Hochschulschriftenserver der Hochschule Furtwangen

Ding, Chenling

Zhu, Yibo

Zhang, Shuyi

Zhao, Zhanqi

Gao, Yuan

Li, Zhe

English

Bedside electrical impedance tomography to assist the management of pulmonary embolism : A case report

Frerichs, Inéz

Becher, Tobias

L E T T E R  TO  T H E  E D I TO R Open Access
© The Author(s) 2024. Open Access  This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, 
sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and 
the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this 
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included 
in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will 
need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Frerichs et al. Annals of Intensive Care           (2024) 14:62 
https://doi.org/10.1186/s13613-024-01288-0
(R/Iest), as calculated from the change in end-expiratory 
lung impedance between PEEP of 5 and 15 cmH2O.
In physiological studies investigating different parame-
ters for individualization of PEEP settings, correct under-
standing and interpretation of the examined parameters 
is of utmost importance. Regarding the R/Iest parameter, 
a previous study suggested that its absolute value does 
not correlate with actual lung recruitment because this 
parameter cannot distinguish atelectasis and overdisten-
sion [2]. In the current study by Pavlovsky and colleagues 
[1], the ’optimum PEEP’ levels determined with the four 
abovementioned methods were not associated with R/Iest, 
leading the authors to the conclusion that ”optimal PEEP 
levels proposed by these four methods were not associ-
ated with recruitability”. This conclusion may be inac-
curate because R/Iest ignores overdistension and can be 
misleading. Since R/Iest is certainly not the gold standard 
for PEEP setting, as implicitly presented by the authors, it 
would be interesting to see if any of the other parameters 
were correlated with respect to ’optimum PEEP’ (e.g., was 
there a positive correlation between the ’positive PLE’ and 
’crossing point’ PEEP values? ).
Besides these general methodological considerations, 
we would like to point out some important inaccuracies 
in the assessment of the EIT-derived CoV parameter. The 
top three panels of Fig. 1 show the individual patient val-
ues of CoV, relative overdistension and collapse. Surpris-
ingly, the highest CoV values were identified at the lowest 
PEEP value of 5 cmH2O. The authors wrote in the Meth-
ods that CoV was ”the percentage of ventilation reach-
ing the dorsal half of the lung”. The dorsal fraction of 
Dear editor
We have read with interest the recently published article 
by Pavlovsky and colleagues [1]. The authors examined 
patients with ARDS during a decremental PEEP trial 
performed with steps of 3 cmH2O from 20 to 5 cmH2O. 
Airway and esophageal pressures as well as electrical 
impedance tomography (EIT) data were recorded simul-
taneously. Four different approaches were then applied to 
identify the ’optimum PEEP’ value from the recordings 
of each patient by determining those PEEP levels of the 
PEEP trial at which (1) the plateau pressure was 28–30 
cmH2O, (2) the positive expiratory transpulmonary pres-
sure reached its minimum (’positive PLE’), (3) the center 
of ventilation (CoV) was closest to 0.5 and (4) the relative 
lung overdistension and collapse curves crossed (’cross-
ing point’). The latter two values were derived from the 
EIT data. With the exception of the  ’crossing point’ and 
’positive PLE’, the calculated ’optimum PEEP’ values var-
ied significantly from each other, and none of the val-
ues was correlated to the recruitment-to-inflation ratio 
Annals of Intensive Care
*Correspondence:
Inéz Frerichs
frerichs@anaesthesie.uni-kiel.de
1Department of Anesthesiology and Intensive Care Medicine, University 
Medical Center, Schleswig-Holstein, Campus Kiel, Kiel, Germany
2School of Biomedical Engineering, Guangzhou Medical University, 
Guangzhou, China
3Department of Critical Care Medicine, Peking Union Medical College 
Hospital, Beijing, China
4Institute of Technical Medicine, Furtwangen University, Villingen-
Schwenningen, Germany
Methodological considerations 
in personalized methods for PEEP 
optimization with electrical impedance 
tomography
Inéz Frerichs1* , Tobias Becher1 and Zhanqi Zhao2,3,4
Page 2 of 2Frerichs et al. Annals of Intensive Care           (2024) 14:62 
ventilation should not be the highest at the lowest PEEP 
level when derecruitment is most likely to have occurred 
in the dependent lung regions. The presented CoV val-
ues also contradict the findings presented in the top right 
