Novel approach to computerized breath detection in lung function diagnostics

https://doi.org/10.1016/j.compbiomed.2018.07.017Get rights and content

Highlights

  • Current state-of-the-art algorithms have insufficient performance in breath detection.

  • Pre-set thresholds limit versatility of the algorithm, smoothening distorts raw data.

  • Algorithm based on general physiological principles works in different patient groups.

  • More complex use of the information in raw data leads to robust and precise algorithm.

  • The proposed breath detection algorithm has accuracy comparable to the human experts.

Abstract

Background

Breath detection, i.e. its precise delineation in time is a crucial step in lung function data analysis as obtaining any clinically relevant index is based on the proper localization of breath ends. Current threshold or smoothing algorithms suffer from severe inaccuracy in cases of suboptimal data quality. Especially in infants, the precise analysis is of utmost importance. The key objective of our work is to design an algorithm for accurate breath detection in severely distorted data.

Methods

Flow and gas concentration data from multiple breath washout test were the input information. Based on universal physiological characteristics of the respiratory tract we designed an algorithm for breath detection. Its accuracy was tested on severely distorted data from 19 patients with different types of breathing disorders. Its performance was compared to the performance of currently used algorithms and to the breath counts estimated by human experts.

Results

The novel algorithm outperformed the threshold algorithms with respect to their accuracy and had similar performance to human experts. It proved to be a highly robust and efficient approach in severely distorted data. This was demonstrated on patients with different pulmonary disorders.

Conclusion

Our newly proposed algorithm is highly robust and universal. It works accurately even on severely distorted data, where the other tested algorithms failed. It does not require any pre-set thresholds or other patient-specific inputs. Consequently, it may be used with a broad spectrum of patients. It has the potential to replace current approaches to the breath detection in pulmonary function diagnostics.

Introduction

Breath detection (i.e. finding the spot where expiration ends and the consecutive inspiration starts) is a crucial step in pulmonary function testing (PFT). It is a starting point for the computing of various clinically significant indices, performing regression analyses or making predictions. With the increasing importance of PFT as a diagnostic tool, new methods of PFT and approaches to data analysis are required especially in infants and toddlers (i.e., uncooperative children). In this age category, precise raw data analysis is of utmost importance, as the infant PFT is very prone to technical errors. Based on our clinical experience, the current PFT algorithms suffer from severe inaccuracy, which may lead to difficult and time-consuming interpretation of results or even raw data rejection.

Although breath detection is a relatively easy task for a physician, the automated detection by a computer remains a challenge, especially in cases of severely distorted data (e.g., as a result of young patients not cooperating well, severe drift etc.). An approach to the breath detection analysis is primarily determined by the signals being measured. Usually, a time-flow signal is captured. In this situation, there exist two basic algorithms for breath detection, that have already been proposed – threshold and smoothing approach, each with numerous modifications and extensions in an attempt to achieve greater reliability and accuracy [1]. The threshold approach rejects any breath having parameters below pre-set threshold. On the other hand, the smoothing approach smooths the signal to eliminate spurious breath endings. Despite the significant progress done in this field, clinicians are still facing situations in which the measured signal is too distorted to be automatically analysed.

Multiple breath washout test (MBW) is an example of a highly sensitive method recently introduced into clinical practice [2] [3], or [4]. It offers an important insight into early stages of several chronic lung diseases [5], [6]. Moreover, it does not require active breath manoeuvres and can be performed on infants during tidal breathing. Consequently, this method may yield clinically extremely relevant information. However, it requires a special approach to raw data analysis which the current algorithms may not offer. In comparison with the conventional methods which are based solely on flow, volume and pressure measurements and estimate primarily airway resistance (e.g. bodypletysmography, tidal breath analysis etc.), MBW brings a new dimension to raw data – the gas concentration signal (O2, CO2, inert gas). A current commercial software (Spiroware, Ecomedics, Duernten, Switzerland) uses concentrations only for constructing washout curves. However, this information may be also used for breath detection. The aim of our study was to design and justify a new and robust algorithm for breath detection using not only time-flow data but also the gas concentration signal. Such a breath detection algorithm can significantly outperform the current threshold-based algorithms. Moreover, its key ideas have the potential to contribute to the general design of the medical algorithms.

