The study described in this master’s thesis is part of the three-year project "Comparative Effectiveness of Socket Casting Methods: Improving Form and Fit" funded by the U.S. Department of Defense (DoD) and was carried out at the INAIL Prosthesis Center in Vigorso di Budrio (BO). The aim of the project is to compare total surface bearing for subject with transfemoral and transtibial amputation realised with two different methods to capture the anatomy of the residual limb: a traditional manual one, and a one that uses a system called 'Aqua System'. Amputation involves the removal of the distal segments of a limb by surgery or trauma. The need for individuals with amputation to regain some or all of their independence has led to the need for continuous improvement of prostheses. The socket represents the core of a prosthesis: it constitutes the interface with the individual’s residual limb. Between residual limb and socket, there is an indissoluble relationship represented by a reciprocal exchange of forces and pressures. The socket has allowed optimal pressure distribution to ensure fit and function. For this reason, the first step in fabricating a socket involves capturing the residuum anatomy through a plaster cast that is properly modelled to guarantee proper pressure distribution. The cast can be obtained either through "manual" or "hydrostatic" techniques. In the first case, the negative cast is properly molded and modelled manually by a certified prosthetist. This technique is certainly the most widespread and is clearly influenced by the experience and knowledge of the prosthetist, through it is difficult to distribute pressure evenly around the residual limb. Therefore, new techniques have been developed that do not rely on manual plaster manipulation and are called "Hydrostatic pressure with patient in monopodal orthostatism". The technique uses a pressure cylinder into which the subject's stump is placed, which allows capturing of the anatomy of the residuum as well as the pressure situation in the negative cast under full weight-bearing load. The literature lacks comparative studies of the effectiveness, efficiency and consistency of these two techniques for both transtibial and transfemoral amputations. The project funded by the U.S. Department of Defense (DoD), in which this thesis is developed, was conducted precisely to fill these gaps. The project involves the collaboration of three centers: Northwestern University of Chicago (NU), Minnesota VA Health Care System (MVAHCS) and INAIL (National Institute for Insurance against Accidents at Work,) Prosthesis Center. In each center, 20 subjects with transtibial amputation and 10 with transfemoral amputation will be recruited for a total of 90 participants. This thesis aims to present the methodologies followed to analyse the plaster and socket scans made with the two different methods to capture the anatomy of the residual limb of the first 30 subjects enrolled at INAIL and NU. To do this, it was necessary to evaluate the accuracy of the scanner devices used. Finally, an analysis of the shapes of the scanned stumps was performed using Principal Component Analysis (PCA) and morphing with the aim of obtaining the average shapes and rectification rules applied by each leading prosthetist. MATERIALS AND METHODS Measurement and Analysis of Scanner Accuracy and Precision Since one of the aims of the study is to understand the consistency of the methods to capture the anatomy of the residual limb, it is necessary to be able to measure the shape of the stump during the phase of casting and the subsequent shapes made by the prosthetists to build the socket. Two three-dimensional scanners have been chosen to complete this objective, they are called "interior" and "exterior" scanner because they are able to scan the interior of a negative model and the external shape of a positive model, respectively. The Echo 3D laser scanner manufactured by Rodin4D was chosen to scan the interior, while the Einscan 2X Pro Plus scanner was chosen to scan the exterior. The ability to make accurate measurements in terms of shape and volume of the residual limb, with good consistency between measurements made by different technicians, is essential for the purposes of the study. Therefore, preliminary accuracy, repeatability and reproducibility analyses were performed on the scanners used, to confirm the quality of the data collected into the project. Two operators with different experience from each center scanned 4 types of sockets with the Einscan Pro 2X Plus scanner. These sockets were also scanned by an operator external to the centers with the Einscan FreeScan 5X Laser metrological scanner which has an accuracy of 0,02 mm and a precision of 0,03 mm. Accuracy analyses were conducted by