panel showing the highest percentage of lung collapse at 
the lowest PEEP. The presented results would only make 
sense if the patients were examined in the prone position, 
however, the authors wrote that the patients were studied 
in the semi-recumbent position.
In addition to these physiologically implausible results, 
the authors also disregarded that CoV and the dor-
sal fraction of ventilation are two separate, not mutu-
ally interchangeable EIT parameters [3]. They are 
calculated differently and they provide dissimilar infor-
mation on ventilation distribution. CoV is determined 
as the weighted geometrical center of ventilation distri-
bution and its location is projected on the ventrodorsal 
axis of the chest and given in percent of chest diameter. 
Values lower than 50% imply more ventral and values 
higher than 50% more dorsal location. The dorsal fraction 
of ventilation, however, gives just the proportion of ven-
tilation identified in the dorsal half of the analyzed func-
tional EIT image. The Consensus Statement on Chest 
EIT [4] (which the authors cite in their article´s reference 
list) provides a clear description of the EIT measures 
and recommendations for unified use and reporting. If 
such recommendations are not adopted, the readers may 
become confused and the users of EIT technology could 
even draw wrong conclusions on the performance of EIT 
parameters.
Previous studies indicated that the dorsal fraction 
of ventilation showed large variation in lung-healthy 
patients and CoV was mostly located in the ventral part 
of the chest [5]. Based on these findings, the authors’ 
decision to define the ’optimum PEEP’ as the value when 
CoV location was at exactly 50% of the ventrodorsal chest 
diameter may not be a physiologically sound approach. 
Indeed, the authors found that CoV reached the value 
close to 50% of the chest diameter only at a PEEP value 
of 20 cmH2O. Since the overdistension most prob-
ably occurred mainly in the non-dependent ventral lung 
regions, this effect ’pushed’ the CoV towards the center of 
the chest. We therefore hypothesize that lower (i.e., more 
ventrally located) CoV values might be more appropriate 
for setting PEEP in future studies or clinical use.
Abbreviations
ARDS  Acute respiratory distress syndrome
CoV  Center of ventilation
EIT  Electrical impedance tomography
PLE  Expiratory transpulmonary pressure
PEEP  Positive end-expiratory pressure
R/Iest  Recruitment-to-inflation ratio
Acknowledgements
Not applicable
Author contributions
IF: first draft of the manuscript, IF, TB, ZZ: revision of the manuscript, IF, TB, ZZ: 
final approval of the manuscript.
Funding
Not applicable.
Data availability
Not applicable.
Declarations
Ethics approval and consent to participate
Not applicable.
Consent for publication
Not applicable.
Competing interests
IF reports funding from the European Commission (project WELMO, grant 
no. 825572). TB reports consulting fees from Löwenstein Medical Innovation 
for institutional funding and lecture fees from Vyaire Medical. ZZ receives a 
consulting fee from Dräger Medical.
Received: 2 February 2024 / Accepted: 2 April 2024
References
1. Pavlovsky B, Desprez C, Richard JC, Fage N, Lesimple A, Chean D, Courtais 
A, Mauri T, Mercat A, Beloncle F. Bedside personalized methods based on 
electrical impedance tomography or respiratory mechanics to set PEEP in 
ARDS and recruitment-to-inflation ratio: a physiologic study. Ann Intensive 
Care. 2024;14(1):1.
2. Zhao Z, Chang MY, Chu CC, Chang HT, Moller K, Frerichs I, Hsu YL. The 
absolute value of recruitment-to-inflation ratio does not correlate with the 
recruited volume. Crit Care. 2023;27(1):246.
3. Frerichs I, Zhao Z, Becher T. Simple electrical impedance tomography 
measures for the assessment of ventilation distribution. Am J Respir Crit Care 
Med. 2020;201(3):386–8.
4. Frerichs I, Amato MB, van Kaam AH, Tingay DG, Zhao Z, Grychtol B, Boden-
stein M, Gagnon H, Bohm SH, Teschner E, Stenqvist O, Mauri T, Torsani V, 
Camporota L, Schibler A, Wolf GK, Gommers D, Leonhardt S, Adler A. Chest 
electrical impedance tomography examination, data analysis, terminology, 
clinical use and recommendations: consensus statement of the TRanslational 
EIT developmeNt stuDy group. Thorax. 2017;72(1):83–93.
5. Yang L, Dai M, Cao X, Moller K, Dargvainis M, Frerichs I, Becher T, Fu F, Zhao 
Z. Regional ventilation distribution in healthy lungs: can reference values be 
established for electrical impedance tomography parameters? Ann Transl 
Med. 2021;9(9):789.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in 
published maps and institutional affiliations. 


