Periodic Reporting for period 1 - DEEPDREAM (A Data-drivEn computational mEthod for PersonalizeD healthcare in chronic REspiratory diseases through big-dAta analytics and dynamical Modelling.)
Okres sprawozdawczy: 2020-03-01 do 2022-02-28
Chronic respiratory diseases (CRDs) affect the structures of the respiratory system and include asthma, chronic obstructive pulmonary disease (EPOC), and sleep apnea syndrome (SAS). The burden of CRDs leads to major adverse effects on the quality of life of affected individuals, causing premature deaths and a serious economic impact on societies. Most recent advances in the study of CRDs are related to the identification of risk factors other than lung function, such as comorbid diseases, seriously affecting prognosis. The DEEPDREAM project aimed at investigating the mechanisms underlying CRDs and their connection with the development and progression of cardiovascular diseases.
Prediction is crucial to reduce the risk of cardiovascular complications caused by CRDs. It is known that intermittent hypoxia caused by obstructive sleep apnea (OSA) contributes to impaired autonomic cardiac function, promoting the development arrhythmias and hypertension. A set of biomarkers can be proposed to perform an accurate prediction of future adverse cardiac events whether the pathophysiological mechanisms linking OSA severity with cardiac function alterations are understood. In the case of COPD, which is characterized by progressive airflow limitation, the patients may often experience a sudden worsening of symptoms, well-known as an “exacerbation”, which represent one of the main causes of mortality. COPD can also cause many other health problems like pneumonia, lung cancer, pulmonary hypertension, and raise the risk of heart disease. A common characteristic of these CRDs is their complexity and multifactorial etiology, involving dynamic interactions between respiration and ventilation control, and among the cardiovascular, respiratory, and nervous systems.
To capture the complex interactions among the involved physiological systems and signals in CRDs, advanced methods were implemented in DEEPDREAM. Data-driven computational modeling was used to figure out relevant predictive biomarkers and to model complex relationships between involved physiological systems, by using statistical learning techniques for all available physiological data. By using probabilistic graphical models such as Bayesian Networks (BN), we can get not only correlations but also cause and effect, therefore, opening the way toward a more precise characterization and interpretation of the cardiorespiratory control system in CRDs.
The analysis of polysomnographic (PSG) recordings represents nowadays the gold-standard for the diagnosis of OSA but it is a time-consuming task. CRDs are within an area where the effective use of the Big-Data technologies holds great promise for risk assessment and personalized medical treatment. In promoting Big-Data as a source of innovation in healthcare, we made use of Big-Data approaches for structuring and organizing our large physiological datasets of OSA patients, as well as for boosting the computational tasks require to produce the insights. On the other hand, since PSG admission is a concern for many patients, accurate and automatic methods represent a great potential to be implemented on lighter hardware or wearable devices with a few sensors, making them suitable for home sleep monitoring.
The combination of advanced computational modeling with signal processing techniques applied to big physiological data, allowed us to assess the relationship between the neural, respiratory, and cardiac systems that take place during chronic respiratory diseases, offering a concrete possibility to go one step further in the early prevention of CVDs and their related complications.
Prediction is crucial to reduce the risk of cardiovascular complications caused by CRDs. It is known that intermittent hypoxia caused by obstructive sleep apnea (OSA) contributes to impaired autonomic cardiac function, promoting the development arrhythmias and hypertension. A set of biomarkers can be proposed to perform an accurate prediction of future adverse cardiac events whether the pathophysiological mechanisms linking OSA severity with cardiac function alterations are understood. In the case of COPD, which is characterized by progressive airflow limitation, the patients may often experience a sudden worsening of symptoms, well-known as an “exacerbation”, which represent one of the main causes of mortality. COPD can also cause many other health problems like pneumonia, lung cancer, pulmonary hypertension, and raise the risk of heart disease. A common characteristic of these CRDs is their complexity and multifactorial etiology, involving dynamic interactions between respiration and ventilation control, and among the cardiovascular, respiratory, and nervous systems.
To capture the complex interactions among the involved physiological systems and signals in CRDs, advanced methods were implemented in DEEPDREAM. Data-driven computational modeling was used to figure out relevant predictive biomarkers and to model complex relationships between involved physiological systems, by using statistical learning techniques for all available physiological data. By using probabilistic graphical models such as Bayesian Networks (BN), we can get not only correlations but also cause and effect, therefore, opening the way toward a more precise characterization and interpretation of the cardiorespiratory control system in CRDs.
The analysis of polysomnographic (PSG) recordings represents nowadays the gold-standard for the diagnosis of OSA but it is a time-consuming task. CRDs are within an area where the effective use of the Big-Data technologies holds great promise for risk assessment and personalized medical treatment. In promoting Big-Data as a source of innovation in healthcare, we made use of Big-Data approaches for structuring and organizing our large physiological datasets of OSA patients, as well as for boosting the computational tasks require to produce the insights. On the other hand, since PSG admission is a concern for many patients, accurate and automatic methods represent a great potential to be implemented on lighter hardware or wearable devices with a few sensors, making them suitable for home sleep monitoring.
