Machine Learning Models for the Automatic Detection of Exercise Thresholds in Cardiopulmonary Exercising Tests: From Regression to Generation to Explanation
- Pattern 1 (moderate intensity domain) is characterized by: (a) increasing VO2 and VCO2; (b) increasing VE; (c) decreasing ventilatory equivalents (i.e., VEVO2 and VEVCO2, computed as VE/VO2 and VE/VCO2); (d) decreasing PetO2 and increasing PetCO2.
- Pattern 2 (heavy intensity domain) is characterized by: (a) increasing VO2 and VCO2; (b) increasing PetO2 and steady PetCO2; (c) increasing VEVO2 and steady VEVCO2; (d) increasing VE (with a slope greater than in Pattern 1).
- Pattern 3 (severe intensity domain) is characterized by: (a) increasing VO2 and VCO2; (b) increasing PetO2 (with a slope greater than in Pattern 2) and decreasing PetCO2; (c) increasing VEVO2 (with a slope greater than in Pattern 2) and VEVCO2; (d) increasing VE (with a slope greater than in Pattern 2).
- The estimated lactate threshold θL (or VT1 in this manuscript, i.e., the transition from Pattern 1 to 2): identifies the highest metabolic rate not associated with acidosis or metabolic homeostasis, and it corresponds to: (a) an increase in VCO2 relative to VO2 (an increase in blood lactate concentration is associated with the increase of H+, which combines with HCO3- to give an additional source of CO2); (b) the first disproportionate increase in VE (VE is regulated by the CO2 delivery to the lungs to minimize CO2 accumulation); (c) an increase in VEVO2 with no increase in VEVO2 (a consequence of the previous two points); (d) an increase in PetO2 with no consequent fall in PetCO2 (onset of the isocapnic period).
- The respiratory compensation point RCP (or VT2 in this manuscript, i.e., the transition from Pattern 2 to 3): identifies the highest metabolic rate at which homeostasis can be maintained despite a metabolic acidosis, and it corresponds to: (a) the second disproportionate increase in VE (hyperventilation relative to both VO2 and VCO2); (b) the first systematic increase in VEVCO2 relative to VO2 (a direct consequence of the previous point); (c) the first systematic decrease in PetCO2 (end of the isocapnic buffering period).
- First, applications that consider the statistics between tests. Hearn et al.  developed a feed-forward neural network (NN) for the prediction of clinical deterioration in patients with heart failure. They included the time dependence of the CPET variables by extracting features with an unsupervised classification algorithm . Inbar et al.  adopted a support vector machine (SVM) to identify chronic heart failure and chronic obstructive pulmonary disease from CPET. Sharma et al.  encoded CPET and processed the output images with a convolutional neural network (CNN) for the classification of heart failure and metabolic syndrome.
- Second, applications that focus on the data within each test. Baralis et al. , for example, implemented both an SVM and a NN with a rolling window technique which considered multiple CPET variables at a time. One of their goals was the online forecast of the VO2 values.
- First, the regression (or classification, or imputation) of an exercise intensity domain from the CPET variables. This challenge has already been taken by Zignoli et al., who developed a recurrent neural network (RNN)  and a CNN  to classify exercise intensity domains from a rolling window of CPET variables.
- Second, the generation of fake-but-realistic examples of CPET while maintaining the possibility to set the exercise thresholds a priori. To the best of the author’s knowledge, CPET data generation counts only one example in the scientific literature. Zignoli et al.  developed a conditional generative adversarial neural network (cGAN) to re-create a window of pre-selected CPET variables corresponding to an intensity-specific pattern.
- Third, the explanation of the why behind the detection of an exercise threshold. On one hand, simple regression models such as the V-slope  and the modified V-slope  can provide the expert with the physiological reason behind the disproportionate increase in VCO2 vs. VO2 and in VE vs. VCO2 at the exercise thresholds. However, their explanatory power comes at the expense of accuracy. On the other hand, the lack of explanatory power is a serious limitation of the use of machine learning models in the medical decision support [20,21]. Therefore, methods that could facilitate the explanation of the output of the machine learning algorithms are mostly needed .
