Introduction Blood-contacting medical devices, such as stents, grafts, catheters, extracorporeal circuits, and ventricular assist devices, are used to treat a variety of cardiovascular and cardiopulmonary diseases. The use of artificial machines to support organ functions has a great advantage related to the almost unlimited availability, in contrast with the limited one of real organs transplanted. On the other hand, artificial organs for circulatory support are associated with thromboembolic, pro-inflammatory and haemolytic complications. Complications resulting from thrombus formation in these devices are, indeed, a frequent source of failure, and thrombosis and haemorrhage are the two dominant clinical issues for mechanical circulatory support patients. The aforementioned complications not only impede the function of the device but pose direct risk to the patient. In order to improve the use of extracorporeal devices and avoid these complications, a lot of studies on clot detection and thrombus formation on different components of bio-machines were performed during the last decades. An extracorporeal device used in certain severe medical situations, when the function of critical organs may temporarily be lost or insufficient to maintain life, is the Extracorporeal Membrane Oxygenation or ECMO, so a life-saving therapy for temporary support in refractory severe lung and/or heart failure. The ECMO circuit is composed of a blood pump, normally centrifugal, that replace the blood pumping function of the heart, a membrane oxygenator used as artificial lungs, a heat exchanger to warm the blood before re-infusing it, cannulas implanted into the patient’s major vessels, and tubing generally made of polyvinylchloride-based plastic compound (PVC) along with multiple tubing connectors. Figure 1 - Example of an ECMO circuit. Source https://www.quora.com/Can-ECMO-be-used-to-replace-breathing-to-buy-time-for-rescue-if-someone-is-trapped-under-water It is commonly recognised that the most important source of enhanced thrombogenicityin the ECMO circuit is of mechanical origin, initiating a vicious biochemical circle. The centrifugal pump is generally considered the most thrombogenic part of the circuit. Clot detection in ECMO is a problem that involves invasive approaches several times and there are currently no methods to detect them prior they can actually become visible, big and problematic. In clinics is not easy to detect clots from the outside and pump thrombosis formation can occur even if the patient apparently has adequate anticoagulation. The main problem from the engineering point of view is related to the fact that a thrombus formation can easily compromise the functionality of the machinery, so when a clot is formed it’s necessary to substitute the thrombotic parts of the circuit in order to solve two problems: first of all the clot can embolize and go inside the circulatory system of the patient, so it has to be removed; secondly, the clot can reduce the efficiency of the machinery interfering with the fluid dynamics of the blood and the functioning of parts of the circuit and has to be removed to save the circuit too. The aim of this work is to improve a non-invasive method for clot detection by infrasound in an ECMO in-vitro circuit and to localise thrombosis in the different components of the bio-machine using clot analogues. This experimental project is an improvement of a previous work, in which a correlation between the presence of clots in the ECMO circuit and the amplification of the signal recorded by Infrasounds in low frequencies was found: the signals recorded come from the vibration of the piping’s walls during the flow of liquid inside the circuit, with or without clots inside the circuit itself. The experimental set-up had the purpose to emulate the interaction between the normal functioning of the ECMO and the presence of particles mimicking the clots. State of the art The idea of using acoustic data for the clot detection into life support systems’ machinery in order to aid clinical decision-making is by no means new. Such means have been proposed by several authors during the last decades and every approach is slightly different from the other in terms of which signals are analysed, how the analysis is performed and which signals’ features are taken into account: the thing in common between all the works is the fact that the acoustic data are mainly related to the pump used into the circuit. The main purpose of the present study is the in vitro validation of the proposed methodology and analysis of its sensitivity using also an additional pump with respect to that used by Fuchs et al. (2018); moreover, the improvement of the method in order to focus also on the localisation of the thrombotic events, not only on the detection of the clots themselves. The work performed by Fuchs et al. (2018) has been the main source of information for the present study. Fuchs et al. (2018) elaborated a new non-invasive method for clot detection in the ECMO circuit: specifically, the flow in an ECMO circuit using a centrifugal pump has been considered. The acoustic approach to detect thrombi has been developed and applied to clinical data gathered at the ECMO centre at Karolinska University Hospital and this method is based on the concept that the presence of blood clot can be identified by considering the low-end of the frequency spectrum, within the range of infrasound, so below 20 Hz. Clot detection has been studied experimentally, both at “bedside” and in a laboratory environment. All the results are based on data gathered by a hydrophone (model 8103, Bruel & Kjaer) and then post processed through Matlab. The circuit was determined to contain a clot if a significant low-frequency component, above a threshold level, was present in the spectrum: this threshold was set to about two orders of magnitude larger than the magnitude of the peak related to the dominating frequency without a clot, most often related to the pump rotation frequency. It is possible to analyse a case in which the presence of a clot was clear enough in the following figure: Figure 2 - Infant patient in the ECMO circuit (Patient 3). The frames are related to the inflow signals.The left and right frame are related to the same patient but the right frame was registered 5 days after the left one. The pump was runningat 2200 rpm (36.67 Hz) and 0.33 LPM at the first registration and at 2100 (35 Hz) and 0.2 LPM in the later registration. The arrows mark the pump rotation rate used for normalization of the spectrum. Note the strong signal at about 2 Hz, which as in the earlier registration is two orders of magnitude stronger than the corresponding pump rotation frequency. In all frames, the clot threshold criterion ismet with a widemargin and hence these registrations indicate a clear presence of a clot. Note also the strong signal at 50 Hz (power line frequency). Source https://www.diva-portal.org/smash/get/diva2:1120190/FULLTEXT01.pdf In this patient’s recording, a clear peak is evident at approximately 2 Hz and this peak is larger by more than an order of magnitude as compared to the peak due to the pump rotation rate indicated with the red arrow. The results indicate that a blood clot was present already during the early recording and it was then confirmed in clinics. Materials and Methods In the present study it was necessary to build up an experimental circuit in order to recreate an ECMO circuit. The main components were tubing, centrifugal pumps to be exchanged during the different measurements to measure signal recorded both from the CentriMag centrigufal pump and the Rotaflow one, an oxygenator and a plastic water tank. On the circuit ten points along the piping were marked with permanent marker, one in each section upstream and downstream of the pump. These points were used as measurements points in the next steps to be sure to have more or less reproducible measurement positions during the whole analysis. Figure 3 – The experimental ECMO circuit built up in the LTH laboratory The CentriMag pump is the pump that was already studied in the previous work of Fuchs et al. (2018): by studying again this pump I wanted to evaluate the repeatability of the non-invasive clot detection method and its sensitivity and also to examine the spectra in more detail to highlight the localization of the clots. This method, instead, was never tested on the Rotaflow pump, or in general in pumps different from the CentriMag, therefore I wanted to evaluate the response of this pump to the method. Once the circuit was assembled, it was decided