Analysis of respiratory pressure–volume curves in intensive care medicine using inductive machine learning
Introduction
Artificial ventilation is one of the key therapies in intensive care medicine. In particular if the lung is ill, it is not unproblematic, and there is always the risk of damaging it by barotrauma or volutrauma, respectively. Therefore, it is important to observe the mechanical status of the artificially ventilated lung. The two quantities most crucial for assessing the pathophysiological mechanical condition of the respiratory system (lung and thorax) are the compliance C (the volume distensibility), and the air-flow resistance R (both will be formally defined in the following section). There are two fundamentally different approaches to measuring compliance C: two-point and multiple-point methods. Two-point methods obtain measurements at two different volumes, usually at the end of inspiration and at the end of expiration, whereas multiple-point methods obtain measurements during the whole breath. In order to find the respective advantages and disadvantages of various two- and multiple-point methods, the university hospitals of Aachen, Berlin, Bonn, Freiburg i. Br., Göttingen, Hamburg, Mannheim-Heidelberg and Munich are conducting a multi-center study in collaboration with the company Drägerwerk AG, Lübeck. Three multi-point methods (LOW-FLOW, SUPER-SYRINGE, SLICE) as well as two two-point methods (static compliance by automated single steps (SCASS) and PEEP-WAVE) are applied in randomized order to each patient participating in the study.
The central issue of the multi-center study is to discover systematic differences among the pressure–volume data generated by these five different measurement methods. Our aim was to show that data mining methods, and in particular methods from inductive machine learning, are generally well-suited to detect structural characteristics of measured pressure–volume curves. Another goal was to obtain interpretable results, such that physicians could inspect them and relate them to prior knowledge. This would advance the state of the art in the area, since at present pressure–volume curves are “just” analyzed visually: important parameters like the lower inflection point LIP (to be defined later) and the upper inflection point UIP, are determined upon visual inspection of the pressure–volume curves. Both the lower and upper inflection point are critical for setting the parameters of artificial ventilatory systems. So, the application of data mining and machine learning tools might not only lead to new insights about the similarities and differences among these measurement methods, but ultimately contribute to more objective decisions.
Many different approaches can be taken to perform data mining. Some data mining techniques originate from the field of machine learning. One advantage of the machine learning methods used here is that they can flexibly handle all sorts of descriptors. For instance, in our domain it is possible to include patient data and see whether there is a connection between properties of the patients and the shape of the pressure–volume curves. In the long run, this might open the door for the individual treatment of patients under artificial ventilation.
From a machine learning perspective, this paper is concerned with the analysis of curves by means of classification and regression techniques. In particular, the classification of curves is an interesting topic, because surprisingly few papers in the literature deal with this type of classification task. Our experience was that normalization and curve fitting techniques dramatically improve the classification results. Apart from these findings, our machine learning approach may be even more widely applicable, since it yields interpretable models of the similarities and differences among curves.
The organization of this paper parallels the one in Cios et al. [1]. Section 2 details the medical problem domain, Section 3 presents the datasets and Section 4 the preparation of the data. Section 5 focuses on the actual data mining step: it describes the data modeling techniques, the training and test procedures, and the results from a quantitative point of view. Subsequently, the discovered knowledge is evaluated in Section 6. Section 7 dicusses the usage and potential usage of the discovered knowledge, before a section touching upon related work. Finally, we conclude the paper and hint at possible directions of further research.
Section snippets
Understanding medical problem domain
In this section, we present the medical problem domain, the medical objectives of this study and the current solution to the addressed problem.
Fig. 1 schematically shows the connection between scanning the lung volume in several breaths and the resulting pressure–volume curves. The x-axis represents the pressure, the y-axis the volume. Each of the loops superimposed on curve A describes one complete breath with its phases of inspiration and expiration. Each of the curves describes the
Understanding the data
In this section, we have a closer look at the pressure–volume data that forms the basis of our study, and the size and format of the datasets.
For all four of the above measurement methods, we had the data from 10 patients available. In addition, we had some incomplete datasets: one for which only the LOW-FLOW method was available, two for which the SCASS method was not applied, and one for which the LOW-FLOW method was not applied. From these 50 example curves, we had to remove one SLICE curve
Preparation of the data
For the applied classification and regression algorithms (see later), the datasets had to be represented as feature vectors of fixed size.
Data mining
In this section, we present and motivate our selection of data mining techniques, and give a quantitative assessment of the data mining results. In general, inductive machine learning techniques are only one possible option for the data mining step in the knowledge discovery process. Other possibilities include clustering techniques and techniques for the discovery of frequently occurring patterns. The choice clearly depends on the problem at hand.