Methodological considerations in personalized methods for PEEP optimization with electrical impedance tomography

Sun, Qianhui

Chase, James G.

Zhou, Cong

Tawhai, Merryn H.

Knopp, Jennifer L.

Möller, Knut

Shaw, Geoffrey M.

Desaive, Thomas

Estimating patient spontaneous breathing effort in mechanical ventilation using a b-splines function approach

Shang, Chen

Yang, Ya

He, Chengcheng

Feng, Junqi

Li, Yan

Tian, Meimei

Quantitative Impact of Traditional Open Surgery and Minimally Invasive Surgery on Patients’ First-Night Sleep Status in the Intensive Care Unit : Prospective Cohort Study

Karime, Ahmad

Battistel, Alberto

Vector Network Analyzer (VNA) Measurements for Electrical Impedance Tomography

Wilkie, Jack

Rauter, Georg

Horizontal Test Stand for Bone Screw Insertion

Abdulbaki Alshirbaji, Tamer

Jalal, Nour A.

Arabian, Herag

Ding, Ning

CoHMed-Final Symposium Abstract Proceedings               DOI: 10.5281/zenodo.14268268 
 
19                                                                                              12 September 2024, Villingen-Schwenningen, Germany.  
DACAPO: Data fusion in operating room 
T. Abdulbaki Alshirbaji1,2*, N. A. Jalal2, H. Arabian1, N. Ding1 and K. Moeller1 
1 Institute of Technical Medicine (ITeM), Furtwangen University, 78054 Villingen-Schwenningen, Germany 
2 Innovation Center Computer Assisted Surgery (ICCAS), University of Leipzig, 04103 Leipzig, Germany 
* Corresponding author, email: Tamer.AbdulbakiAlshirbaji@hs-furtwangen.de 
Abstract: Modern operating rooms (ORs) are equipped with numerous devices that generate large amounts of data, creating 
opportunities and challenges for medical teams. Current systems rely on the expertise of medical staff to process and interpret 
these data streams in real time. The OR of the future requires intelligent systems capable of interpreting and analysing multi-
source data to provide context-aware support. This project aims to develop approaches for surgical workflow recognition, a 
curial component in such systems, targeting three key tasks. Machine learning and deep learning approaches were employed 
to tackle surgical smoke detection, surgical tool classification, and surgical phase recognition. Achieved results demonstrate 
high performance of the developed models and the potential of integrating them into real practice. 
© Copyright 2024 
This is an Open Access article distributed under the terms of the Creative Commons Attribution License CC-BY 4.0., which permits unrestricted 
use, distribution, and reproduction in any medium, provided the original work is properly cited. 
 