that the measurements would be done at the measuring points, previously marked, and that mainly two rotational speeds of the centrifugal pump would be evaluated: 2200rpm and 3200rpm. To acquire the signals, it was used a hydrophone connected to an amplifier and then to a laptop: the audio signal was recorded using Audacity and then exported as a .wav file into Matlab; here the signal was converted in the frequency domain and the analysis was done based on the power spectral density of each signal. First of all, it was necessary to characterise the clean circuit and so acquire the audio signal simply letting run water into the circuit: the recordings were performed in each of the ten marked points and with the two different rotational speeds of the pump. This characterisation was done both for the circuit with the Rotaflow pump and the circuit with the CentriMag pump. After that, the measurements in the circuit with the clot analogues started. During experimental works is not always possible to have access to real animal or human clots, due to ethical problems and the necessity for ethical approvals: for this reason, there is the need of mimic blood clots using clot analogues. The important aspect in the choice of clot analogues is related to the fact that the materials chosen have to interact with the flow of water and not only being a passive presence in the circuit: in particular these clots analogues have to absorb a bit of water and react to the shear stresses exerted by the flow. It was chosen to use cotton yarn threads to mainly mimic fibrin-based clots and those clots that normally extends due to the rotation of the pump; other tested materials were silicone, slime and clay in order to try to mimic RBCs-based clots with different rounded shape, dimension and location. So, for this part of the project it was decided to use two hydrophones in the experimental ECMO circuit, in order to simultaneously record the signal in different points in the circuit. In order to deeply post process the dataset, an algorithm, given by the group led by Beren Semiz at the Georgia Institute of Technology and modified by me ad hoc for this research, was implied. In particular, this algorithm is used to extract features of the signal that can highlight distinguishable differences between the normal and thrombosis recordings. The extracted features were stored in a matrix, where each row represents a single frame and the columns represent the 61 features extracted. The values of these features were compared between the normal and thrombosis recordings in order to find a correlation among the change in value of some features and the presence of clots in the circuit. All of this was done using Excel applying a t-test on the two groups of recordings and evaluating the related the p-value. Results The results of this work need to be subdivided into different groups. Firstly, the results regarding how often the clots get stuck in some part of the circuit, the areas in which they get stuck and a first comparison between the frequency spectra (power spectral density) of the CentriMag and Rotaflow pumps. Subsequently, the results of the post processing of the signals further analysed thanks to the use of Excel are presented. It is necessary to underline that the most important results are related to four groups of recordings that are also compared one with respect to the other. These groups of recordings are the ones from the CentriMag pump with yarn clot analogues 1cm long, the ones from the CentriMag pump with yarn clot analogues 2cm long, the ones from the Rotaflow pump with yarn clot analogues 1cm long and the ones from the Rotaflow pump with yarn clot analogues 2cm long. CentriMag pump 1cm yarn threads During the measurements, the clot analogues got trapped mainly in three different positions inside the circuit: the tongue of the pump, that is the edge between the pump housing and the outlet of the pump; the hole of the magnet house in the pump, that is the hole present in the magnet structure of the blades; the Luer lock of injection, that is the Luer lock used in the circuit to inject the yarn threads. In this specific group of recordings the trapped clots cases account for around the 9% of the total valid measurement cases. The preferential position is the hole in the magnet of the pump, that represents the 55% of the total trapped cases, followed by the tongue of the pump that represents the 41% and lastly the Luer lock of injection representing the 5%. These positions are not casual events, indeed this localization of thrombotic events is also confirmed by computational simulations: for example, Fuchs et al. (2018) identified the hole of the magnet house and the tongue of the pump as places in the pump where platelets can activate coagulation due to high shear stress and residence time; moreover, at the outlet of the pump, where the outlet circular tube merges with the volute, that is the tongue of the pump, an unsteady flow separation was observed. Regarding the Luer lock, this event happened only one time and it could be due to a contact between the thin threads of the yarn and the Luer lock nozzle. CentriMag pump 2cm yarn threads During the measurements all the clot analogues got trapped in the same way, so they started rotating in the house of the pump and between the blades and then they detached and went out from the outlet in small pieces: indeed, during the test the thread flaked off and stretched in the direction of the flow. From these measurements is possible to underline that the length of the clot analogue is a parameter that must be taken into account when analysing the data, as it is decisive for the number of cases in which the clots got trapped. When the threads are 2cm-long they are captured by the flow and the blades easier and it is more difficult for the clot analogues to go outside the flow and through all the circuit downstream the pump. Rotaflow pump 1cm yarn threads During the measurements, the clot analogues got trapped only into the tongue of the pump, that is the edge between the pump housing and the outlet of the pump. In these measurements, the trapped clots cases account for around the 43% of the total valid measurement cases. This percentage is higher than the one related to the CentriMag pump with 1cm-long threads, meaning that the Rotaflow pump seemed to be more likely to promote trapped clots events. This can be explained by the different design of the two pumps: the Rotaflow pump is flatter and the outlet tube is longer resulting in a more angular tongue that can easily catch the threads flowing inside the pump; moreover, in this pump the blades are not exposed as in the CentriMag, so after every measurement it was necessary to stop the pump and take it out from the driving unit in order to check if some threads or pieces of them got stuck in some areas not visible from the driving unit. Rotaflow pump 2cm yarn threads During the measurements all the clot analogues got trapped in the same way, so they started rotating in the house of the pump and between the blades and then they have to be pulled out from the pump manually: indeed, during the test the thread flaked off and stretched in the direction of the flow and it is possible for them to spontaneously go out from the outlet of the pump. From these measurements is possible to underline, again, that the length of the clot analogue is a parameter that must be taken into account when analysing the data, as it is decisive for the number of cases in which the clots got trapped. When the threads are 2cm-long they are captured by the flow and the blades easier and it is more difficult for the clot analogues to go outside the flow and through all the circuit downstream the pump. In this specific case the design of the Rotaflow pump is crucial because the threads specifically got entangled in a small edge on the bottom of the magnet that connects the blades to the covering structure of the blades. For this reason, after every measurement it was necessary to stop the pump and take it put from the driving unit in order to manually remove the threads using a small metal stick or cotton swab. Furthermore, after every cleaning operation it was important to eliminate air bubbles created due to the disconnection of the pump inlet tube. After this first set of results, it was important to see if it were possible to apply the method of Fuchs et al. (2018) also in this experimental circuit: to understand this, graphs, that compare the power spectral density of the spectra of the thrombotic recordings with the spectrum of the clean circuit, were