comparing the scans obtained at the centers to those obtained with the metrological scanner. Repeatability and reproducibility analyses were conducted by comparing the scans obtained by the two operators in the centers. Distance maps were generated and two parameters, RE-INAIL and SNAE, were derived showing the deviation and offset between normals of homologous points belonging to two meshes. From these, the mean errors (MRE) and root mean square error (RMSE) were extracted. In addition, shape and volume were evaluated using methods specific to the measurement of "agreement" developed by Bland and Altman. The accuracy of the 3D Echo scanner (which has a plate value of 1 mm across the diameter) was evaluated using a hollow calibration cylinder provided by the manufacturer and a hemisphere printed with a 3D printer. The experimental set-up was scanned in all three centers and multiple scanner configurations were analysed to see the best one to use. In particular, the absolute errors of each scan with respect to the nominal diameter and nominal circumference of cylinder and semi-sphere were reported in a Bland-Altman plot. Data acquisition The individual, that is eligible with respect to the inclusion/exclusion criteria and provides informed consent, will undergo three visits, each with a specific purpose. Visit 1 - general information regarding the subject, cause, and time since amputation is collected. Functional mobility, residual limb length and tissue type, tenderness or pain areas of the residual limb, and current type of prosthetic socket are assessed. The subject is also assessed on the Ampute Mobility Predictor, a standardized rating scale to understand the subject's level of mobility. Visit 2 - For each individual, two prosthetists created two casts, one with the manual one with the hydrostatic technique. The order in which the prosthetists begin and the order in which they perform the two casts in succession is randomized. In addition, the time required to perform each cast is recorded, and so 4 times were collected in total. The stump was scanned at the beginning and at the end of the visit with the Einscan Pro 2X Plus scanner (scans that were referred to as Start and End). The 4 negative casts were filled with a mixture of water and plaster to obtain the corresponding unrectified positive casts. At this point, only one of the two prosthetists, referred to as the "leader", proceeds with the rectification of their unrectified positive casts obtaining 2 rectified models, one for each technique. This involves adding and removing material to obtain a model having the same circumferences as those measured on the subject's stump. The time taken for the rectification is recorded. From each positive model, a PETG test socket was obtained by thermoforming. Visit 3 - the amputee tries on the two sockets, developed by the lead prosthetist, in both a seated and orthostatic position. The subject, who is unaware of which socket they are trying on, rates their comfort by assigning a grade of on a scale 0-10 (Socket Confort Score - SCS). The perceived comfort level for each socket is the primary outcome of the study. In addition, the subject can express the need to modify the sockets to alleviate any discomfort. The time to complete any modifications is also recorded. Data analysis by extraction of distance maps and scalar parameters To trace the evolution of the stump shape from the negative plasters to the socket, the cast and socket were scanned using the Echo 3D and Einscan 3D at each stage of the process: the stump at the beginning and at the end of visit 1, the negative casts, unrectified positive casts, rectified positive casts and sockets. Because there are multiple possible comparisons, the data analysis was organized into a sequence of 28 hypotheses divided into 7 groups. In general, they relate to comparisons between Start and End, between unrectified positive casts, between rectified positive casts, between unrectified positive casts obtained with the Aqua System technique and rectified positive casts obtained with the manual technique, between Start and unrectified positives, and between Start and rectified positives. Of these, the last two will be reported only in the appendix B and therefore will not be discussed in this paper. In addition, statistical hypotheses were added on the time and on the votes associated with socket given by subjects during the 3rd visit. For shape analysis and extraction of summary scalar parameters (circumferences on transverse sections, shape heights, and volume), the Socket Factory (SF) software developed at the INAIL Prosthesis Center was used and further developed. For the 3D comparison of the shapes, distance maps were generated from which two parameters RE-INAIL and SNAE were extracted. Bland- Altman plots were used for the analysis of summary scalar parameters. Bias and repeatability coefficient were extracted from these plots. Finally, descriptive statistical indicators were used first for the analysis of construction time and comfort ratings, and then t-test analysis and ANOVA were used. Data analysis by Morphing and PCA To extract information about the "typical" manual and Aqua System socket fabrication technique used at INAIL, an innovative technique has been used, based on two essential elements: 1. the geometric fitting of a reference model on all the scans performed, by means of the mesh morphing library named RBF-Morph, which we integrated into SocketFactory. 