The given problem is to find similarities and
Classification
In this section, we present an example tree from the classification experiments and provide an interpretation of the model. We picked the best settings from our experiments, with normalization II, including the background features and parameter b. This setting achieved a mean error rate of 32.5% in 10 runs of 10-fold cross-validation. Fig. 4 shows a tree learned from all 49 examples using these settings.
Although the curve fitting was not too exact, we had a good fit in the central part of the
Using the discovered knowledge
From a medical perspective, the discovered knowledge enables an objective analysis of the pressure–volume relationship. At present, in the daily routine, physicians analyze pressure–volume curves only visually, if they are measured at all. In contrast to this approach, data mining and machine learning offer the opportunity to gain a more objective and general understanding of these measurement methods.
Furthermore, both classification and regression techniques from machine learning allow for
Related work
Inductive machine learning techniques have been applied to data from artificial ventilation before: Muller et al. [8], [9] describe the development of a knowledge-based alarm system for ventilatory therapy using techniques from inductive machine learning. Starting with a mathematical model, the authors generated data for “simulated patients”, and subsequently applied a machine learning algorithm to induce rules linking signal features from each simulated breath to events that occurred while the
Conclusion
To the best of our knowledge, this is the first application of inductive machine learning techniques to real, measured respiratory data from intensive care medicine. We compared different methods of measuring pressure–volume curves of artificially ventilated patients suffering from the ARDS. The goal was to find structural differences and similarities among the different methods. Our approach introduces an objective dimension along which the measurement methods can be analyzed: we induced
Acknowledgements
We would like to thank Drägerwerk AG for supporting the multi-center study, especially. E.-W. Schubert and T. Handzsuj. Special thanks also go to all our scientific collaborators in this multi-center study.
References (13)
- et al.
Multi-dimensional curve classification using passing-through regions
Pattern Recog. Lett.
(1999) - et al.
Total respiratory pressure–volume curves in the adult respiratory distress syndrome
Chest
(1984) - et al.
Building intelligent alarm systems by combining mathematical models and inductive machine learning techniques
Int. J. Biomed. Comput.
(1996) - et al.
Building intelligent alarm systems by combining mathematical models and inductive machine learning techniques. Part 2. Sensitivity analysis
Int. J. Biomed. Comput.
(1996) - et al.
Diagnosing myocardial perfusion SPECT bull’s-eye maps—a knowledge discovery approach
IEEE Eng. Med. Biol. Mag.
(2000) - Džeroski S, Schulze-Kremer S, Heidtke KR, Siems K, Wettschereck D. Applying ILP to diterpene structure elucidation from...
Cited by (33)
Artificial intelligence in acute respiratory distress syndrome: A systematic review
2022, Artificial Intelligence in MedicineCitation Excerpt :This precision medicine approach, which is widely applicable to perioperative medicine and critical care medicine research, may help refine the design of future trials. Ganzert S et al. [28] developed a Regression model by CUBIST and C5.0 using features such as three multi-point methods (LOW-FLOW, SUPER-SYRINGE, SLICE) as well as two two-point methods (static compliance by automated single steps; SCASS and PEEP-WAVE) to compare the various methods of measuring pressure-volume curves in artificially-ventilated ARDS patients. Data from artificially-ventilated ARDS patients' records were used.
A hover view over effectual approaches on pandemic management for sustainable cities – The endowment of prospective technologies with revitalization strategies
2021, Sustainable Cities and SocietyCitation Excerpt :For example, the CT scans possess good sensitivity towards lung diseases and, if the radiographical changes are more pronounced, then the person has an increased probability of being infected with COVID-19 when compared to others, and subsequent testing procedures can be performed to confirm the infection. The potential of AI and ML is not utilized in the field of treatment, but it can be actively involved in tasks such as automatically adjusting the parameters in the ventilators for each patient as it would possibly minimize or eliminate the time spent by the healthcare professionals to do this (Ganzert et al., 2002). Apart from these areas, AI can be useful in creating and updating the research information as and when new research is published regarding COVID-19.
Breathing process monitoring with a biaxially oriented polypropylene film based fiber Fabry–Perot sensor
2020, Optics CommunicationsRestructured society and environment: A review on potential technological strategies to control the COVID-19 pandemic
2020, Science of the Total EnvironmentCitation Excerpt :It might not possible at present to implement AI in the surgical field to treat COVID-19 infected patients. But at least, in the near future, we can make use of AI to treat the patient and also, for adjusting ventilator parameters for individual patients (Ganzert et al., 2002). AI technology is proficient for clinical trials and provides effective monitoring to attain a precise result.
Clinical decision support systems to improve the diagnosis and management of respiratory diseases
2020, Artificial Intelligence in Precision Health: From Concept to Applications