elaborated; if the method works and if the clot analogues are suitable, then I would have seen an amplification of the peaks in the low frequencies. An example of these types of graphs is shown in figure: Figure 4 - The logarithm of the amplitude of the signal with respect to the frequency rage [Hz]. The graph refers to the signal acquired in the circuit with the CentriMag pump, in point#1 and with a rotational speed of 2200 rpm. In particular, there is a comparison between the recordings during the injections of clots 1cm long and the clean spectrum. The y-axis is normalized with respect to the amplitude of the peak related to the frequency associated to the rotational speed. From the graph is possible to see that there is an actual amplification of the 4 Hz peak when a clot is inserted inside the circuit, even if this amplification is more evident in clinical data so using real patients, blood and real clots. Moreover, due to the normalization is possible to see that on the y-axis the value 0 corresponds to the peak at the frequency related to the rotational speed of the pump. The maximum amplification, evaluated between the thrombotic recordings and the clean recording, reached in the four groups analysed was the following: 2 orders of magnitude for CentriMag pump 1cm yarn threads; 7 orders of magnitude for CentriMag pump 2cm yarn threads; 5 orders of magnitude for Rotaflow pump 1cm yarn threads; 7 orders of magnitude for Rotaflow pump 2cm yarn threads. As explained before, a Matlab algorithm was used to extract some specific features of the signals recorded. To understand which features were more correlated to the transition from non-thrombotic to thrombotic recordings, each feature of these two groups of recordings were directly compared using histograms. Furthermore, it is necessary to underline that the difference between two groups, such as an experiment vs. control group, is judged to be statistically significant when p=0.05 or less and the p-value is exactly the value calculated by the Excel function. In this phase, five peculiar recordings for each pump were chosen to be compared: considering the CentriMag pump, the signal’s features correlated the most were zero crossing rate, energy of the signal, entropy of the signal and spectral flux; considering the Rotaflow pump, the signal’s features correlated the most were entropy, spread of a frame, total harmonic distortion, total harmonic distortion of the even harmonics and total harmonic distortion of the odd harmonics. Discussion and conclusions The results obtained provide an indication on the preferential localization of the clots, different depending on the type of pump used in the circuit. Moreover, regarding the 1cm-long yarn threads, the percentage of clots trapped into the circuit with the Rotaflow pump is five times higher than the one related to the circuit with the CentriMag pump. Concerning this aspect of the results, surely the design of the two centrifugal pumps is crucial. Indeed, in the Rotaflow pump the threads specifically got entangled in a small edge on the bottom of the magnet that connects the blades to the covering structure of the blades. The second type of results are related to the evaluation of the amplification of the low frequency peaks: the detection method is working with both pumps because it is possible to see an amplification of the peaks in the range 2-6Hz. This amplification is more evident for longer threads and also in the circuit with the Rotaflow pump. The last type of results are the ones related to the post processing algorithm used to underline which features of the signals are actually changing if there is a clot in the circuit and how much statistical significance has these changes. These feature extraction and evaluation is important in a future work to better analyse the differences, in terms of signal recorded, between thrombotic and non-thrombotic recordings. All of this can give more significance to the clot detection method improved in this project. The peculiar differences into these specific features can allow to evaluate the presence or absence of clots in the circuit, analysing the spectra of the recorded signals. Obviously, the present study is affected by various limitations but in conclusions, thanks to the results obtained in the present study, it was possible to validate and help the optimisation of a novel methodology to detect blood clots in ECMO machinery, in particular it was possible to obtain information about the preferential positioning of the clots within the ECMO circuit, through a thousand tests in the laboratory that provided statistical value to the results and sensitivity of the methodology. Furthermore, the results given by the post processing and related to the features extracted gave new information about the characteristics of the thrombotic signals in the whole spectrum and in the low frequency range and have provided new insights for the development of this research area.
Introduzione I dispositivi medici a contatto con il sangue, come stent, innesti, cateteri, circuiti extracorporei e dispositivi di assistenza ventricolare, sono usati per trattare una varietà di malattie cardiovascolari e cardiopolmonari. L'utilizzo di macchine artificiali a supporto delle funzioni degli organi ha un grande vantaggio legato alla loro disponibilità pressoché illimitata, contrariamente a quella limitata degli organi trapiantati. D'altra parte, gli organi artificiali per il supporto circolatorio sono associati a complicanze tromboemboliche, pro-infiammatorie ed emolitiche. Le complicanze derivanti dalla formazione di trombi in questi dispositivi sono, infatti, una frequente fonte di fallimento della terapia e la trombosi e l'emorragia sono i due problemi clinici dominanti per i pazienti connessi ad un supporto circolatorio meccanico. Le suddette complicazioni non solo ostacolano il funzionamento del dispositivo ma rappresentano un rischio diretto per il paziente. Al fine di migliorare l'uso dei dispositivi extracorporei ed evitare queste complicazioni, negli ultimi decenni sono stati condotti molti studi sulla rilevazione dei coaguli e sulla formazione di trombi su diversi componenti delle bio-macchine. Un dispositivo extracorporeo utilizzato in alcune situazioni mediche gravi, quando la funzione di organi critici può essere temporaneamente persa o insufficiente per mantenere il paziente in vita, è l'ossigenazione a membrana extracorporea o ECMO, quindi una terapia salvavita per il supporto temporaneo in caso di insufficienza polmonare e/o insufficienza cardiaca. Il circuito ECMO è composto da una pompa, normalmente centrifuga, che sostituisce la funzione di pompaggio del sangue del cuore, un ossigenatore a membrana utilizzato come polmone artificiale, uno scambiatore di calore per riscaldare il sangue prima di reinfonderlo, cannule impiantate nei vasi sanguigni principali del paziente e tubi generalmente realizzati in composto plastico a base di polivinilcloruro (PVC) insieme a più connettori per i tubi stessi. Figure 5 - Esempio di un circuito ECMO. Fonte https://www.quora.com/Can-ECMO-be-used-to-replace-breathing-to-buy-time-for-rescue-if-someone-is-trapped-under-water È comunemente riconosciuto che la fonte più importante di potenziale trombogenicità nel circuito ECMO è di origine meccanica, la cui dà inizio a un circolo vizioso biochimico. La pompa centrifuga è generalmente considerata la parte più trombogenica del circuito. Il rilevamento di coaguli nell'ECMO è un problema che coinvolge più volte approcci invasivi e attualmente non ci sono metodi per rilevare i coaguli prima che possano effettivamente diventare visibili, grandi e problematici. In clinica non è facile rilevare i coaguli dall'esterno e può verificarsi la formazione di trombosi all’interno della pompa anche se il paziente ha apparentemente un'adeguata terapia anticoagulante. Il problema principale dal punto di vista ingegneristico è legato al fatto che la formazione di trombi può facilmente compromettere la funzionalità del macchinario, quindi quando si forma un coagulo è necessario sostituire le parti del circuito coinvolte per risolvere due problemi: prima di tutto il coagulo potrebbe embolizzare ed entrare nel sistema circolatorio del paziente, quindi deve essere rimosso; in secondo luogo, il coagulo potrebbe ridurre l'efficienza del macchinario interferendo con la fluidodinamica del sangue e il funzionamento di parti del circuito e deve essere rimosso anche per salvaguardare il circuito stesso. Lo scopo del presente lavoro è migliorare un metodo non invasivo per il rilevamento di coaguli mediante l’analisi