2. the use of a machine learning technique based on Principal Component Analysis (PCA), able to generate the average shape from a set of shapes (anatomical PCA) and to learn the difference between an unrectified positive and a rectified positive by the prosthetists (differential PCA). The ultimate purpose of this technique is to allow the software to learn the rectification technique of each leader prosthetist so that they can use this tool either as a support during manual rectification, or as a method to directly obtain the desired rectified positive from the unrectified positive. Since one of the fundamental aspects for the principal components analysis is that the meshes considered have the same topology, i.e. the same number of vertices and interconnections, it was necessary to use morphing. The morphing allows to project a reference mesh on a target mesh to obtain a mesh that has the reference mesh topology but target mesh morphology. It is necessary to identify two reference meshes, one for transtibial subjects and one for transfemoral ones. In addition, the best morphing method among those provided by the RBF-Morph library was identified. Once the rectification rule was learned, it was applied to each unrectified positive cast obtaining the possible rectified positive ones. These were then compared with those obtained by the prosthetists using RE-INAIL distance maps. Similarly, the rectification rule was also applied to the mean shape of unrectified positives obtained with anatomic PCA to obtain the mean shape of rectified positive. Finally, this rule was applied to a new unrectified positive Aqua System to obtain a possible rectified positive that was evaluated by the subject during the 3rd visit. RESULTS Einscan Pro 2X Plus scanner The MRE RE-INAIL data obtained from the distance maps between the scans of the 4 models made at each centre show a maximum mean error of 0,099 mm. Moreover, the MRE SNAE values show a maximum mean deviation between normals of 2,41°. Bland-Altman plots regarding circumferences show a bias for accuracy of -0,17 mm. Bland-Altman plots regarding volume show a consistent tendency to exceed the values obtained with the metrology scanner by 1,56 ml. Echo 3D Scanner For the 90°cylinder configuration, most of the differences between measurements are between limits of agreement of -0,26 mm and 0,3 0 mm. On the contrary, the 60° cylinder configuration shows higher differences than the nominal value, since 95% of the differences are between 0,57 mm and 1,93 mm. The values obtained for the semi-sphere in both configurations of the scanner head inclination are found to deviate from the nominal value by an average of 1,45 mm for the 90° configuration and by 0,94 for the 60° configuration. Data analysis The outcomes for each hypothesis have been reported in the table below, subdivided by transtibial (TT) and transfemoral (TF) amputations considering INAIL and NU data. Since the study is still in progress, for each statistical hypothesis (Hp. 2) we indicate whether it is true or false as a trend and specify whether or not the trend is supported by statistically significant differences based on the currently available sample. This allows for easier reading for the reader. Morphing and PCA In order to carry out the morphing the reference mesh was identified: TT reference has 31000 vertices and 62000 faces while TF reference has about 18000 vertices and 36000 faces. The best method of morphing to use is the implicit_vertex_gradient method in term of computational cost and a better accuracy. Anatomical PCA allowed to obtain the average shapes of unrectified and rectified positive casts using morphed mesh dataset. The principal components were then analysed: the first 3 components mostly influence shape variation for subjects with transtibial amputation while the first 4 mostly influence subjects with transfemoral amputation. The differential PCA allowed to calculate the rectification rule and apply it to the mean shape of the unrectified positive casts obtaining the possible rectified models. Also in this case, the shape