Stroke classification using Virtual Hybrid Edge Detection from in silico electrical impedance tomography data

Juan Pablo Agnelli, Fernando S. Moura, Siiri Rautio, Melody Alsaker, Rashmi Murthy, Matti Lassas, Samuli Siltanen · 2025

Read Stroke classification using Virtual Hybrid Edge Detection from in silico electrical impedance tomography data by Juan Pablo Agnelli, Fernando S. Moura, Siiri Rautio, Melody Alsaker, Rashmi Murthy, Matti Lassas, Samuli Siltanen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-14704.txt.

Introduction

Juan Pablo Agnelli and six co-authors test a safer way to sort stroke types from electrical scans, in a 2025 paper. Electrical impedance tomography reads how a body conducts from surface electrodes, with no radiation and no costly magnet. Paired with machine learning, it could one day sort strokes at the bedside, where minutes matter.

Raw voltage inputs had been the norm for the learning models. The team instead fed special edge-detection functions that stay steady under noise. They built detailed, realistic virtual patients to test the pipeline honestly, a step past the toy models of earlier work. Strong results on realistic data strengthen the case for cheap stroke triage in ambulances and rural wards.

Catalog record

CREATORJuan Pablo Agnelli, Fernando S. Moura, Siiri Rautio, Melody Alsaker, Rashmi Murthy, Matti Lassas, Samuli Siltanen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-14704

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Stroke classification using Virtual Hybrid Edge Detection from in silico electrical impedance tomography data Juan Pablo Agnelli, Fernando S. Moura, Siiri Rautio, Melody Alsaker, Rashmi Murthy, Matti Lassas, Samuli Siltanen Abstract Electrical impedance tomography (EIT) is a non-invasive imaging method for recovering the internal conductivity of a physical body from electric boundary measurements. EIT combined with machine learning has shown promise for the classification of strokes. However, most previous works have used raw EIT voltage data as network inputs. We build upon a recent development which suggested the use of special noise-robust Virtual Hybrid Edge Detection (VHED) functions as network inputs, although that work used only highly simplified and mathematically ideal models. In this work we strengthen the case for the use of EIT, and VHED functions especially, for stroke classification. We design models with high detail and mathematical realism to test the use of VHED functions as inputs. Virtual patients are created using a physically detailed 2D head model which includes features known to create challenges in real-world imaging scenarios. Conductivity values are drawn from statistically realistic distributions, and phantoms are afflicted with either hemorrhagic or ischemic strokes of various shapes and sizes. Simulated noisy EIT electrode data, generated using the realistic Complete Electrode Model (CEM) as opposed to the mathematically ideal continuum model, is processed to obtain VHED functions. We compare the use of VHED functions as inputs against the alternative paradigm of using raw EIT voltages. Our results show that (i) stroke classification can be performed with high accuracy using 2D EIT data from physically detailed and mathematically realistic models, and (ii) in the presence of noise, VHED functions outperform raw data as network inputs. 1 Introduction Stroke is a serious medical condition in which blood supply to part of the brain is blocked or significantly decreased. This prevents oxygen and other nutrients from reaching that part of the brain. Strokes are mainly classified into two groups: ischemic and hemorrhagic [ 15 ] . Ischemic stroke occurs when brain blood vessels become narrowed or blocked, impeding part of the brain from receiving blood. Hemorrhagic stroke occurs when a blood vessel ruptures or leaks, releasing blood inside the brain. Brain cells start to die in minutes without proper blood supply, and therefore stroke is an emergency situation that requires quick medical intervention to reduce brain damage and improve the prognosis of the patient [ 20 , 30 , 31 ] . The treatment depends on the type of stroke. Ischemic strokes are treated with tissue plasminogen activator medication to dissolve the clots, or, in some cases, thrombectomy to mechanically remove the clots from the affected blood vessel [ 6 , 16 , 21 ] . Hemorrhagic strokes, on the other hand, require medicine to reduce blood pressure and, if possible, a medical procedure to control the bleed, such as aneurysm clipping, coiling, or AVM endovascular embolization [ 13 , 22 , 28 ] . The medicine administered to treat one type of stroke causes damaging effects in the other type. That is, plasminogen activator medication (for ischemic treatment) causes reduction in coagulation capability, which is a serious issue for the hemorrhagic stroke patient, while medicine to reduce blood pressure (for hemorrhagic treatment) increases obstruction problems in the ischemic stroke patient. Because of the antagonistic effects of the medicines, early diagnosis is key, as diagnosis must be made prior to beginning treatment. Unfortunately, ischemic and hemorrhagic strokes present very similar symptoms in visual clinical inspection [ 26 ] . At the time of this writing, X-ray computed tomography (CT) and magnetic resonance imaging (MRI) are the brain imaging techniques used to make a diagnosis, and therefore diagnosis can only be performed after transporting a patient to the hospital [ 10 , 27 , 29 ] . If a portable diagnostic method were developed, however, the necessary diagnostic equipment could be carried in the ambulance to permit earlier treatment, which could greatly improve patient outcomes. Electrical impedance tomography (EIT) is a relatively new tomographic imaging technique which is a good candidate for this purpose, since EIT devices are portable and do not require radiation shielding or other bulky and expensive setups. EIT generates tomographic images of the electrical conductivity distribution on a target by measuring the electrical potential on its external surface when an electrical current is applied to the target [ 11 ] . In this work, we combine specially constructed VHED functions processed from EIT measurements with a neural network for stroke classification, and we validate the method using numerically simulated phantoms which are much more physically detailed and mathematically realistic than those used in a previous preliminary work on this topic [ 1 ] . Other machine learning-based approaches have been proposed for the classification of strokes using EIT data, but these approaches have all considered learning directly from measured EIT voltage data. In [ 24 ] a Support Vector Machine (SVM) was applied directly to simulated EIT voltage measurements to detect the presence or absence of a hemorrhage in the brain. A 3D model was considered, and a ring of electrodes was positioned in the exterior of the model. The highly simplified head model consisted of only 2 layers: an inner brain layer and an outer layer modeling tissues external to the brain. Therefore, the important effects produced by the highly resistive skull and the highly conductive cerebrospinal fluid (CSF) were not considered. The impact of noise, lesion location and size, electrode positioning and different anatomies of the head were considered to test the methodology. A similar approach was presented in [ 14 ] , but considered a more realistic 3D head model consisting of 4 layers. This model consists of a first layer that aggregates the scalp and skull layers, followed by separate layers representing the CSF, brain tissue (aggregating gray and white matter), and finally, the ventricle layer. Additionally, the study also explored the impact of four different measurement frequencies on stroke identification and differentiation using SVM. Multi-frequency symmetry difference EIT combined with SVM was explored in [ 23 ] . This study employed a 3D head model comprising scalp, skull, CSF, and brain layers. The methodology was applied to both simulated data and a limited set of human data (35 individuals). In [ 32 ] a Residual Neural Network was proposed for the classification of strokes in different regions of the head, again using minimally processed simulated EIT voltage measurements as inputs. A simplified 2D head model consisting of three layers representing the scalp, skull, and parenchyma (brain) was used. However, the highly conductive CSF and other anatomical tissues were not considered in this model. The performance of the method was evaluated when the data was corrupted by noise, when shape deformations occurred, and when the conductivity varied in the three layers of the model. However, these conditions were considered individually, while in real-world scenarios these variations would be simultaneous. Furthermore, in that study, difference EIT data was considered as an input to the network, so the proposed method requires healthy data that may not be available in practice. In [ 9 ] , fully connected neural networks (FCNN) and convolutional neural networks (CNN) were employed for detection of brain hemorrhages from simulated

Read the complete text