2. Materials and Methods
4.4. Practical Applications
4.5. Final Considerations
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
- Wasserman, K.; Hansen, J.E.; Sue, D.Y.; Stringer, W.W.; Whipp, B.J. Principles of Exercise Testing and Interpretation: Including Pathophysiology and Clinical Applications; Lippincott Williams & Wilkins Philadelphia: Philadelphia, PA, USA, 2005. [Google Scholar]
- Bentley, D.J.; Newell, J.; Bishop, D. Incremental Exercise Test Design and Analysis: Implications for Performance Diagnostics in Endurance Athletes. Sport. Med. 2007, 37, 575–586. [Google Scholar] [CrossRef] [PubMed]
- Jones, A.M.; Poole, D.C. Oxygen uptake dynamics: From muscle to mouth–an introduction to the symposium. Med. Sci. Sport. Exerc. 2005, 37, 1542–1550. [Google Scholar] [CrossRef] [PubMed][Green Version]
- Keir, D.A.; Iannetta, D.; Maturana, F.M.; Kowalchuk, J.M.; Murias, J.M. Identification of Non-Invasive Exercise Thresholds: Methods, Strategies, and an Online App. Sport. Med. 2022, 52, 237–255. [Google Scholar] [CrossRef] [PubMed]
- Reeves, T.; Bates, S.; Sharp, T.; Richardson, K.; Bali, S.; Plumb, J.; Anderson, H.; Prentis, J.; Swart, M.; Levett, D. Cardiopulmonary exercise testing (CPET) in the United Kingdom—A national survey of the structure, conduct, interpretation and funding. Perioper. Med. 2018, 7, 2. [Google Scholar] [CrossRef] [PubMed][Green Version]
- Garrard, C.S.; Das, R. Sources of Error and Variability in the Determination of Anaerobic Threshold in Healthy Humans. Respiration 1987, 51, 137–145. [Google Scholar] [CrossRef]
- Prud’Homme, D.; Bouchard, C.; Leblance, C.; Landry, F.; Lortie, G.; Boulay, M.R. Reliability of assessments of ventilatory thresholds. J. Sport. Sci. 1984, 2, 13–24. [Google Scholar] [CrossRef]
- Gladden, L.B.; Yates, J.W.; Stremel, R.W.; Stamford, B.A. Gas exchange and lactate anaerobic thresholds: Inter- and intraevaluator agreement. J. Appl. Physiol. 1985, 58, 2082–2089. [Google Scholar] [CrossRef]
- Ekkekakis, P.; Lind, E.; Hall, E.E.; Petruzzello, S.J. Do regression-based computer algorithms for determining the ventilatory threshold agree? J. Sport. Sci. 2008, 26, 967–976. [Google Scholar] [CrossRef]
- Hearn, J.; Ross, H.J.; Mueller, B.; Fan, C.-P.; Crowdy, E.; Duhamel, J.; Walker, M.; Alba, A.C.; Manlhiot, C. Neural Networks for Prognostication of Patients with Heart Failure: Improving Performance Through the Incorporation of Breath-by-Breath Data From Cardiopulmonary Exercise Testing. Circ. Heart Fail. 2018, 11, e005193. [Google Scholar] [CrossRef]
- Christ, M.; Kempa-Liehr, A.W.; Feindt, M. Distributed and parallel time series feature extraction for industrial big data applications. arXiv 2016, arXiv:1610.07717. [Google Scholar]
- Inbar, O.; Inbar, O.; Reuveny, R.; Segel, M.J.; Greenspan, H.; Scheinowitz, M. A Machine Learning Approach to the Interpretation of Cardiopulmonary Exercise Tests: Development and Validation. Pulm. Med. 2021, 2021, 5516248. [Google Scholar] [CrossRef] [PubMed]
- Sharma, Y.; Coronato, N.; Brown, D.E. Encoding Cardiopulmonary Exercise Testing Time Series as Images for Classification using Convolutional Neural Network. arXiv 2022, arXiv:2204.12432. [Google Scholar]