The combination of advanced computational modeling with signal processing techniques applied to big physiological data, allowed us to assess the relationship between the neural, respiratory, and cardiac systems that take place during chronic respiratory diseases, offering a concrete possibility to go one step further in the early prevention of CVDs and their related complications.
A study of heart rate variability (HRV) during recurrent apnea was performed through several linear and non-linear indices to characterize its evolution in an experimental model developed in rats. HRV analysis was performed considering different scenarios to evaluate transients and global effects of intermittent hypoxia (IH). Moreover, several electrophysiological markers related to ventricular depolarization and repolarization were also investigated besides HRV markers. These parameters were evaluated on ECG signals acquired during the rat model to determine the effects caused by recurrent apnea in their temporal evolution. Likewise, an automatic detector for obstructive apnea episodes was also implemented using a single-ECG channel. It estimated the occurrence probability of apnea episodes based on the effects caused by respiratory obstruction on depolarization ECG markers. These markers were used to train Dynamic Bayesian Networks (DBNs), providing a suitable performance when classifying 15-s epochs as normal or apneic.
The relationship between sleep stages and HRV markers was also studied in OSA patients. In this study, the autonomic response was assessed through HRV spectral markers in patients with mild-moderate and severe OSA during normal and abnormal respiratory segments. The different sleep stages were considered to assess their impact on the occurrence and duration of apneas, besides the level of sympathetic modulation according to OSA severity. This provided a better characterization of the sleep stages’ influence on OSA severity and autonomic control.
Regarding COPD patients, we proposed a multivariate model capable of estimating/predicting the main outputs of a standard 6-minute walking test (6MWT), including the walked distance during the test, the maximum exercise-induced heart rate, and the recovery capacity. In this study, the 6MWT outputs were simultaneously estimated by training a Bayesian Network (BN) that integrated three multivariate models. The final model allowed to assess the patients’ functional status by predicting the 6MWT outcomes just from baseline parameters, but also to infer disease severity based on actual patient's 6MWT outcomes. This powerful two-way assessment tool for COPD patients opens the way to implement a continuous home monitoring system and provide more personalized care.
DEEPDREAM results were disseminated through several international conferences and journal articles, local and international symposiums.
The relationship between sleep stages and HRV markers was also studied in OSA patients. In this study, the autonomic response was assessed through HRV spectral markers in patients with mild-moderate and severe OSA during normal and abnormal respiratory segments. The different sleep stages were considered to assess their impact on the occurrence and duration of apneas, besides the level of sympathetic modulation according to OSA severity. This provided a better characterization of the sleep stages’ influence on OSA severity and autonomic control.
Regarding COPD patients, we proposed a multivariate model capable of estimating/predicting the main outputs of a standard 6-minute walking test (6MWT), including the walked distance during the test, the maximum exercise-induced heart rate, and the recovery capacity. In this study, the 6MWT outputs were simultaneously estimated by training a Bayesian Network (BN) that integrated three multivariate models. The final model allowed to assess the patients’ functional status by predicting the 6MWT outcomes just from baseline parameters, but also to infer disease severity based on actual patient's 6MWT outcomes. This powerful two-way assessment tool for COPD patients opens the way to implement a continuous home monitoring system and provide more personalized care.
DEEPDREAM results were disseminated through several international conferences and journal articles, local and international symposiums.
DEEPDREAM project has led to the development of new predictive models for personalized therapies, prognosis, and monitoring in CRDs such as COPD and OSA, potentially suitable for ambulatory and home-based systems to provide more personalized care.
We developed a powerful tool for COPD patients’ assessment, able to estimate the major 6MWT outcomes without physical exercise. The 6MWT outcomes are commonly used to evaluate the functional exercise capacity in these patients and are associated with an increased risk of hospitalization and mortality. The tool allowed to predict these outcomes from only baseline parameters, but also served to infer disease severity based on actual patient's 6MWT outcomes.
Furthermore, several electrophysiological and autonomic-nervous-system markers were explored in OSA applications to stratify potential biomarkers that predict and reduce the risk of cardiovascular complications. These biomarkers were used to implement an automatic detector for obstructive apnea episodes using a single ECG channel, providing an overall a suitable performance for online apnea detection and diagnosis in OSA patients.
Overall, the DEEPDREAM project has been developed in the context of Computational Healthcare, which is at the interface of biomedical signal processing, computational modeling, and health informatics.
We developed a powerful tool for COPD patients’ assessment, able to estimate the major 6MWT outcomes without physical exercise. The 6MWT outcomes are commonly used to evaluate the functional exercise capacity in these patients and are associated with an increased risk of hospitalization and mortality. The tool allowed to predict these outcomes from only baseline parameters, but also served to infer disease severity based on actual patient's 6MWT outcomes.
Furthermore, several electrophysiological and autonomic-nervous-system markers were explored in OSA applications to stratify potential biomarkers that predict and reduce the risk of cardiovascular complications. These biomarkers were used to implement an automatic detector for obstructive apnea episodes using a single ECG channel, providing an overall a suitable performance for online apnea detection and diagnosis in OSA patients.
Overall, the DEEPDREAM project has been developed in the context of Computational Healthcare, which is at the interface of biomedical signal processing, computational modeling, and health informatics.