variations induced by each principal component were studied. The calculated rectification rule was applied to all unrectified positive casts thus obtaining the respective rectified shapes that were compared with those made by each prostehtists. Mean error values related to RE-INAIL and SNAE parameters show deviations between -0,47 ÷ 0,59 mm and 3 ÷ 6,41° for transtibial and -0,77 ÷ 0,87 mm and 4,25 ÷ 6,97° for transfemoral. As an example, the calculated rectification rule was then applied to the Aqua System unrectified positive cast of INAIL thirteenth amputee TT to obtain a rectified positive cast on the basis of which the socket was made. This rectified positive cast was compared with the rectified positive cast made by the leader prosthetist in term of volume and circumferences on seven sections. DISCUSSIONS Einscan Pro 2X Plus scanner The maximum RE-INAIL MREs and SNAE MREs are lower than the clinical accuracy ranges found in the literature by da Sanders et al. , which are respectively 0,25 mm and 4°. The relative volume accuracy bias is 0,05%, which is lower than the data in the literature for Artec Eva and VIUScan. In addition, the Einscan Pro 2X Plus scanner showed high repeatability and reproducibility values for the volume with repeatability coefficients of 0,49% and 0,21%, respectively. Echo 3D Scanner The errors found in both cylinder configurations at 90° are smaller than those at 60°. The values obtained for the semi-sphere in both scanner head tilt configurations are greater than for the cylinder. All configurations for the cylinder, both at 60° and 90°, show good accuracy as they allow scanning with errors lower than those stated by the manufacturer, equal to 1 mm on the diameter and 3,14 mm on the circumference. However, the 90° configuration for the cylinder, both high and low, shows the best accuracy and was therefore used in this project. Data analysis Discussions are reported in the table in section RESULTS - Data analysis. They are divided between transtibial and transfemoral amputations considering both INAIL and NU data. Morphing and PCA PCA results allow to study the main components that mostly influence the shape variation and to determine the main effects that they have on the shape. The first main component mainly affects the length of the unrectified positive cast while the others affect its profile. The rectification rule was obtained thanks to differential PCA. The shape variations, induced by each principal component on the average shapes, were studied. The mean error values related to RE-INAIL and SNAE for the comparison between the rectified positive casts obtained with the software and those made by the prosthetists indicate that there are differences that would require further investigation of this result including tests on subjects. Finally, application of the rectification rule was tested to obtain the rectified positive cast and the socket of the thirteenth subject TT INAIL. Despite having a greater shape and volume than the thermoformed socket on the rectified positive made by the prosthetists, it was perceived as the most comfortable by the subject (associated rating 9,5). CONCLUSIONS The scanner accuracy analysis shows that both the Einscan Pro 2X Plus and the Echo 3D have good accuracy. The evaluations on the assumptions reported in the tables of the section RESULTS - Data analysis, however, will need to be expanded on all subjects participating in the study in order to evaluate consistency, efficiency, and outcomes. Finally, about morphing, anatomical and differential PCA, the outcomes obtained are hopeful but the dataset, from which the rectification rule is obtained, needs to be expanded.
Questa tesi si colloca all’interno del progetto “Comparative Effectiveness of Socket Casting Methods: Improving Form and Fit” della durata di tre anni finanziato dal Dipartimento della Difesa (DoD) degli Stati Uniti d’America ed è stata svolta presso il Centro Protesi INAIL a Vigorso di Budrio (BO). L’obiettivo del progetto è comparare invasature “a contatto totale” per soggetti con amputazione transtibiale e transfemorale realizzate con una tecnica manuale e mediante un sistema denominato “Aqua System”. L’amputazione consiste nell’asportazione dei segmenti distali di un arto per via chirurgica o traumatica. La necessità di soggetti con amputazione di riacquisire in parte o totalmente la loro indipendenza ha portato alla necessità di migliorare continuamente le protesi. L’invasatura rappresenta la parte più critica, in quanto costituisce l’interfaccia con il moncone del soggetto. Tra moncone e invasatura esiste una relazione indissolubile rappresentata da un reciproco scambio di forze e pressioni. L’invasatura deve consentire una distribuzione ottimale della pressione per garantirne l’adattamento