Neural networks for classification of strokes in electrical impedance tomography on a 3D head model

Abstract

We consider the problem of the detection of brain hemorrhages from three-dimensional (3D) electrical impedance tomography (EIT) measurements. This is a condition requiring urgent treatment for which EIT might provide a portable and quick diagnosis. We employ two neural network architectures -a fully connected and a convolutional one -for the classification of hemorrhagic and ischemic strokes. The networks are trained on a dataset with 40 000 samples of synthetic electrode measurements generated with the complete electrode model on realistic heads with a 3-layer structure. We consider changes in head anatomy and layers, electrode shape and position, measurement noise and conductivity values. We then test the networks on several datasets of unseen EIT data, with more complex stroke modeling (different shapes and volumes), higher levels of noise and different amounts of electrode misplacement. On most test datasets we achieved ≥ 90% average accuracy with fully connected neural networks, while the convolutional ones had worse performances. Despite the use of simple neural network architectures, the results obtained are very promising and motivate the applications of EIT-based classification methods on real phantoms and ultimately on human patients.

Anatomical atlas of the upper part of the human head for electroencephalography and bioimpedance applications

“…There are works about impedance cardiography to determine stroke volume (Bernstein 2010), electrical bioimpedance sensing to determine the central aortic pressure (CAP) curves (Min et al 2019), and pulmonary artery pressure estimation using EIT (Proença et al 2020). There is an increasing interest in brain monitoring using electrical measurements, such as to monitor ventricular volume (Wembers et al 2019), rheoencephalography to assess cerebral blood flow (Bodo et al 2018, Meghdadi et al 2019, brain perfusion of rats (Dowrick et al 2016, Song et al 2018, and stroke identification (Goren et al 2018, Agnelli et al 2020, Candiani and Santacesaria 2020, Candiani et al 2019.…”

Approximation error method for imaging the human head by electrical impedance tomography

“…This work considers stroke detection and classification between ischemic and hemorrhagic strokes by EIT. The leading idea is to combine a principal component model for the variations in the anatomy of the human head [12,13] with the approximation error approach [26,27] and a certain edge-promoting reconstruction algorithm [23] in order to introduce an image reconstruction method that is robust with respect to geometric uncertainties in the measurement configuration and carries potential to identify stroke events despite the partial shielding by the skull.…”

“…In this work we combine the principal component head model from [12,13] with a 'hybrid' reconstruction approach, taking elements from both the Bayesian framework, i.e. approximation errors and likelihood models, and the regularization framework, i.e.…”

“…The learning data is formed by computing simulated EIT measurements for an ensemble of random head and skull shapes and electrode locations. The sought-for statistics are then obtained by considering the deviation of these measurements from the ones corresponding to the intended electrode positions and the mean anatomy inside the average head of the employed principal component model [13]. The actual inversion is performed by reconstructing the deviation of the conductivity (inside the examined head) from an expected conductivity using the average head model 1 , accounting for both the actual measurement noise and the approximation error noise within the Bayesian paradigm.…”

“…Section 2 recalls the complete electrode model (CEM) [15,50] that is the most accurate model known for real-world EIT measurements. Section 3 discusses the principal component parametrization for the human head, introduced originally in [12] and supplemented with a treatment of the skull in [13]. The approximation error methodology and the edge-promoting reconstruction algorithm are reviewed in Sections 5 and 6, respectively.…”