  
Electrical Impedance Tomography ◽  
Electrical Impedance ◽  
Cardiac Activity ◽  
Pulseless Electrical Activity ◽  
Disease State ◽  
Training Data ◽  
Electrode Arrays ◽  
Impedance Tomography

Introduction: Pseudo-Pulseless Electrical Activity (p-PEA) is a form of profound cardiac shock defined as measurable cardiac activity without clinically detectable pulses. p-PEA has a distinct physiology and etiology from VF and true-PEA, and may constitute up to 40% of reported cases of cardiac arrest. Electrical impedance tomography (EIT) uses cutaneous electrodes to generate images based on cross sectional resistance. We utilized EIT to predict the number of interventions required to achieve ROSC from p-PEA. Methods: Female swine (N = 14) under intravenous anesthesia were instrumented with aortic and central venous micromanometer catheters. p-PEA was induced by ventilation with 6% oxygen in 94% nitrogen and was defined as a systolic aortic pressure less than 40 mmHg. Continuous EIT renderings were obtained from circumferential cutaneous thoracic and abdominal electrode arrays. A deep learning model was utilized to detect features within the EIT video clips of the p-PEA disease state to predict the number of treatments required to achieve ROSC. Twelve pigs were randomly selected as training data and 2 pigs as a test set. EIT images were saved as 30 second clips, resulting in 1630 clips generated. To increase generalizability, random epochs ranging from 30 - 100% of the total clip length were generated, resulting in a model capable of detecting this disease state with limited video fragments. Data were labeled based on the number of interventions required to achieve ROSC (100% O 2 , 100% O 2 + CPR, 100% O 2 + CPR + Epi, ROSC not achieved). Results: This approach yielded receiver operator characteristic curves - area under the curve (ROC-AUC, Figure 1) values of 0.75 for micro (weighted) AUC and 0.78 for macro (unweighted) AUC on a 4 class prediction model. Conclusion: EIT combined with machine learning may differentiate the required treatments needed to achieve ROSC in p-PEA.

Sign in / Sign up

Export Citation Format

Share Document