e la funzionalità. Per questo motivo, il primo passo nella fabbricazione di un’invasatura consiste nel realizzare un calco in gesso che rappresenti fedelmente l’anatomia dell’arto residuo. L’invasatura finale è influenzata in primo luogo dal calco che può essere ottenuto mediante tecnica “manuale” oppure “idrostatica”. La presa misura del moncone manuale consiste nella realizzazione di un negativo del moncone opportunamente conformato da parte del tecnico ortopedico solo attraverso le proprie mani. Questa tecnica è sicuramente la più diffusa ed è chiaramente influenzata dall’esperienza e dalle conoscenze del tecnico ortopedico e con essa è difficile distribuire uniformemente la pressione intorno all'arto residuo. Sono state quindi sviluppate nuove tecniche che non si basano sulla manipolazione manuale del gesso e che vengono denominate “A pressione idrostatica con paziente in ortostatismo monopodalico". La tecnica è basata sull’uso di un cilindro in pressione dentro al quale viene posto il moncone del soggetto, che permette di ottenere un negativo del moncone che tenga conto della pressione uniforme generata dal sistema e della forza peso del soggetto. In letteratura non si trovano studi comparativi riguardanti l’efficacia (effectiveness), l’efficienza (efficiency) e la consistenza (consistency) fra queste due tecniche, sia per amputazioni transtibiali che transfemorali. Il progetto finanziato dal Dipartimento della Difesa (DoD) degli Stati Uniti, nell’ambito del quale si sviluppa il presente lavoro di tesi, è stato condotto proprio per colmare queste lacune. Il progetto prevede la collaborazione di tre centri: Northwestern University di Chicago (NU), Minnesota VA Health Care System (MVAHCS) e Centro Protesi INAIL (Istituto nazionale assicurazione contro gli infortuni sul lavoro). In ognuno dei quali verranno reclutati 20 soggetti con amputazione transtibiale e 10 con amputazione transfemorale per un totale di 90 partecipanti. Questa tesi ha lo scopo di presentare le metodologie seguite per analizzare le scansioni dei gessi e delle invasature realizzate con le due tecniche di presa misura dei primi 30 soggetti arruolati presso INAIL e NU. Per fare questo è stato necessario valutare l’accuratezza dei dispositivi di scansione utilizzati. Infine, è stata effettuata anche un’analisi delle forme dei monconi scansionati mediante Analisi delle Componenti Principali (PCA) e morphing al fine di ottenere le forme medie e le regole di stilizzazione applicate da ogni tecnico leader. MATERIALI E METODI Misura e analisi dell’accuratezza e precisione degli scanner Poiché una delle finalità dello studio è comprendere la consistenza delle tecniche di presa misura, è necessario potere misurare la forma del moncone durante la fase di presa misura e le successive forme realizzate dal tecnico ortopedico per costruire l’invasatura. Per potere completare questo obiettivo sono stati scelti due scanner tridimensionali, denominati “da interni” e da “esterni” in quanto in grado di scansionare, rispettivamente, l’interno di un modello negativo e la forma esterna di un modello positivo. Per scansionare gli interni è stato scelto lo scanner laser Echo 3D prodotto da Rodin4D, mentre per scansionare gli esterni è stato scelto lo scanner Einscan Pro 2X Plus. La capacità di effettuare misurazioni accurate in termini di forma e volume dell’arto residuo, con buona consistenza fra le misure effettuate da tecnici diversi, è essenziale per le finalità dello studio. Sono state, quindi, effettuate delle analisi preliminari sugli scanner utilizzati per confermare la qualità dei dati raccolti. Mediante lo scanner Einscan Pro 2X Plus due operatori, con diversa esperienza in ognuno dei tre centri, hanno scansionato 4 tipologie di invasature stilizzate. Queste sono anche state scansionate anche da un operatore esterno ai centri con uno scanner metrologico (Einscan FreeScan 5X Laser) che ha un’accuratezza di 0,02 mm e una precisione di 0,03 mm. Sono state, quindi, effettuate delle analisi di accuratezza delle scansioni ottenute dai due operatori per ogni centro rispetto a quelle ottenute con lo scanner metrologico e delle analisi di ripetibilità e riproducibilità sulle scansioni ottenute da due operatori. A questo scopo sono state generate delle mappe di distanza da cui sono stati estratti due parametri RE-INAIL e SNAE che mostrano rispettivamente lo scostamento e la deviazione tra le normali di punti omologhi appartenenti a due mesh. Da queste ultime sono stati ricavati gli errori medi (MRE) e la radice dell’errore quadratico medio (RMSE). Inoltre, sono stati valutati forma e volume mediante l’uso dei metodi propri della misura di ‘’agreement ‘’sviluppati da Bland e Altman. Per effettuare una valutazione sull’accuratezza dell’Echo 3D (che ha un valore di targa pari a 