del range degli infrasuoni in un circuito ECMO in vitro e, inoltre, localizzare la trombosi nei diversi componenti della bio-macchina utilizzando sostitutici naturali o sintetici di coaguli. Questo progetto sperimentale consiste nel miglioramento di un lavoro di ricerca svolto in precedenza, in cui è stata trovata una correlazione tra la presenza di coaguli nel circuito ECMO e l'amplificazione del segnale registrato nelle basse frequenze: i segnali registrati provengono dalla vibrazione delle pareti dei tubi del circuito durante il flusso di liquido al suo interno, con o senza coaguli nel circuito stesso. Il set-up sperimentale aveva lo scopo di emulare l’interazione tra il normale funzionamento dell’ECMO e la presenza di particelle che vanno a mimare i coaguli. Stato dell’arte L'idea di utilizzare l’acustica per il rilevamento dei coaguli nei macchinari dei sistemi di supporto alla vita, al fine di aiutare la decisione clinica dei medici rispetto ad un paziente non è affatto nuova. Tali mezzi sono stati proposti da diversi autori negli ultimi decenni e ogni approccio è leggermente diverso dall'altro in termini di tipo di segnali analizzati, di modalità di esecuzione di tale analisi e tipo di caratteristiche dei segnali prese in considerazione: la cosa in comune tra tutte le ricerche è il fatto che i dati acustici sono principalmente relativi alla pompa utilizzata nel circuito. Lo scopo principale del presente studio è la validazione in vitro della metodologia proposta e l'analisi della sua senstività utilizzando anche una pompa aggiuntiva rispetto a quella utilizzata da Fuchs et al. (2018); inoltre, il perfezionamento della metodologia al fine di focalizzarsi anche sulla localizzazione degli eventi trombotici, non solo sulla rilevazione degli stessi. Il lavoro svolto da Fuchs et al. (2018) è stata la principale fonte di informazioni per il presente studio. Questo gruppo di ricerca ha elaborato un nuovo metodo non invasivo per il rilevamento di coaguli nel circuito ECMO: in particolare, è stato considerato il flusso in un circuito ECMO utilizzando una pompa centrifuga. L'approccio acustico per rilevare i trombi è stato sviluppato e applicato ai dati clinici raccolti presso il centro ECMO del Karolinska University Hospital e questo metodo si basa sul fatto che i coaguli di sangue possono essere identificati considerando la fascia iniziale dello spettro di frequenza (basse frequenze), nella gamma degli infrasuoni, quindi con frequenza inferiore a 20 Hz. La rilevazione dei coaguli è stata studiata sperimentalmente, sia attraverso dati provenienti da pazienti reali, che in un ambiente di simulazione in laboratorio. Tutti i risultati si basano su dati raccolti utilizzando un idrofono (modello 8103, Bruel & Kjaer) e poi post elaborati tramite Matlab. Il circuito è stato considerato trombotico se una significativa componente a bassa frequenza, al di sopra di un certo livello di soglia, era presente nello spettro: tale soglia è stata impostata a circa due ordini di grandezza maggiore della grandezza del picco relativo alla frequenza dominante senza coaguli, molto spesso correlato alla frequenza di rotazione della pompa. È possibile analizzare un caso, in cui la presenza del coagulo era abbastanza chiara, nella figura seguente: Figure 6 - Paziente neonato nel circuito ECMO (Paziente 3). Le immagini sono relative ai segnali di nell’ingresso alla pompa. L’immagine di sinistra e destra sono relative allo stesso paziente ma quella di destra è stato registrata 5 giorni dopo quella sinistra. La pompa funzionava a 2200 giri / min (36,67 Hz) e 0,33 LPM alla prima registrazione e a 2100 (35 Hz) e 0,2 LPM nella registrazione successiva. Le frecce indicano la velocità di rotazione della pompa utilizzata per la normalizzazione dello spettro. Notare il segnale forte a circa 2 Hz, che come nella registrazione precedente è due ordini di grandezza più forte della corrispondente frequenza di rotazione della pompa. Nei grafici, la soglia utilizzata come criterio spartiacque è soddisfatta con un margine ampio e quindi queste registrazioni indicano una chiara presenza di un coagulo. Notare anche il forte segnale a 50 Hz (frequenza della linea di alimentazione). Fonte https://www.diva-portal.org/smash/get/diva2:1120190/FULLTEXT01.pdf Nella registrazione effettuata nel circuito di questo paziente, un picco chiaro è evidente a circa 2 Hz e questo picco è maggiore di oltre un ordine di grandezza rispetto al picco legato alla velocità di rotazione della pompa, indicato con una freccia rossa. Materiali e Metodi In questo lavoro di tesi è stato necessario costruire un circuito sperimentale per ricreare un circuito ECMO. I componenti principali erano tubi, pompe centrifughe da scambiare durante le diverse misurazioni per misurare il segnale registrato sia dalla pompa centrigufa CentriMag che da quella Rotaflow, un ossigenatore e un serbatoio d'acqua in plastica. Sul circuito sono stati contrassegnati con un pennarello indelebile dieci punti lungo la superficie dei tubi, uno in ogni tratto a monte e a valle della pompa. Questi punti sono stati utilizzati come punti di misurazione nelle fasi successive per essere sicura di avere posizioni di misurazione più o meno riproducibili durante l'intera analisi. Figure 7 - Il circuito ECMO sperimentale costruito nel laboratorio del KTH La pompa CentriMag è la pompa che era già stata studiata nel precedente lavoro di Fuchs et al. (2018): studiando nuovamente questa pompa ho voluto valutare la ripetibilità del metodo non invasivo di rilevamento dei coaguli e la sua sensitività e anche esaminare gli spettri in più dettaglio per evidenziare la localizzazione dei coaguli. Questo metodo, invece, non è mai stato testato sulla pompa Rotaflow, o in generale su pompe diverse dalla CentriMag, quindi ho voluto valutare la risposta di questa pompa al suddetto metodo. Una volta assemblato il circuito, si è deciso che le misurazioni sarebbero state fatte nei dieci punti di misura precedentemente contrassegnati, e che sarebbero state valutate principalmente due velocità di rotazione della pompa centrifuga: 2200rpm e 3200rpm. Per acquisire i segnali acustici è stato utilizzato un idrofono collegato ad un amplificatore e poi ad un computer: il segnale audio è stato registrato utilizzando l’applicativo Audacity e poi esportato come file .wav in Matlab; qui il segnale è stato convertito nel dominio della frequenza e l'analisi è stata effettuata in base alla densità spettrale di potenza di ciascun segnale. Innanzitutto era necessario caratterizzare il circuito “pulito”, ovvero senza coaguli all’interno, e acquisire così il segnale audio semplicemente facendo scorrere acqua nel circuito stesso: le registrazioni sono state eseguite in ciascuno dei dieci punti contrassegnati e con le due diverse velocità di rotazione della pompa. Questa caratterizzazione è stata eseguita sia per il circuito con la pompa Rotaflow che per il circuito con la pompa CentriMag. Successivamente, sono iniziate le misurazioni nel circuito con all’interno i sostituti dei coaguli. Durante i lavori sperimentali non è sempre possibile avere accesso a coaguli reali animali o umani, a causa di problemi etici e della necessità di approvazioni burocratiche: per questo motivo c'è la necessità di imitare i coaguli di sangue utilizzando sostitutici dei coaguli. L'aspetto importante nella scelta di tali sostituti è legato al fatto che i materiali scelti devono interagire con il flusso dell'acqua e non solo essere una presenza passiva nel circuito: in particolare questi sostituti devono assorbire dell’acqua, come farebbero I coaguli reali, e reagire alle sollecitazioni di taglio esercitate dal flusso di liquido. Si è scelto di utilizzare fili di cotone per imitare principalmente i coaguli a base di fibrina e quei coaguli che normalmente si estendono in lunghezza a causa della rotazione della pompa; altri materiali testati sono stati silicone, slime e plastilina per cercare di imitare i coaguli a base di globuli rossi con forma, dimensione e posizione diverse. Quindi, per questa parte del progetto si è deciso di utilizzare due idrofoni nel circuito ECMO, in modo da registrare simultaneamente il segnale in diversi punti del circuito. Al fine di post processare nel dettaglio il set di dati, è stato utilizzato un algoritmo, fornito dal gruppo di ricerca guidato da Beren Semiz presso il Georgia Institute of Technology e modificato da me ad hoc per questa ricerca. In particolare, questo algoritmo viene utilizzato per estrarre caratteristiche del segnale che possono evidenziare differenze distinguibili tra un segnale non trombotico e uno trombotico. Le caratteristiche