I. Introduction 
Operating rooms (ORs) are becoming increasingly 
complex environments due to the incorporation of 
advanced medical devices, each designed to support 
medical teams inside the operating room. These devices 
generate large amount of data, which provide critical 
information about the patient’s condition and surgical 
progress. However, this abundance of technology presents 
the challenge of managing and interpreting these data 
streams in real-time. Currently, surgical teams rely on their 
experience and knowledge to interpret the data coming 
from various devices.  The future of operating operations 
will involve intelligent systems that can analyse this data in 
real time, provide relevant information to human operators 
and support medical teams in decision-making. Such 
systems can promote better communication and 
collaboration between medical teams. Furthermore, data-
driven assistance systems have the potential to optimize 
workflow, avoid possible complications or hazards, and 
improve the patient outcome. 
Analysing the surgical workflow is a crucial component to 
develop such intelligent systems. This project presents 
approaches for analysing the surgical workflow using 
laparoscopic videos. Laparoscopic videos offer important 
source of information about surgical actions, tackled organs 
or tissues and surgical tools. Analysing these data allows 
understanding surgical processes. This project handled 
three tasks which are surgical smoke detection, surgical 
tool classification and surgical phase recognition. 
II. Material and methods 
This project aims to recognise surgical workflow of 
laparoscopic procedures by focusing on three key tasks: 
smoke detection, tool classification and phase recognition. 
II.I. Data 
Videos were recorded at 25 Hz during laparoscopic 
procedures. The data collection was caried out at a clinic in 
Villingen-Schwenningen, Germany [1]. The collected 
videos were anonymised by replacing image frames when 
the camera is outside the patient body with blank images. 
After anonymisation, videos were annotated with the class 
and location of surgical tools presented in the scene. To 
evaluate the performance of the proposed smoke detection 
approach, 258333 laparoscopic images were annotated for 
smoke presence. 
Additionally, publicly-access datasets such as the Cholec80 
dataset [2] were used for training and validating the 
developed models. This dataset includes 80 laparoscopic 
videos and labels for surgical tools and phases. 
II.II. Smoke detection 
Surgical smoke, emitted from electrosurgical devices, can 
obstruct the laparoscopic camera’s view. Smoke detection 
is a crucial step toward the development of an automated 
smoke removal system, which will ensure a clear surgical 
field and improve procedural visibility for surgeons. 
Handcrafted features such as texture and color features 
were developed to distinguish smoke in the scene. Then an 
SVM model was trained using these features to detect 
smoke presence in laparoscopic images [3,4]. 
II.III. Surgical tool classification 
Identifying surgical tools in laparoscopic images is a 
challenging task, as they can be occluded by smoke, blood 
or anatomical tissues. Thus, deep learning models, in 
particular convolutional neural networks (CNNs), were 
employed to perform surgical tool classification. CNNs are 
well-suited for image recognition tasks due to their ability 
to automatically learn complex features from training data. 
To further enhance the performance of CNN, attention 
modules were integrated, allowing the model to focus on 
the most relevant regions in the image. Additionally, a 
feature fusion approach was applied to improve robustness 
and generalisability of the model [5,6]. To model the 
temporal dependencies along laparoscopic videos, a long 
short-term memory (LSTM) model was combined with the 
CoHMed-Final Symposium Abstract Proceedings                                                                 DOI: 10.5281/zenodo.14268268 
20                                                                                              12 September 2024, Villingen-Schwenningen, Germany. 
CNN model [7,8]. Moreover, a weakly-supervised tool 
localization approach was proposed to detect the location 
of surgical tools in the images using only tool presence 
binary labels [9,10]. 
II.IV. Surgical phase recognition 
Surgical phases represent the main surgical steps within a 
procedure, each involving the use of specific surgical tools 
and the execution of certain surgical actions. While CNN 
models effectively extract spatial information from 
individual frames, phase recognition is time-dependent, 
requiring the integration of temporal models to capture the 
sequential dependencies over time. Therefore, CNNs were 
combined with temporal models to leverage spatio-
temporal information encoded in the laparoscopic videos. 
Several temporal models were explored for surgical phase 
recognition, including NARX neural network [11], Hidden 
Markov model (HMM) [12,13] and LSTM network [14]. 
Since there is a strong correlation between surgical tools 
used and the surgical phase, a joint model was developed 
to recognize both tools and phases simultaneously [15]. The 
model used CNN, attention modules and LSTM. 
Additionally, a multi-map convolutional layer was 
integrated, allowing the model to perform tool localisation. 
III. Results and discussion 
This project introduces video-based approaches for 
analysing surgical workflow of the laparoscopic 
procedures. The proposed approaches achieved high 
performance for various recognition tasks. The SVM model 
achieved an accuracy of 91% for detecting smoke in 
laparoscopic images. Experimental results show that the 
model can detect the smoke regardless of its density in the 
