In clinical environments, such as the Intensive Care Unit (ICU), continuous and uninterrupted monitoring of vital signs is critical for the early detection of patient deterioration and prompt intervention. Since data collected in these settings are often corrupted by noise, artifacts, or recording gaps, it is important to estimate missing data for a more accurate signal assessment. In this study, we propose a novel automatic algorithm for the reconstruction of arterial blood pressure waveforms and electrocardiogram signals. The methodological core of the algorithm is based on the concept of statistical shape modeling, which basically estimates the shape variation of preceding beat waveforms in order to reconstruct noisy segments. The waveform reconstruction is achieved by combining the average beat template from a 90-second segment of clean signal preceding the gap with the main shape variations of the estimated waveform. The algorithm was validated using 35 10-minutes arterial blood pressure recordings acquired from bedside ICU patients provided by The PhysioNet / Computing in Cardiology Challenge 2010, as well as 17 1-hour blood pressure recordings from subjects admitted in the ICU and collected in the MIMIC-III Waveform Database. The validation process was carried out with the metrics proposed within the PhysioNet / Computing in Cardiology Challenge 2010 and considering the RMSE between reconstructed and original segments. All results were compared with those obtained from a challenge's winner. A further validation of the algorithm's performance was conducted using recording from MIMIC database creating fictitous gap to simulate missing data. As final validation, on ten selected signals from MIMIC-III Dataset, a set of HRV indices derived from the analysis of physiological data are calculated for both the original signal, the reconstructed signal by the presented algorithm and by the challenge winner algorithm and the signal with the unreconstructed gap. The overall obtained results are promising, as our algorithm proves to be more efficient than the challenge winner's algorithm when using the same number of reference channels. The algorithm's overall performance is remarkable, achieving high Q1 and Q2 scores and low RMSE. The Q1 and Q2 score tends to decrease as the gap interval duration increases while the RMSE increases. Furthermore, the HRV indices obtained by processing signal reconstructed by our algorithm are comparably similar to those of the original signal. Furthermore, when compared to the approach in which the gap is not reconstructed and the one reconstructed by the winner algorithm, our approach obtains better results, showing that it is always more useful try to recover missing information, rather than not considering it at all.
In ambienti clinici, come l'Unità di Terapia Intensiva, il monitoraggio continuo e ininterrotto dei segnali vitali è fondamentale per individuare precocemente il deterioramento del paziente e intervenire tempestivamente. Poiché i dati raccolti in questi ambienti sono spesso corrotti da rumore, artefatti o lacune di registrazione, è importante stimare i dati mancanti per una valutazione più accurata del segnale. In questo studio proponiamo un nuovo algoritmo automatico per la ricostruzione delle forme d'onda della pressione arteriosa e dei segnali dell'elettrocardiogramma. Il nucleo metodologico dell'algoritmo si basa sul concetto di modellazione statistica della forma, che fondamentalmente stima la variazione di forma delle forme d'onda di battito precedenti per ricostruire i segmenti rumorosi. La ricostruzione della forma d'onda si ottiene combinando il template medio del battito dalle informazioni di un segmento di 90 secondi di segnale pulito che precede il buco considerato con le principali variazioni di forma della forma d'onda stimata. L'algoritmo è stato validato utilizzando 35 registrazioni di pressione arteriosa di 10 minuti acquisite da pazienti in terapia intensiva, fornite dal PhysioNet / Computing in Cardiology Challenge 2010, nonché 17 registrazioni di pressione arteriosa di 1 ora da soggetti ricoverati in terapia intensiva e raccolte nel database delle forme d'onda MIMIC-III. Il processo di validazione è stato effettuato con le metriche proposte dalla PhysioNet / Computing in Cardiology Challenge 2010 e considerando l'RMSE tra i segmenti ricostruiti e quelli originali. Tutti i risultati sono stati confrontati con quelli ottenuti da uno dei vincitori della challenge. Un'ulteriore validazione delle prestazioni dell'algoritmo è stata condotta utilizzando le registrazioni del database MIMIC, creando gap fittizi per simulare i dati mancanti. Come convalida finale, su 10 segnali selezionati dal dataset di forme d'onda MIMIC-III, sono stati calcolati una serie di indici HRV derivati dall'analisi dei dati fisiologici sia per il segnale originale, sia per il segnale ricostruito dall'algoritmo presentato e dall'algoritmo vincitore della challenge, sia per il segnale con il bucp non ricostruito. I risultati complessivi ottenuti sono promettenti, in quanto il nostro algoritmo si dimostra più efficiente di quello del vincitore della challenge quando si utilizza lo stesso numero di canali di riferimento. Le prestazioni complessive dell'algoritmo sono notevoli, con punteggi Q1 e Q2 elevati e RMSE bassi. I punteggi Q1 e Q2 tendono a diminuire all'aumentare della durata di intervallo del buco, mentre l'RMSE aumenta. Inoltre, gli indici HRV ottenuti dall'elaborazione del segnale ricostruito dal nostro algoritmo sono comparabilmente simili a quelli del segnale originale. Inoltre, rispetto all'approccio in cui il gap non viene ricostruito e a quello ricostruito dall'algoritmo vincitore, il nostro approccio ottiene risultati migliori, dimostrando che è sempre più utile cercare di recuperare le informazioni mancanti, piuttosto che non considerarle affatto.
