Cardiovascular (CV) deconditioning is a major consequence of spaceflight and consists in changes in function and structure of heart and blood vessels in response to prolonged microgravity exposure to which the human body is subjected. This translates, with marked inter-individual variability, in reduced exercise capacity and orthostatic intolerance (OI), quantified by maximum aerobic power (VO2max [ml/kg/min]) and Head-Up Tilt (HUT) test duration [min], which may affect astronauts' ability to perform mission-related tasks or constitute a concern when re-exposed to a non-terrestrial gravity field. The driving hypothesis of this thesis was that, since a relationship was demonstrated between microgravity-induced CV deconditioning and changes in cardiac electrical activity and variability, information obtainable from a 24-hour Holter ECG recording, easily acquirable during spaceflight, might be predictive of key indicators of CV deconditioning, allowing for its frequent monitoring. Based on data from 87 subjects participating in six campaigns of Head-Down Tilt bed rest with different duration (i.e., the main ground-based microgravity analogue), the research hypothesis was tested by developing Machine Learning models based on predictive features calculated from beat-to-beat and time series extracted from the 24-hour ECG trace, aimed at predicting the level of CV deconditioning in terms of VO2max and HUT test duration, measured the day following the ECG acquisition. Short- and long-term time-domain, frequency-domain, and non-linear features were calculated from each ECG and used in two predictive strategies: one based on a single ECG recording, thus without the need to perform any maximal exercise or HUT tests, and one specifically designed for aerospace applications in which the pre-spaceflight results (i.e., baseline condition of the astronaut) were also considered. Results of the different proposed predictive models demonstrated the feasibility of the hypothesized prediction approach, with an ability to distinguish subjects into different risk intervals of OI (around 80% accuracy), obtained by a-posteriori classification of predictions, and an estimation of VO2max values (RMSE < 5 ml/kg/min) comparable with results of previous studies and that could potentially be used to assess mission feasibility for individual astronauts and to correct in-flight countermeasures.
Il decondizionamento cardiovascolare (CV) è una delle principali conseguenze dei voli spaziali e consiste in cambiamenti nella funzione e nella struttura del cuore e dei vasi sanguigni in risposta all’esposizione prolungata alla microgravità a cui il corpo umano è sottoposto. Ciò si traduce, con una marcata variabilità interindividuale, in ridotta capacità di esercizio e intolleranza ortostatica (OI), quantificate rispettivamente tramite massima potenza aerobica (VO2max [ml/kg/min]) e durata dell’Head-Up Tilt (HUT) test [min], che possono influenzare la capacità degli astronauti di svolgere compiti legati alla missione o costituire un problema quando vengono riesposti a un campo di gravità non terrestre. Poiché è stata dimostrata una relazione tra il decondizionamento CV indotto dalla microgravità e variazioni dell’attività e della variabilità elettrica cardiaca, in questa tesi è stato ipotizzato che informazioni ottenibili da una registrazione Holter ECG di 24 ore, facilmente ottenibile durante i voli spaziali, potrebbero essere predittive degli indicatori chiave del decondizionamento CV, permettendone un frequente monitoraggio. Utilizzando dati relativi a 87 soggetti partecipanti in sei campagne di riposo a letto Head-Down Tilt di diversa durata (cioè il principale analogo terrestre di microgravità), l'ipotesi di ricerca è stata testata sviluppando modelli di Machine Learning basati su variabili predittive calcolate da serie battito-battito e serie temporali estratte dal tracciato ECG di 24 ore, con l’obiettivo di predire il livello di decondizionamento CV in termini di VO2max e di durata dell’HUT test, misurati il giorno successivo all’acquisizione ECG. Variabili di breve e lungo termine del dominio del tempo, della frequenza, e non-lineari sono state calcolate da ogni tracciato ECG e impiegate in due strategie predittive: una basata su una singola registrazione ECG, senza quindi necessità di eseguire nessun test di esercizio massimale o di stimolazione ortostatica, e una specificamente progettata per applicazioni aerospaziali in cui sono stati considerati anche i risultati pre-volo (cioè la condizione base dell'astronauta). I risultati dei diversi modelli predittivi proposti hanno dimostrato la praticabilità dell'approccio predittivo ipotizzato, con una capacità di distinzione dei soggetti in diverse fasce di rischio di OI (circa 80% di accuratezza), ottenute tramite classificazione a posteriori delle predizioni, e una stima dei valori di VO2max (RMSE < 5 ml/kg/min) confrontabili con risultati di studi precedenti e potenzialmente utilizzabili per valutare la fattibilità delle missioni per i singoli astronauti e correggere le contromisure applicate durante il volo.
Machine learning approaches for predicting cardiovascular deconditioning induced by prolonged microgravity exposure
Bendandi, Riccardo
2020/2021
Abstract
Cardiovascular (CV) deconditioning is a major consequence of spaceflight and consists in changes in function and structure of heart and blood vessels in response to prolonged microgravity exposure to which the human body is subjected. This translates, with marked inter-individual variability, in reduced exercise capacity and orthostatic intolerance (OI), quantified by maximum aerobic power (VO2max [ml/kg/min]) and Head-Up Tilt (HUT) test duration [min], which may affect astronauts' ability to perform mission-related tasks or constitute a concern when re-exposed to a non-terrestrial gravity field. The driving hypothesis of this thesis was that, since a relationship was demonstrated between microgravity-induced CV deconditioning and changes in cardiac electrical activity and variability, information obtainable from a 24-hour Holter ECG recording, easily acquirable during spaceflight, might be predictive of key indicators of CV deconditioning, allowing for its frequent monitoring. Based on data from 87 subjects participating in six campaigns of Head-Down Tilt bed rest with different duration (i.e., the main ground-based microgravity analogue), the research hypothesis was tested by developing Machine Learning models based on predictive features calculated from beat-to-beat and time series extracted from the 24-hour ECG trace, aimed at predicting the level of CV deconditioning in terms of VO2max and HUT test duration, measured the day following the ECG acquisition. Short- and long-term time-domain, frequency-domain, and non-linear features were calculated from each ECG and used in two predictive strategies: one based on a single ECG recording, thus without the need to perform any maximal exercise or HUT tests, and one specifically designed for aerospace applications in which the pre-spaceflight results (i.e., baseline condition of the astronaut) were also considered. Results of the different proposed predictive models demonstrated the feasibility of the hypothesized prediction approach, with an ability to distinguish subjects into different risk intervals of OI (around 80% accuracy), obtained by a-posteriori classification of predictions, and an estimation of VO2max values (RMSE < 5 ml/kg/min) comparable with results of previous studies and that could potentially be used to assess mission feasibility for individual astronauts and to correct in-flight countermeasures.File | Dimensione | Formato | |
---|---|---|---|
2022_04_Bendandi_2.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
1.04 MB
Formato
Adobe PDF
|
1.04 MB | Adobe PDF | Visualizza/Apri |
2022_04_Bendandi_1.pdf
accessibile in internet per tutti
Descrizione: Thesis
Dimensione
76.33 MB
Formato
Adobe PDF
|
76.33 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/186653