Tetralogy of Fallot (TOF) is one of the most common causes of cyanotic congenital heart disease beyond the neonatal period. Surgical repair is typically completed before the age of one and requires some days in the Intensive Care Unit (ICU) for recovery. During this critical period, patients are continuously monitored. Among the recorded vital signs, Central Venous Pressure (CVP), which measures the pressure in the vena cava near the right atrium of the heart, provides valuable insights into hemodynamic status. However, the literature on CVP in critically ill neonates is limited, and even less information is available for healthy neonates, as this parameter is not routinely monitored in neonatal care. The present thesis investigates CVP waveforms in infants and the association with their conditions during the first 24 hours following TOF repair. The study is divided into two main phases. Firstly, a pipeline for waveform analysis is developed, including time series extraction, template generation, and unsupervised classification using the Dynamic Time Warping metric. This procedure aims to verify a possible association between CVP variations and patient condition or outcome. Secondly, we aimed at identifying patient clusters based on differences of vital signs in their post-operative course. The proposed methodology involves computation of descriptive vital signs indices and corresponding 24-hour time series, feature selection and unsupervised clustering. Initially, this procedure is applied exclusively to CVP indexes. Subsequently, additional indexes from physiological signals, including electrocardiogram, arterial blood pressure and respiratory rate are included to enhance classification and interpretation. Waveform analysis identifies three distinct clusters of CVP waveforms with different occurrences during the observational time period. Time series clustering analysis reveals although the clusters converge to a similar hemodynamic state at the end of the 24-hour period, their trajectories are markedly different. This divergence is particularly evident in the first six hours.
La tetralogia di Fallot è una delle cause più comuni di cardiopatia congenita cianogena oltre il periodo neonatale. La riparazione chirurgica viene in genere completata prima del compimento dell'anno di età e richiede alcuni giorni di ricovero in terapia intensiva. Durante questo periodo critico, i pazienti sono costantemente monitorati. Tra i segni vitali registrati, la pressione venosa centrale (PVC), che misura la pressione nella vena cava vicino all'atrio destro, fornisce indicazioni preziose sullo stato emodinamico. Tuttavia, indicazioni riguardanti i valori di PVC nei neonati gravemente malati sono limitate e ancora meno informazioni sono disponibili per i neonati sani, poiché questo parametro non viene monitorato di routine nelle cure neonatali. Questo lavoro di tesi si prefigge di studiare le forme d'onda della PVC nei neonati e di verificare possibili correlazioni con le loro condizioni durante le prime 24 ore dopo l'intervento chirurgico. Lo studio è suddiviso in due fasi principali. In primo luogo, è stata sviluppata una "pipeline" per l'analisi delle forme d'onda, che comprende l'estrazione delle serie temporali, la generazione di un "template" e un clustering non supervisionato utilizzando la metrica "Dynamic Time Warping". In secondo luogo, abbiamo cercato di identificare i cluster di pazienti in base alle differenze dei segni vitali nel loro decorso post-operatorio. La metodologia proposta prevede il calcolo di indici a partire dai segni vitali e delle corrispondenti serie temporali, la selezione delle variabili più importanti e il clustering non supervisionato. Inizialmente, questa procedura viene applicata esclusivamente agli indici PVC. Successivamente, per migliorare la classificazione e l'interpretazione, sono stati inclusi altri indici provenienti da segnali fisiologici, tra cui elettrocardiogramma, pressione arteriosa e frequenza respiratoria. L'analisi delle forme d'onda identifica tre cluster distinti di PVC, con una diversa distribuzione nelle 24 ore post chirurgiche. L'analisi del clustering delle serie temporali rivela che, sebbene i cluster convergano verso uno stato emodinamico simile alla fine delle 24 ore, le loro traiettorie sono notevolmente diverse. Questa divergenza è particolarmente evidente nelle prime sei ore.
Waveform analysis of Central Venous Pressure in infants undergoing tetralogy of Fallot repair: a time series clustering approach to stratify post-operative course
COLOMBO, LUCREZIA;BERTESAGO, BENEDETTA
2023/2024
Abstract
Tetralogy of Fallot (TOF) is one of the most common causes of cyanotic congenital heart disease beyond the neonatal period. Surgical repair is typically completed before the age of one and requires some days in the Intensive Care Unit (ICU) for recovery. During this critical period, patients are continuously monitored. Among the recorded vital signs, Central Venous Pressure (CVP), which measures the pressure in the vena cava near the right atrium of the heart, provides valuable insights into hemodynamic status. However, the literature on CVP in critically ill neonates is limited, and even less information is available for healthy neonates, as this parameter is not routinely monitored in neonatal care. The present thesis investigates CVP waveforms in infants and the association with their conditions during the first 24 hours following TOF repair. The study is divided into two main phases. Firstly, a pipeline for waveform analysis is developed, including time series extraction, template generation, and unsupervised classification using the Dynamic Time Warping metric. This procedure aims to verify a possible association between CVP variations and patient condition or outcome. Secondly, we aimed at identifying patient clusters based on differences of vital signs in their post-operative course. The proposed methodology involves computation of descriptive vital signs indices and corresponding 24-hour time series, feature selection and unsupervised clustering. Initially, this procedure is applied exclusively to CVP indexes. Subsequently, additional indexes from physiological signals, including electrocardiogram, arterial blood pressure and respiratory rate are included to enhance classification and interpretation. Waveform analysis identifies three distinct clusters of CVP waveforms with different occurrences during the observational time period. Time series clustering analysis reveals although the clusters converge to a similar hemodynamic state at the end of the 24-hour period, their trajectories are markedly different. This divergence is particularly evident in the first six hours.File | Dimensione | Formato | |
---|---|---|---|
2025_04_Bertesago_Colombo_Tesi.pdf
accessibile in internet per tutti a partire dal 13/03/2026
Descrizione: Tesi
Dimensione
14.17 MB
Formato
Adobe PDF
|
14.17 MB | Adobe PDF | Visualizza/Apri |
2025_04_Bertesago_Colombo_Executive_Summary.pdf
accessibile in internet per tutti a partire dal 13/03/2026
Descrizione: Executive Summary
Dimensione
3.89 MB
Formato
Adobe PDF
|
3.89 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/10589/236038