estratte sono state memorizzate in una matrice, dove ogni riga rappresenta un singolo pezzo del segnale e le colonne rappresentano le 61 caratteristiche estratte. I valori di queste caratteristiche sono stati confrontati tra il segnale non trombotico e uno trombotico al fine di trovare una correlazione tra la variazione del valore di alcune di queste caratteristiche e la presenza di coaguli nel circuito. Tutto ciò è stato fatto utilizzando Excel, in particolare effettuando un t-test sui due gruppi di registrazioni e valutando il relativo p-value. Risultati I risultati di questo lavoro devono essere suddivisi in diversi gruppi. In primo luogo, i risultati riguardanti la frequenza con cui i coaguli restano incastrati in qualche parte del circuito, le zone in cui ciò avviene e un primo confronto tra gli spettri di frequenza (densità spettrale di potenza) delle pompeCentriMag e Rotaflow. Successivamente vengono presentati i risultati della post elaborazione dei segnali analizzati grazie all’utilizzo di Excel. È necessario sottolineare che I risultati più importanti sono relative a quattro gruppi di registrazioni che vengono anche confrontati l’uno rispetto all’altro: questi gruppi sono composti dai dati della pompa CentriMag con fili di cotone da 1cm, dai dati della pompa CentriMag con fili di cotone da 2cm, dai dati della pompa Rotaflow con fili di cotone da 1cm e dai dati della pompa Rotaflow con fili di cotone da 2cm. Pompa CentriMag fili da 1cm Durante le misurazioni, i fili di cotone sono rimasti intrappolati principalmente in tre diverse posizioni all'interno del circuito: la linguetta della pompa, cioè il bordo tra il corpo della pompa e l'uscita della pompa; il foro dell'alloggiamento del magnete nella pompa, cioè il foro presente nella struttura megnetica delle pale; il Luer lock di iniezione, cioè il Luer lock utilizzato nel circuito per iniettare i fili di cotone. In questo primo gruppo di registrazioni i casi di coaguli rimasti intrappolati rappresentano circa il 9% del totale dei casi di misurazione validi. La posizione preferenziale è il foro nel magnete della pompa, che rappresenta il 55% dei casi di coaguli intrappolati totali, seguito dalla linguetta della pompa che rappresenta il 41% ed infine il Luer lock di iniezione che rappresenta il 5%. Queste posizioni non sono eventi casuali, anzi questa localizzazione di eventi trombotici è confermata anche da simulazioni computazionali: ad esempio, Fuchs et al. (2018) ha identificato il foro della struttura magnetica e la linguetta della pompa come punti della pompa in cui le piastrine possono attivare la coagulazione a causa dell'elevato sforzo di taglio e dell’elevato tempo di permanenza in questa zona; inoltre, all'uscita della pompa, dove il tubo circolare di uscita si fonde con la voluta, cioè la linguetta della pompa, si è osservata una separazione di flusso instabile. Per quanto riguarda il Luer lock, questo evento è avvenuto una sola volta e potrebbe essere dovuto ad un contatto tra i fili sottili di cotone e l'ugello Luer lock. Pompa CentriMag fili da 2cm Nelle misurazioni, tutti i sostituti del coagulo sono rimasti completamente intrappolati allo stesso modo, quindi hanno iniziato a ruotare all'interno della pompa e tra le lame per poi staccarsi ed uscire dall'uscita in piccoli pezzi: infatti durante la prova il filo si è sfaldato e allungato nella direzione del flusso.Da queste misurazioni è possibilesottolineare che la lunghezza dell'analogo del coagulo è sicuramente un parametro di cui tenere conto nell'analisi dei dati, in quanto determinante per il numero di casi in cui i coaguli sono rimasti intrappolati. Pompa Rotaflow fili da 1cm Nelle misurazioni, i sostituti dei coaguli sono rimasti completamente intrappolati solo nella linguetta della pompa, ovvero il bordo tra il corpo della pompa e lauscita della pompa. In queste misurazioni, i casi di coaguli intrappolati rappresentano circa il 43% del totale dei casi di misurazione validi. Questa percentuale è superiore a quella relativa alla pompa CentriMag con fili di lunghezza 1cm, il che significa che la pompa Rotaflow sembra avere maggiori probabilità di promuovere l’intrappolamento di coaguli. Una spiegazione di questa differenza potrebbe venire dal diverso design delle due pompe: la pompa Rotaflow è più piatta e il tubo di uscita è più lungo, quindi si crea una una linguetta più spigolosa che può facilmente catturare fili che scorrono all'interno dellapompa; un’altra osservazione importante è legata al fatto che in questa pompa le lame non sono esposte come nella CentriMag, quindi dopo ogni misura è stato necessariofermare la pompa ed estrarla dal motore per riuscire a verificare se qualche filetto o pezzi di essi fossero incastrati in alcune zone non visibili con la pompa in azione. Pompa Rotaflow fili da 2cm Durante le misurazioni tutti i sostituti dei coaguli sono rimasti intrappolati allo stesso modo, quindi hanno iniziato a ruotare all'interno della pompa e tra le lame e quindi è stato necessario estrarli manualmente dalla pompa: infatti durante la prova il filo si è sfaldato e allungato nella direzione del flusso ed è impossibile che fuoriescano spontaneamente dall'uscita della pompa. Da queste misurazioni si può sottolineare, ancora, che la lunghezza di questi fili è un parametro di cui tenere conto nell'analisi dei dati, in quanto determinante per il numero di casi in cui i coaguli sono rimasti intrappolati. In questo caso specifico il design della pompa Rotaflow è cruciale perché i fili si sono specificamente impigliati in un piccolo bordo sul fondo del magnete che collega le pale alla struttura di copertura delle pale. Per questo motivo dopo ogni misura è stato necessario fermare la pompa ed estrarla dall'unità motrice per poter rimuovere manualmente i fili utilizzando un bastoncino di metallo o un cotton fioc. Inoltre, dopo ogni operazione di pulizia, era importante eliminare le bolle d'aria create a causa dello scollegamento del tubo di ingresso della pompa. Dopo questa prima serie di risultati, era importante vedere se fosse possibile applicare il metodo di Fuchs et al. (2018) anche in questo circuito sperimentale: per comprenderlo, sono stati elaborati dei grafici che confrontano la densità spettrale di potenza degli spettri delle registrazioni trombotiche con lo spettro del circuito pulito; se il metodo funziona e se i sostituti dei coaguli sono adatti, allora si sarebbe vista una amplificazione dei picchi nelle basse frequenze. Un esempio di questi tipi di grafici è mostrato nella figura sottostante: Figure 8 - Il logaritmo dell'ampiezza del segnale rispetto alla banda di frequenza [Hz]. Il grafico si riferisce al segnale acquisito nel circuito con la pompa CentriMag, nel punto#1 e con una velocità di rotazione di 2200rpm. In particolare, c'è un confronto tra le registrazioni durante le iniezioni di coaguli lunghi 1 cm e lo spettro del segnale del circuito pulito. L'asse y è normalizzato rispetto all'ampiezza del picco relativo alla frequenza associata alla velocità di rotazione Dal grafico è possibile osservare che c'è un'effettiva amplificazione del picco di 4 Hz nel momento in cui viene inserito un coagulo all'interno del circuito, anche se questa amplificazione è più evidente nei dati clinici quindi utilizzando pazienti reali, sangue e coaguli reali. Inoltre, grazie alla normalizzazione è possibile vedere che sull'asse y il valore 0 corrisponde al picco della frequenza relativa alla velocità di rotazione della pompa. L'amplificazione massima, valutata tra le registrazioni trombotiche e la registrazione del circuito pulito, raggiunta nei quattro gruppi analizzati è stata la seguente: 2 ordini di grandezza in più per fili di cotone da 1cm nella pompa CentriMag; 7 ordini di grandezza in più per i fili di cotone da 2 cm nella pompa CentriMag; 5 ordini di grandezza per i fili di cotone da 1 cm nella pompa Rotaflow; 7 ordini di grandezza per fili di cotone da 2 cm nella pompa Rotaflow. Come spiegato precedentemente, è stato utilizzato un algoritmo in Matlab per estrarre alcune caratteristiche specifiche dei segnali registrati. Per capire quali caratteristiche fossero più correlate alla transizione da registrazioni non trombotiche a registrazioni trombotiche, ciascuna caratteristica di questi due gruppi di registrazioni è stata confrontata direttamente utilizzando istogrammi. Inoltre, è necessario sottolineare che la differenza tra due gruppi, in particolare un gruppo di dati provenienti da misurazioni sperimentali rispetto ad un gruppo di controllo, è giudicata statisticamente significativa quando p-value = 0.05 o inferiore e il p-value è esattamente il valore