- Baralis, E.; Cerquitelli, T.; Chiusano, S.; D’elia, V.; Molinari, R.; Susta, D. Early prediction of the highest workload in incremental cardiopulmonary tests. ACM Trans. Intell. Syst. Technol. 2013, 4, 1–20. [Google Scholar] [CrossRef][Green Version]
- Zignoli, A.; Fornasiero, A.; Stella, F.; Pellegrini, B.; Schena, F.; Biral, F.; Laursen, P.B. Expert-level classification of ventilatory thresholds from cardiopulmonary exercising test data with recurrent neural networks. Eur. J. Sport Sci. 2019, 19, 1221–1229. [Google Scholar] [CrossRef]
- Zignoli, A.; Fornasiero, A.; Rota, P.; Muollo, V.; Peyré-Tartaruga, L.A.; Low, D.A.; Fontana, F.Y.; Besson, D.; Pühringer, M.; Ring-Dimitriou, S.; et al. Oxynet: A collective intelligence that detects ventilatory thresholds in cardiopulmonary exercise tests. Eur. J. Sport Sci. 2021, 22, 425–435. [Google Scholar] [CrossRef] [PubMed]
- Zignoli, A.; Fruet, D. Automatic generation of realistic cardiopulmonary exercise test data with a conditional generative adversarial neural network. In Proceedings of the 2022 IEEE International Workshop on Sport, Technology and Research (STAR), Trento-Cavalese, Italy, 6–8 July 2022; pp. 29–34. [Google Scholar]
- Beaver, W.L.; Wasserman, K.; Whipp, B.J. A new method for detecting anaerobic threshold by gas exchange. J. Appl. Physiol. 1986, 60, 2020–2027. [Google Scholar] [CrossRef]
- Schneider, D.A.; Phillips, S.E.; Stoffolano, S. The simplified V-slope method of detecting the gas exchange threshold. Med. Sci. Sport. Exerc. 1993, 25, 1180–1184. [Google Scholar] [CrossRef]
- Longoni, C.; Bonezzi, A.; Morewedge, C.K. Resistance to Medical Artificial Intelligence. J. Consum. Res. 2019, 46, 629–650. [Google Scholar] [CrossRef]
- Ehrmann, D.E.; Gallant, S.N.; Nagaraj, S.; Goodfellow, S.D.; Eytan, D.; Goldenberg, A.; Mazwi, M.L. Evaluating and reducing cognitive load should be a priority for machine learning in healthcare. Nat. Med. 2022, 28, 1331–1333. [Google Scholar] [CrossRef]
- Watson, D.S.; Krutzinna, J.; Bruce, I.N.; Griffiths, C.E.; McInnes, I.B.; Barnes, M.R.; Floridi, L. Clinical applications of machine learning algorithms: Beyond the black box. BMJ 2019, 364, l886. [Google Scholar] [CrossRef][Green Version]
- Goodfellow, I.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; Bengio, Y. Generative adversarial networks. Commun. ACM 2020, 63, 139–144. [Google Scholar] [CrossRef]
- Mirza, M.; Osindero, S. Conditional Generative Adversarial Nets. arXiv 2014, arXiv:1411.1784. [Google Scholar]
- Smith, K.E.; Smith, A.O. Conditional GAN for timeseries generation. arXiv 2020, arXiv:2006.16477. [Google Scholar]
- Lipovetsky, S.; Conklin, M. Analysis of regression in game theory approach. Appl. Stoch. Model. Bus. Ind. 2001, 17, 319–330. [Google Scholar] [CrossRef]
- Shrikumar, A.; Greenside, P.; Kundaje, A. Learning Important Features Through Propagating Activation Differences. arXiv 2017, arXiv:1704.02685. [Google Scholar]
- LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015, 521, 436–444. [Google Scholar] [CrossRef]
- Ismail Fawaz, H.; Forestier, G.; Weber, J.; Idoumghar, L.; Muller, P.-A. Deep learning for time series classification: A review. Data Min. Knowl. Discov. 2019, 33, 917–963. [Google Scholar] [CrossRef][Green Version]