Section snippets

Materials and methods

Raw data from nitrogen multiple breath washout test were used as an input for the breath end analysis. Data were captured by the machine Exhalyzer D, Ecomedics, Duernten, Switzerland with software Spiroware 3.2.0, following the relevant recommendations by European cystic fibrosis society (ECFS) [7] and European Respiratory Society (ERS) [8]. The raw data were stored in. txt files with a specific structure containing time, flow, oxygen (O2), carbon dioxide (CO2) and molar mass (MM) signals, all

Results

All the A-files included in our testing could be successfully analysed by all the implemented algorithms. The analysis time was longer for Alg-OUR than for the threshold algorithms (1.35 ± 0.23s vs. 0.12 ± 0.01s, p<0.001). The manual analysis took much longer; the average analysis time was roughly estimated to be between 100 and 180s.

The two specialists in PFT working independently detected the same number of breaths in 35 out of 47 A-files (74%). In the remaining cases, differences were not

Discussion

We propose an innovative algorithm for breath detection that has similar accuracy to that of human experts. In comparison with the existing threshold based algorithms and commercial software algorithm, it exhibits a significantly higher success rate in recognition of true breaths, especially in severely distorted data. The algorithm addresses both the problems of false negative and false positive breath detection. We are convinced that its high performance is because it simultaneously uses more

Conclusion

We developed and tested a new and robust algorithm which extends the possibility to reliably detect breaths during lung function testing. It deals only with presumptions based on general human physiology with no need of arbitrarily pre-set thresholds or limitations. It can be used without modifications throughout the whole age spectrum and in different disease groups. Alg-OUR also shows higher accuracy in breath detection than the previously mentioned algorithms, even in the cases of severely

Specification and availability

Our software is written for GNU Octave. It was tested under Windows and Linux operating systems. There are no specific hardware requirements. However, processing large MBW files may take a while. The current version (with user manual and demo data files) demonstrating the aforementioned ideas is available at: http://kam.mff.cuni.cz/∼horacek/lungo.

Author's contributions

Václav Koucký contributed to the manuscript in following areas: literature search, design of the algorithm and its clinical validation - performing the study on real patients including data collection, analysis and interpretation. Jaroslav Horáček contributed to the manuscript in following areas: literature search, design of the algorithm and its software implementation. He also participated in the data interpretation. He created all the Figures. Milan Hladík contributed to the literature

Conflicts of interest

Jaroslav Horáček, Václav Koucký and Milan Hladík have nothing relevant to disclose.

Acknowledgements

We thank to all the patients and their legal representatives for the participation in the study. We thank to prof. Petr Pohunek, MD., Ph.D. from Department of Paediatrics, 2nd Faculty of Medicine, Charles University for medical consultations and Petra Vančurová from Department of Paediatrics, 2nd Faculty of Medicine, Charles University for the support during measurement. We are grateful to John Wilson and James Cragg from the Department of Languages, 2nd Faculty of Medicine, Charles University

References (20)

There are more references available in the full text version of this article.

Cited by (7)

  • Application of deep learning technique to manage COVID-19 in routine clinical practice using CT images: Results of 10 convolutional neural networks

    2020, Computers in Biology and Medicine
    Citation Excerpt :

    In this regard, Barbosa et al. compared the quantitative CT (qCT), pulmonary function testing (PFT), and semi-quantitative image scores (SQS) metrics and found that pulmonary function testing and qCT metrics demonstrated the highest accuracy for monitoring bi- and unilateral lung transplantation with an AUC of 0.771 and 0.817, respectively [19]. The potential of CAD systems can go even further and help radiologists to detect lung abnormalities by computerized analysis of pulmonary sounds [20] and flow and gas concentration of breath [21]. In this study, we propose a CAD system based on deep learning to classify COVID-19 infection versus other atypical and viral pneumonia diseases.

  • Automatic detection of breathing disorder from ballistocardiography signals

    2020, Knowledge-Based Systems
    Citation Excerpt :

    There has been research in the past five years regarding health problems related to the analysis of heart function and respiration using various methods. Specifically, they are Acharya et al. [6, 7], Fujita and Cimr [8], Hassan and Subasi [9], Horáček et al. [10] and Acharya et al. [11]. A few systems wherein ECG signal was used to detect sleep apnea are reported in Sharma et al. [12] and Sharma et al. [13].

View all citing articles on Scopus
View full text