Physiological monitoring in the icu: signal reconstruction by a novel statistical shape modeling beat algorithm
GARAGNANI, MARGHERITA
2022/2023
Abstract
In clinical environments, such as the Intensive Care Unit (ICU), continuous and uninterrupted monitoring of vital signs is critical for the early detection of patient deterioration and prompt intervention. Since data collected in these settings are often corrupted by noise, artifacts, or recording gaps, it is important to estimate missing data for a more accurate signal assessment. In this study, we propose a novel automatic algorithm for the reconstruction of arterial blood pressure waveforms and electrocardiogram signals. The methodological core of the algorithm is based on the concept of statistical shape modeling, which basically estimates the shape variation of preceding beat waveforms in order to reconstruct noisy segments. The waveform reconstruction is achieved by combining the average beat template from a 90-second segment of clean signal preceding the gap with the main shape variations of the estimated waveform. The algorithm was validated using 35 10-minutes arterial blood pressure recordings acquired from bedside ICU patients provided by The PhysioNet / Computing in Cardiology Challenge 2010, as well as 17 1-hour blood pressure recordings from subjects admitted in the ICU and collected in the MIMIC-III Waveform Database. The validation process was carried out with the metrics proposed within the PhysioNet / Computing in Cardiology Challenge 2010 and considering the RMSE between reconstructed and original segments. All results were compared with those obtained from a challenge's winner. A further validation of the algorithm's performance was conducted using recording from MIMIC database creating fictitous gap to simulate missing data. As final validation, on ten selected signals from MIMIC-III Dataset, a set of HRV indices derived from the analysis of physiological data are calculated for both the original signal, the reconstructed signal by the presented algorithm and by the challenge winner algorithm and the signal with the unreconstructed gap. The overall obtained results are promising, as our algorithm proves to be more efficient than the challenge winner's algorithm when using the same number of reference channels. The algorithm's overall performance is remarkable, achieving high Q1 and Q2 scores and low RMSE. The Q1 and Q2 score tends to decrease as the gap interval duration increases while the RMSE increases. Furthermore, the HRV indices obtained by processing signal reconstructed by our algorithm are comparably similar to those of the original signal. Furthermore, when compared to the approach in which the gap is not reconstructed and the one reconstructed by the winner algorithm, our approach obtains better results, showing that it is always more useful try to recover missing information, rather than not considering it at all.| File | Dimensione | Formato | |
|---|---|---|---|
|
2023_05_Garagnani_Tesi_01.pdf
solo utenti autorizzati dal 18/04/2024
Descrizione: tesi
Dimensione
9.84 MB
Formato
Adobe PDF
|
9.84 MB | Adobe PDF | Visualizza/Apri |
|
2023_05_Garagnani_Executive_Summary_02.pdf
solo utenti autorizzati dal 18/04/2024
Descrizione: executive summary
Dimensione
614.29 kB
Formato
Adobe PDF
|
614.29 kB | 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/10589/208259