- Längkvist, M.; Karlsson, L.; Loutfi, A. A review of unsupervised feature learning and deep learning for time-series modeling. Pattern Recognit. Lett. 2014, 42, 11–24. [Google Scholar] [CrossRef][Green Version]
- Lundberg, S.M.; Nair, B.; Vavilala, M.S.; Horibe, M.; Eisses, M.J.; Adams, T.; Liston, D.E.; Low, D.K.-W.; Newman, S.-F.; Kim, J.; et al. Explainable machine-learning predictions for the prevention of hypoxaemia during surgery. Nat. Biomed. Eng. 2018, 2, 749–760. [Google Scholar] [CrossRef]
- Aristidou, A.; Jena, R.; Topol, E.J. Bridging the chasm between AI and clinical implementation. Lancet 2022, 399, 620. [Google Scholar] [CrossRef]
- Lundberg, S.M.; Lee, S.-I. Advances in Neural Information Processing Systems; Guyon, I., Luxburg, U.V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., Garnett, R., Eds.; Curran Associates, Inc.: Red Hook, NY, USA, 2017; Volume 30. [Google Scholar]
- Jablonski, J.A.; Angadi, S.S.; Sharma, S.; Brown, D.E. Enabling clinically relevant and interpretable deep learning models for cardiopulmonary exercise testing. In Proceedings of the 2022 IEEE Healthcare Innovations and Point of Care Technologies (HI-POCT), Houston, TX, USA, 10–11 March 2022; pp. 50–53. [Google Scholar]
- Selvaraju, R.R.; Cogswell, M.; Das, A.; Vedantam, R.; Parikh, D.; Batra, D. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proceedings of the 2017 IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 22–29 October 2017; pp. 618–626. [Google Scholar]
- Portella, J.J.; Andonian, B.J.; Brown, D.E.; Mansur, J.; Wales, D.; West, V.L.; Kraus, W.E.; Hammond, W.E. Using machine learning to identify organ system specific limitations to exercise via cardiopulmonary exercise testing. IEEE J. Biomed. Health Inform. 2022, 26, 4228–4237. [Google Scholar] [CrossRef] [PubMed]
- Topol, E.J. High-performance medicine: The convergence of human and artificial intelligence. Nat. Med. 2019, 25, 44–56. [Google Scholar] [CrossRef] [PubMed]
- Baralis, E.; Cerquitelli, T.; Chiusano, S.; Giordano, A.; Mezzani, A.; Susta, D.; Xiao, X. Predicting cardiopulmonary response to incremental exercise test. In Proceedings of the 2015 IEEE 28th International Symposium on Computer-Based Medical Systems, Sao Carlos, Brazil, 22–25 June 2015; pp. 135–140. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
© 2023 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Zignoli, A. Machine Learning Models for the Automatic Detection of Exercise Thresholds in Cardiopulmonary Exercising Tests: From Regression to Generation to Explanation. Sensors 2023, 23, 826. https://doi.org/10.3390/s23020826
Zignoli A. Machine Learning Models for the Automatic Detection of Exercise Thresholds in Cardiopulmonary Exercising Tests: From Regression to Generation to Explanation. Sensors. 2023; 23(2):826. https://doi.org/10.3390/s23020826Chicago/Turabian Style
Zignoli, Andrea. 2023. "Machine Learning Models for the Automatic Detection of Exercise Thresholds in Cardiopulmonary Exercising Tests: From Regression to Generation to Explanation" Sensors 23, no. 2: 826. https://doi.org/10.3390/s23020826