1 mm sul diametro), invece, è stato utilizzato un cilindro cavo di calibrazione fornito dall’azienda produttrice e una semi-sfera stampata con stampante 3D. Il set-up sperimentale è stato scansionato in tutti e tre i centri che partecipano al progetto e sono state analizzate più configurazioni dello scanner, per capire quale fosse la migliore da utilizzare. In particolare, gli errori assoluti di ogni scansione rispetto al diametro e alla circonferenza nominale di cilindro e semi-sfera sono stati riportati in un grafico di Bland-Altman. Acquisizione dei dati Una volta acquisito il consenso informato di un soggetto idoneo rispetto ai criteri di inclusione/esclusione, si procede con tre visite, ciascuna con una specifica finalità. Visita 1 – Vengono raccolte le informazioni generali relative al soggetto, la causa e il tempo trascorso dall’amputazione. Si valutano la mobilità funzionale, la lunghezza residua dell’arto e il tipo di tessuto, la sensibilità o le aree dolorose dell’arto residuo e il tipo attuale di invasatura protesica. Il soggetto viene anche sottoposto alla valutazione dell’Ampute Mobility Predictor, una scala di valutazione standardizzata per comprendere il livello di mobilità del soggetto. Visita 2 – Vengono eseguiti 4 calchi gessati, detti negativi, da parte di due tecnici ortopedici, utilizzando sia la tecnica manuale che quella idrostatica. L’ordine d’inizio dei tecnici ortopedici e quello con cui questi effettuano i due gessi in successione è randomizzato. Inoltre, il tempo necessario ad eseguire ognuno di essi viene cronometrato e di conseguenza si raccolgono 4 tempi in totale. Il moncone viene scansionato all’inizio ed alla fine della visita con lo scanner Einscan Pro 2X Plus (scansioni che sono state denominate Start ed End). I 4 negativi vengono riempiti con una miscela di acqua e gesso ottenendo i positivi grezzi corrispondenti. Successivamente solo uno dei due tecnici, sempre il medesimo e denominato “leader”, procede con la stilizzazione dei suoi positivi ottenendo 2 positivi stilizzati, uno per ciascuna tecnica di presa misura. Il tempo impiegato per la stilizzazione è stato registrato. Questo consiste nell’aggiunta o rimozione di materiale al fine di ottenere le stesse circonferenze misurate durante la presa misura. Da ciascuno dei positivi, si ottiene mediante termoformatura di una lastra di PETG un’invasatura di prova Visita 3 – Il soggetto prova le due invasature, sviluppate dal protesista leader, sia da seduto che in posizione ortostatica. Il soggetto non è a conoscenza di quale stia provando e a quale tecnico appartenga e ne valuta il comfort assegnando un voto da 0 a 10 (SCS). Questo livello di comfort percepito per ciascuna invasatura è l’outcome primario dello studio. Inoltre, il soggetto può esprimere la necessità di effettuare delle modifiche per alleviare eventuali fastidi. Anche il tempo per completare le eventuali modifiche è stato cronometrato. Analisi dei dati mediante estrazione di mappe distanza e parametri scalari Per potere tracciare l’evoluzione della forma del moncone dalla fase di presa misura all’invasatura dopo la prova, è stata effettuata la scansione tridimensionale mediante gli scanner Echo3D e Einscan 3D di ogni fase del processo: il moncone all’inizio ed alla fine della visita 1, i negativi, i positivi grezzi, gli stilizzati e le invasature. Poiché i possibili confronti sono molteplici, l’analisi dei dati è stata organizzata in una sequenza di 28 ipotesi suddivise in 7 gruppi. In generale sono relative al confronto tra Start e End, tra i positivi grezzi, tra gli stilizzati, tra i positivi ottenuti con la tecnica Aqua System e gli stilizzati ottenuti con quella manuale, tra Start e positivi grezzi e tra Start e stilizzati. Di queste, le ultime due saranno riportate solo in appendice B e quindi non saranno discusse all’interno di questo elaborato. Inoltre, sono state aggiunte ipotesi sulla valutazione dei tempi di costruzione di un’invasatura e sui voti associati alle invasature dati dai soggetti durante la visita 3. Per l’analisi delle forme e l’estrazione di parametri scalari riassuntivi (circonferenze su sezioni trasverse, altezze delle forme e volume), è stato utilizzato e ulteriormente sviluppato il software SocketFactory (SF) sviluppato presso il centro protesi INAIL. Per il confronto 3D delle forme sono state estratte delle mappe distanza da cui sono stati estratti due parametri RE-INAIL e SNAE. Per l’analisi dei parametri scalari riassuntivi sono stati utilizzati dei grafici di Bland- Altman da cui sono stati estratti bias e coefficiente di ripetibilità. Infine, per l’analisi dei tempi di costruzione e dei voti sul comfort si è invece fatto ricorso in prima battuta a indicatori statistici descrittivi, per poi procedere con analisi t-test e ANOVA. Analisi dei dati