calcolato dalla funzione Excel. In questa fase si è deciso di confrontare cinque registrazioni peculiari per ciascuna pompa: considerando la pompa CentriMag, le caratteristiche estratte del segnale maggiormente correlate alla variazione circuito pulito – circuito trombotico erano lo “zero crossing rate”, l’energia del segnale, l’entropia del segnale e il flusso spettrale; considerando la pompa Rotaflow, invece, tali caratteristiche erano l’entropia, lo “spread” del segnale, la distorsione armonica totale, la distorsione armonica delle armoniche pari e la distorsione armonica delle armoniche dispari. Discussione e conclusioni I risultati ottenuti forniscono un'indicazione sulla localizzazione preferenziale dei coaguli, diversa a seconda del tipo di pompa utilizzata nel circuito. Inoltre, per quanto riguarda i fili da 1 cm, la percentuale di coaguli rimasti intrappolati nel circuito con la pompa Rotaflow è cinque volte superiore a quella relativa al circuito con la pompa CentriMag. Per quanto riguarda questo aspetto dei risultati, sicuramente il design delle due pompe centrifughe è fondamentale. Nella pompa Rotaflow, infatti, i fili rimagono impigliati in modo specifico in un piccolo bordo sul fondo del magnete che collega le pale alla struttura di copertura delle pale. Il secondo tipo di risultati è relativo alla valutazione dell'amplificazione dei picchi di bassa frequenza: il metodo di rilevamento funziona con entrambe le pompe perché è possibile vedere un'amplificazione dei picchi nel range 2-6Hz. Questa amplificazione è più evidente utilizzando sostituti di coaguli più lunghi e anche nel circuito con la pompa Rotaflow. L'ultimo tipo di risultati sono quelli relativi all'algoritmo di post-elaborazione utilizzato per sottolineare quali caratteristiche dei segnali tendono effettivamente a cambiare il proprio valore se c'è un coagulo nel circuito e quanta significatività statistica hanno questi cambiamenti. L'estrazione e la valutazione di queste caratteristiche è importante in un lavoro futuro per analizzare meglio le differenze, in termini di segnale registrato, tra registrazioni trombotiche e non trombotiche. Tutto ciò può dare più significato al metodo di rilevamento dei coaguli migliorato in questo progetto di tesi. Le peculiari differenze in queste specifiche caratteristiche possono consentire di valutare la presenza o l'assenza di coaguli nel circuito, analizzando gli spettri dei segnali registrati. Ovviamente, il presente studio è affetto da varie limitazioni ma in conclusione, grazie ai risultati ottenuti nel presente studio, è stato possibile validare e ottimizzare questa una nuova metodologia di rilevazione dei coaguli di sangue nelle macchine ECMO; in particolare è stato possibile ottenere informazioni sul posizionamento preferenziale dei coaguli all'interno del circuito ECMO, attraverso innumerevoli test in laboratorio che hanno fornito valore statistico ai risultati e sensitività della metodologia. Inoltre, I risultati forniti dalla post elaborazione e relative alle caratteristiche estratte dai segnali, hanno fornito nuove informazioni sulle caratteristiche acustiche dei segnali trombotici nel range delle basse frequenze e hanno fornito nuovi spunti per lo sviluppo di quest’area di ricerca.
Improvement of a non-invasive method for clot detection in ECMO circuit and localization of thrombotic events
Cairelli, Adriana Gaia
2019/2020
Abstract
Introduction Blood-contacting medical devices, such as stents, grafts, catheters, extracorporeal circuits, and ventricular assist devices, are used to treat a variety of cardiovascular and cardiopulmonary diseases. The use of artificial machines to support organ functions has a great advantage related to the almost unlimited availability, in contrast with the limited one of real organs transplanted. On the other hand, artificial organs for circulatory support are associated with thromboembolic, pro-inflammatory and haemolytic complications. Complications resulting from thrombus formation in these devices are, indeed, a frequent source of failure, and thrombosis and haemorrhage are the two dominant clinical issues for mechanical circulatory support patients. The aforementioned complications not only impede the function of the device but pose direct risk to the patient. In order to improve the use of extracorporeal devices and avoid these complications, a lot of studies on clot detection and thrombus formation on different components of bio-machines were performed during the last decades. An extracorporeal device used in certain severe medical situations, when the function of critical organs may temporarily be lost or insufficient to maintain life, is the Extracorporeal Membrane Oxygenation or ECMO, so a life-saving therapy for temporary support in refractory severe lung and/or heart failure. The ECMO circuit is composed of a blood pump, normally centrifugal, that replace the blood pumping function of the heart, a membrane oxygenator used as artificial lungs, a heat exchanger to warm the blood before re-infusing it, cannulas implanted into the patient’s major vessels, and tubing generally made of polyvinylchloride-based plastic compound (PVC) along with multiple tubing connectors. Figure 1 - Example of an ECMO circuit. Source https://www.quora.com/Can-ECMO-be-used-to-replace-breathing-to-buy-time-for-rescue-if-someone-is-trapped-under-water It is commonly recognised that the most important source of enhanced thrombogenicityin the ECMO circuit is of mechanical origin, initiating a vicious biochemical circle. The centrifugal pump is generally considered the most thrombogenic part of the circuit. Clot detection in ECMO is a problem that involves invasive approaches several times and there are currently no methods to detect them prior they can actually become visible, big and problematic. In clinics is not easy to detect clots from the outside and pump thrombosis formation can occur even if the patient apparently has adequate anticoagulation. The main problem from the engineering point of view is related to the fact that a thrombus formation can easily compromise the functionality of the machinery, so when a clot is formed it’s necessary to substitute the thrombotic parts of the circuit in order to solve two problems: first of all the clot can embolize and go inside the circulatory system of the patient, so it has to be removed; secondly, the clot can reduce the efficiency of the machinery interfering with the fluid dynamics of the blood and the functioning of parts of the circuit and has to be removed to save the circuit too. The aim of this work is to improve a non-invasive method for clot detection by infrasound in an ECMO in-vitro circuit and to localise thrombosis in the different components of the bio-machine using clot analogues. This experimental project is an improvement of a previous work, in which a correlation between the presence of clots in the ECMO circuit and the amplification of the signal recorded by Infrasounds in low frequencies was found: the signals recorded come from the vibration of the piping’s walls during the flow of liquid inside the circuit, with or without clots inside the circuit itself. The experimental set-up had the purpose to emulate the interaction between the normal functioning of the ECMO and the presence of particles mimicking the clots. State of the art The idea of using acoustic data for the clot detection into life support systems’ machinery in order to aid clinical decision-making is by no means new. Such means have been proposed by several authors during the last decades and every approach is slightly different from the other in terms of which signals are analysed, how the analysis is performed and which signals’ features are taken into account: the thing in common between all the works is the fact that the acoustic data are mainly related to the pump used into the circuit. The main purpose of the present study is the in vitro validation of the proposed methodology and analysis of its sensitivity using also an additional pump with respect to that used by Fuchs et al. (2018); moreover, the improvement of the method in order to focus also on the localisation of the thrombotic events, not only on the detection of the clots themselves. The work performed by Fuchs et al. (2018) has been the main source of information for the present study. Fuchs et al. (2018) elaborated a new non-invasive method for clot detection in the ECMO circuit: specifically, the flow in an ECMO circuit using a centrifugal pump has been considered. The acoustic approach to detect thrombi has been developed and applied to clinical data gathered