scene, i.e. low density or fog. That indicates that the 
designed handcrafted features are effective for 
distinguishing the smoke. 
CNN architectures have demonstrated strong performance 
in classifying surgical tools, with ResNet-50 achieving an 
average precision of 92% on the Cholec80 dataset. 
Incorporating temporal information further enhanced the 
classification performance, except for the grasper. Since 
graspers are frequently employed to hold or manipulate 
tissues during the entire procedure, temporal models cannot 
learn distinctive temporal features for this tool. However, 
the frequent presence of the grasper allows the CNN to 
learn discriminative spatial features for accurate 
identification.  
Moreover, combining features at multiple stages of the 
CNN model has significantly improved its generalizability. 
Results showed a 7% improvement in the model’s 
performance when applied to data from unseen surgical 
sites and different types of procedures. This demonstrates 
the benefit of multi-stage feature fusion in enhancing the 
robustness and adaptability of CNN models across varied 
surgical environments. 
Recurrent neural networks demonstrate superior 
performance compared to NARX and Hidden Markov 
Models (HMMs) for surgical phase recognition, due to their 
ability to better capture temporal dependencies. 
Additionally, multi-task learning significantly enhances 
performance in target tasks, as the model learns more 
effective and shared features across tasks. The developed 
CNN-LSTM-based approach jointly performs surgical tool 
classification, localization, and phase recognition. This 
model achieved a mean average precision of 96% for tool 
classification, a 70% F1-score for tool localization, and 
89% precision for surgical phase recognition. 
IV. Conclusions 
This project presents video-based approaches for analysing 
surgical workflow in laparoscopic procedures. The 
proposed models for smoke detection, tool classification, 
and phase recognition achieved high performance, paving 
the way for the development of many potential applications 
to assist healthcare providers intra- and post-operatively.  
AUTHOR’S STATEMENT 
Research funding: This research has been funded by the German Federal 
Ministry of Education and Research (BMBF) under grant agreement no. 
13FH5I01IA (CoHMed / DACAPO). Conflict of interest: Authors state no 
conflict of interest. Ethical approval: The research related to human use 
complies with all the relevant national regulations, institutional policies 
and was performed in accordance with the tenets of the Helsinki 
Declaration, and has been approved by the authors’ institutional review 
board or equivalent committee. 
REFERENCES 
[1] Abdulbaki Alshirbaji T, Aldeen Jalal N, et al. Data Recording 
Framework for Physiological and Surgical Data in Operating 
Theatres. Current Directions in Biomedical Engineering 
2020;6:364–7. 
[2] Twinanda AP, Shehata S, et al. Endonet: a deep architecture for 
recognition tasks on laparoscopic videos. IEEE Transactions on 
Medical Imaging 2016;36:86–97. 
[3] Alshirbaji TA, Jalal NA, et al. Classifying smoke in laparoscopic 
videos using SVM. Current Directions in Biomedical Engineering 
2017;3:191–4. 
[4] Jalal NA, Alshirbaji TA, et al. Features for detecting smoke in 
laparoscopic videos. Current Directions in Biomedical Engineering 
2017;3:521–4. 
[5] Alshirbaji TA, Jalal NA, et al. Improving the Generalisability of 
Deep CNNs by Combining Multi-stage Features for Surgical Tool 
Classification, IEEE; 2022, p. 533–6. 
[6] Tamer AA, Jalal NA, et al. Robustness of Convolutional Neural 
Networks for Surgical Tool Classification in Laparoscopic Videos 
from Multiple Sources and of Multiple Types 2022. 
[7] Alshirbaji TA, Jalal NA, et al. A deep learning spatial-temporal 
framework for detecting surgical tools in laparoscopic videos. 
Biomedical Signal Processing and Control 2021;68:102801. 
[8] Jalal NA, Abdulbaki Alshirbaji T, et al. Surgical Tool Detection in 
Laparoscopic Videos by Modeling Temporal Dependencies Between 
Adjacent Frames, Springer; 2020, p. 1045–52. 
[9] Jalal NA, Alshirbaji TA, et al. Surgical tool classification & 
localisation using attention and multi-feature fusion deep learning 
approach. IFAC-PapersOnLine 2023;56:5626–31. 
[10] Zia A, Bhattacharyya K, et al. Surgical tool classification and 
localization: results and methods from the MICCAI 2022 
SurgToolLoc challenge. arXiv Preprint arXiv:230507152 2023. 
[11] Jalal NA, Alshirbaji TA, et al. Predicting surgical phases using 
CNN-NARX neural network. Current Directions in Biomedical 
Engineering 2019;5:405–7. 
[12] Jalal NA, Abdulnaki Alshirbaji T, et al. Evaluating convolutional 
neural network and hidden markov model for recognising surgical 
phases in sigmoid resection. Current Directions in Biomedical 
Engineering 2018;4:415–8. 
[13] Jalal NA, Abdulbaki Alshirbaji T, et al. The effect of Hierarchical 
Hidden Markov Model on Phase Recognition in Laparoscopic 
Videos, Pabst; 2018, p. 61–3. 
[14] Jalal NA, Alshirbaji TA, et al. A deep learning framework for 
recognising surgical phases in laparoscopic videos. IFAC-
PapersOnLine 2021;54:334–9. 
[15] Jalal NA, Alshirbaji TA, et al. Laparoscopic video analysis using 
temporal, attention, and multi-feature fusion based-approaches. 
Sensors 2023;23:1958. 


DACAPO : Data fusion in operating room

Belloni, Paola

Fundamentals of Illumination Optics

Wrzus, Cornelia

Frenkel, Marie O.

Schöne, Benjamin

Social Virtual Reality : Person perception, cognition, and interpersonal behavior in iVR (Part II)

Uhlenbrock, Friederike

Baetzner, Anke

Strahler, Jana

VR on duty – The potential of a scenario-based police training in virtual reality (VR) for improving officers’ attention, decision-making and acting under stress

Furtwangen University

Not a member yet