at the ECMO centre at Karolinska University Hospital and this method is based on the concept that the presence of blood clot can be identified by considering the low-end of the frequency spectrum, within the range of infrasound, so below 20 Hz. Clot detection has been studied experimentally, both at “bedside” and in a laboratory environment. All the results are based on data gathered by a hydrophone (model 8103, Bruel & Kjaer) and then post processed through Matlab. The circuit was determined to contain a clot if a significant low-frequency component, above a threshold level, was present in the spectrum: this threshold was set to about two orders of magnitude larger than the magnitude of the peak related to the dominating frequency without a clot, most often related to the pump rotation frequency. It is possible to analyse a case in which the presence of a clot was clear enough in the following figure: Figure 2 - Infant patient in the ECMO circuit (Patient 3). The frames are related to the inflow signals.The left and right frame are related to the same patient but the right frame was registered 5 days after the left one. The pump was runningat 2200 rpm (36.67 Hz) and 0.33 LPM at the first registration and at 2100 (35 Hz) and 0.2 LPM in the later registration. The arrows mark the pump rotation rate used for normalization of the spectrum. Note the strong signal at about 2 Hz, which as in the earlier registration is two orders of magnitude stronger than the corresponding pump rotation frequency. In all frames, the clot threshold criterion ismet with a widemargin and hence these registrations indicate a clear presence of a clot. Note also the strong signal at 50 Hz (power line frequency). Source https://www.diva-portal.org/smash/get/diva2:1120190/FULLTEXT01.pdf In this patient’s recording, a clear peak is evident at approximately 2 Hz and this peak is larger by more than an order of magnitude as compared to the peak due to the pump rotation rate indicated with the red arrow. The results indicate that a blood clot was present already during the early recording and it was then confirmed in clinics. Materials and Methods In the present study it was necessary to build up an experimental circuit in order to recreate an ECMO circuit. The main components were tubing, centrifugal pumps to be exchanged during the different measurements to measure signal recorded both from the CentriMag centrigufal pump and the Rotaflow one, an oxygenator and a plastic water tank. On the circuit ten points along the piping were marked with permanent marker, one in each section upstream and downstream of the pump. These points were used as measurements points in the next steps to be sure to have more or less reproducible measurement positions during the whole analysis. Figure 3 – The experimental ECMO circuit built up in the LTH laboratory The CentriMag pump is the pump that was already studied in the previous work of Fuchs et al. (2018): by studying again this pump I wanted to evaluate the repeatability of the non-invasive clot detection method and its sensitivity and also to examine the spectra in more detail to highlight the localization of the clots. This method, instead, was never tested on the Rotaflow pump, or in general in pumps different from the CentriMag, therefore I wanted to evaluate the response of this pump to the method. Once the circuit was assembled, it was decided that the measurements would be done at the measuring points, previously marked, and that mainly two rotational speeds of the centrifugal pump would be evaluated: 2200rpm and 3200rpm. To acquire the signals, it was used a hydrophone connected to an amplifier and then to a laptop: the audio signal was recorded using Audacity and then exported as a .wav file into Matlab; here the signal was converted in the frequency domain and the analysis was done based on the power spectral density of each signal. First of all, it was necessary to characterise the clean circuit and so acquire the audio signal simply letting run water into the circuit: the recordings were performed in each of the ten marked points and with the two different rotational speeds of the pump. This characterisation was done both for the circuit with the Rotaflow pump and the circuit with the CentriMag pump. After that, the measurements in the circuit with the clot analogues started. During experimental works is not always possible to have access to real animal or human clots, due to ethical problems and the necessity for ethical approvals: for this reason, there is the need of mimic blood clots using clot analogues. The important aspect in the choice of clot analogues is related to the fact that the materials chosen have to interact with the flow of water and not only being a passive presence in the circuit: in particular these clots analogues have to absorb a bit of water and react to the shear stresses exerted by the flow. It was chosen to use cotton yarn threads to mainly mimic fibrin-based clots and those clots that normally extends due to the rotation of the pump; other tested materials were silicone, slime and clay in order to try to mimic RBCs-based clots with different rounded shape, dimension and location. So, for this part of the project it was decided to use two hydrophones in the experimental ECMO circuit, in order to simultaneously record the signal in different points in the circuit. In order to deeply post process the dataset, an algorithm, given by the group led by Beren Semiz at the Georgia Institute of Technology and modified by me ad hoc for this research, was implied. In particular, this algorithm is used to extract features of the signal that can highlight distinguishable differences between the normal and thrombosis recordings. The extracted features were stored in a matrix, where each row represents a single frame and the columns represent the 61 features extracted. The values of these features were compared between the normal and thrombosis recordings in order to find a correlation among the change in value of some features and the presence of clots in the circuit. All of this was done using Excel applying a t-test on the two groups of recordings and evaluating the related the p-value. Results The results of this work need to be subdivided into different groups. Firstly, the results regarding how often the clots get stuck in some part of the circuit, the areas in which they get stuck and a first comparison between the frequency spectra (power spectral density) of the CentriMag and Rotaflow pumps. Subsequently, the results of the post processing of the signals further analysed thanks to the use of Excel are presented. It is necessary to underline that the most important results are related to four groups of recordings that are also compared one with respect to the other. These groups of recordings are the ones from the CentriMag pump with yarn clot analogues 1cm long, the ones from the CentriMag pump with yarn clot analogues 2cm long, the ones from the Rotaflow pump with yarn clot analogues 1cm long and the ones from the Rotaflow pump with yarn clot analogues 2cm long. CentriMag pump 1cm yarn threads During the measurements, the clot analogues got trapped mainly in three different positions inside the circuit: the tongue of the pump, that is the edge between the pump housing and the outlet of the pump; the hole of the magnet house in the pump, that is the hole present in the magnet structure of the blades; the Luer lock of injection, that is the Luer lock used in the circuit to inject the yarn threads. In this specific group of recordings the trapped clots cases account for around the 9% of the total valid measurement cases. The preferential position is the hole in the magnet of the pump, that represents the 55% of the total trapped cases, followed by the tongue of the pump that represents the 41% and lastly the Luer lock of injection representing the 5%. These positions are not casual events, indeed this localization of thrombotic events is also confirmed by computational simulations: for example, Fuchs et al. (2018) identified the hole of the magnet house and the tongue of the pump as places in the pump where platelets can activate coagulation due to high shear stress and residence time; moreover, at the outlet of the pump, where the outlet circular tube merges with the volute, that is the tongue of the pump, an unsteady flow separation was observed. Regarding the Luer lock, this event happened only one time and it could be due to a contact between the thin threads of the yarn and the Luer lock nozzle. CentriMag pump 2cm yarn threads During the measurements all the clot analogues got trapped in the same way, so they started rotating in the house of the pump and between the blades and then they detached and went out from the outlet in small pieces: indeed, during the test the thread flaked off and stretched in the direction of the flow. From these measurements is possible to underline that the length of the clot analogue is a parameter that must be taken into account when analysing the data, as it is decisive for the number of cases in which the clots got trapped. When the threads are 2cm-long they are captured by the flow and the blades easier and it is more difficult for the clot analogues to go outside the flow and through all the circuit downstream the pump. Rotaflow pump 1cm yarn threads During the measurements, the clot analogues got trapped only into the tongue of the pump, that is the edge between the pump housing and the outlet of the pump. In these measurements, the trapped clots cases account for around the 43% of the total valid measurement cases. This percentage is higher than the one related to the CentriMag pump with 1cm-long threads, meaning that the Rotaflow pump seemed to be more likely to promote trapped clots events. This can be explained by the different design of the two pumps: the Rotaflow pump is flatter and the outlet tube is longer resulting in a more angular tongue that can easily catch the threads flowing inside the pump; moreover, in this pump the blades are not exposed as in the CentriMag, so after every measurement it was necessary to stop the pump and take it out from the driving unit in order to check if some threads or pieces of them got stuck in some areas not visible from the driving unit. Rotaflow pump 2cm yarn threads During the measurements all the clot analogues got trapped in the same way, so they started rotating in the house of the pump and between the blades and then they have to be pulled out from the pump manually: indeed, during the test the thread flaked off and stretched in the direction of the flow and it is possible for them to spontaneously go out from the outlet of the pump. From these measurements is possible to underline, again, that the length of the clot analogue is a parameter that must be taken into account when analysing the data, as it is decisive for the number of cases in which the clots got trapped. When the threads are 2cm-long they are captured by the flow and the blades easier and it is more difficult for the clot analogues to go outside the flow and through all the circuit downstream the pump. In this specific case the design of the Rotaflow pump is crucial because the threads specifically got entangled in a small edge on the bottom of the magnet that connects the blades to the covering structure of the blades. For this reason, after every measurement it was necessary to stop the pump and take it put from the driving unit in order to manually remove the threads using a small metal stick or cotton swab. Furthermore, after every cleaning operation it was important to eliminate air bubbles created due to the disconnection of the pump inlet tube. After this first set of results, it was important to see if it were possible to apply the method of Fuchs et al. (2018) also in this experimental circuit: to understand this, graphs, that compare the power spectral density of the spectra of the thrombotic recordings with the spectrum of the clean circuit, were elaborated; if the method works and if the clot analogues are suitable, then I would have seen an amplification of the peaks in the low frequencies. An example of these types of graphs is shown in figure: Figure 4 - The logarithm of the amplitude of the signal with respect to the frequency rage [Hz]. The graph refers to the signal acquired in the circuit with the CentriMag pump, in point#1 and with a rotational speed of 2200 rpm. In particular, there is a comparison between the recordings during the injections of clots 1cm long and the clean spectrum. The y-axis is normalized with respect to the amplitude of the peak related to the frequency associated to the rotational speed. From the graph is possible to see that there is an actual amplification of the 4 Hz peak when a clot is inserted inside the circuit, even if this amplification is more evident in clinical data so using real patients, blood and real clots. Moreover, due to the normalization is possible to see that on the y-axis the value 0 corresponds to the peak at the frequency related to the rotational speed of the pump. The maximum amplification, evaluated between the thrombotic recordings and the clean recording, reached in the four groups analysed was the following: 2 orders of magnitude for CentriMag pump 1cm yarn threads; 7 orders of magnitude for CentriMag pump 2cm yarn threads; 5 orders of magnitude for Rotaflow pump 1cm yarn threads; 7 orders of magnitude for Rotaflow pump 2cm yarn threads. As explained before, a Matlab algorithm was used to extract some specific features of the signals recorded. To understand which features were more correlated to the transition from non-thrombotic to thrombotic recordings, each feature of these two groups of recordings were directly compared using histograms. Furthermore, it is necessary to underline that the difference between two groups, such as an experiment vs. control group, is judged to be statistically significant when p=0.05 or less and the p-value is exactly the value calculated by the Excel function. In this phase, five peculiar recordings for each pump were chosen to be compared: considering the CentriMag pump, the signal’s features correlated the most were zero crossing rate, energy of the signal, entropy of the signal and spectral flux; considering the Rotaflow pump, the signal’s features correlated the most were entropy, spread of a frame, total harmonic distortion, total harmonic distortion of the even harmonics and total harmonic distortion of the odd harmonics. Discussion and conclusions The results obtained provide an indication on the preferential localization of the clots, different depending on the type of pump used in the circuit. Moreover, regarding the 1cm-long yarn threads, the percentage of clots trapped into the circuit with the Rotaflow pump is five times higher than the one related to the circuit with the CentriMag pump. Concerning this aspect of the results, surely the design of the two centrifugal pumps is crucial. Indeed, in the Rotaflow pump the threads specifically got entangled in a small edge on the bottom of the magnet that connects the blades to the covering structure of the blades. The second type of results are related to the evaluation of the amplification of the low frequency peaks: the detection method is working with both pumps because it is possible to see an amplification of the peaks in the range 2-6Hz. This amplification is more evident for longer threads and also in the circuit with the Rotaflow pump. The last type of results are the ones related to the post processing algorithm used to underline which features of the signals are actually changing if there is a clot in the circuit and how much statistical significance has these changes. These feature extraction and evaluation is important in a future work to better analyse the differences, in terms of signal recorded, between thrombotic and non-thrombotic recordings. All of this can give more significance to the clot detection method improved in this project. The peculiar differences into these specific features can allow to evaluate the presence or absence of clots in the circuit, analysing the spectra of the recorded signals. Obviously, the present study is affected by various limitations but in conclusions, thanks to the results obtained in the present study, it was possible to validate and help the optimisation of a novel methodology to detect blood clots in ECMO machinery, in particular it was possible to obtain information about the preferential positioning of the clots within the ECMO circuit, through a thousand tests in the laboratory that provided statistical value to the results and sensitivity of the methodology. Furthermore, the results given by the post processing and related to the features extracted gave new information about the characteristics of the thrombotic signals in the whole spectrum and in the low frequency range and have provided new insights for the development of this research area.File | Dimensione | Formato | |
---|---|---|---|
AdrianaGaia_Cairelli_10538810_915574.pdf
non accessibile
Descrizione: Tesi Magistrale
Dimensione
32.27 MB
Formato
Adobe PDF
|
32.27 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/10589/166354