mediante Morphing e PCA Per potere estrarre informazioni circa la “tipica” tecnica impiegata da INAIL per realizzare un’invasatura a partire dal modello manuale e quello Aqua System è stata impiegata una tecnica innovativa, basata su due elementi essenziali: 1. il fitting geometrico di un modello di riferimento su tutte le scansioni effettuate, mediante la libreria di mesh morphing denominata RBF-Morph, da noi integrata in SocketFactory; 2. l’impiego di un una tecnica di machine learning basata sulla Principal Component Analysis (PCA), in grado di generare la forma media a partire da un set di forme (PCA anatomica) e di apprendere la differenza fra un positivo grezzo e un modello stilizzato dal tecnico (PCA differenziale). Questo ha come scopo quello di far apprendere al software la tecnica di stilizzazione di ogni tecnico ortopedico in modo che esso possa usare tale strumento come supporto durante la stilizzazione manuale, oppure uno strumento che permetta di ottenere, a partire dal positivo grezzo, direttamente lo stilizzato desiderato. Dato uno degli aspetti fondamentali per l’analisi delle componenti principali è che le mesh considerate abbiano la stessa topologia, cioè stesso numero di vertici e interconnessioni, è stato necessario utilizzare il morphing. Quest’ultimo permette di proiettare una mesh detta reference su una detta target in modo da ottenere una mesh che abbia la stessa topologia della reference ma morfologia della target. È stato, quindi, necessario individuare due mesh di riferimento, una per soggetti TT e una per quelli TF. Inoltre, è stato individuato quale fosse il metodo di morphing migliore tra quelli forniti dalla libreria RBF-Morph. Questi metodi permettono di ottenere mesh con la stessa topologia, aspetto fondamentale per poter attuare la PCA. Una volta appresa la regola di stilizzazione è stata applicata ad ogni positivo grezzo ottenendo i possibili stilizzati. Questi sono stati poi confrontati con quelli ottenuti dal tecnico ortopedico mediante mappe distanza RE-INAIL. Allo stesso modo, la regola di stilizzazione è anche stata applicata alla forma media dei positivi grezzi ottenuta con la PCA anatomica per ottenere lo stilizzato medio. Infine, questa regola è stata applicata ad un nuovo positivo grezzo Aqua System per ottenere un suo possibile stilizzato di cui è poi stato valutato il comfort dal soggetto durante la 3° visita. RISULTATI Scanner Einscan Pro 2X Plus I dati di MRE RE-INAIL ricavati dalle mappe di distanza tra le scansioni dei 4 modelli effettuate in ogni centro mostrano un errore medio massimo di 0,099 mm. Invece, i valori di MRE SNAE ottenuti mostrano una deviazione media massima tra le normali di 2,41°. Inoltre, i grafici di Bland-Altman relativi alle circonferenze mostrano un bias per la accuratezza di -0,17 mm. I grafici di Bland-Altman relativi ai volumi una tendenza costante di questi a superare di 1,56 ml i valori ottenuti con lo scanner metrologico. Scanner Echo 3D Per la configurazione a 90° del cilindro la maggior parte delle differenze tra le misurazioni è compresa tra i limiti di agreement pari a -0,26 mm e 0,30 mm. Invece, quella a 60° per il cilindro presenta differenze superiori rispetto al valore nominale in quanto il 95% delle differenze è compreso tra 0,57 mm e 1,93 mm. I valori ottenuti per la semi-sfera in entrambe le configurazioni dell’inclinazione della testa dello scanner risultano discostarsi dal valore nominale mediamente di 1,45 mm per la configurazione 90° e di 0,94 per quella a 60°. Analisi dei dati Gli esiti per ogni ipotesi sono stati riportati nelle tabelle sotto, suddivisi per amputazioni transtibiali (TT) e transfemorali (TF) considerando i dati di INAIL e NU. Visto che lo studio risulta ancora in corso con l’arruolamento di nuovi soggetti, per ogni ipotesi statistica (Hp. 2) si indica se questa sia vera o falsa come tendenza e si specifica se la tendenza sia supportata o meno da differenze statisticamente significative sulla base del campione attualmente disponibile. Ciò consente di offrire al lettore un quadro il più ampio possibile. Morphing e PCA Per effettuare il morphing le mesh reference individuate presentano per i TT circa 31000 vertici e 62000 facce mentre per i TF circa 18000 vertici e 36000 facce. Le analisi condotte per individuare il miglior metodo di morphing da utilizzare hanno evidenziato come il metodo implicit_vertex_gradient sia quello che ha un minor costo computazionale e una miglior accuratezza. Dopo aver effettuato il morphing sul dataset di mesh desiderate, la PCA anatomica ha permesso di ottenere le forme medie di positivi grezzi e stilizzati. Sono state poi analizzate le componenti principali: quelle che maggiormente influenzano la variazione di forma per i soggetti TT sono le prime 3 mentre per quelli TF sono le prime 4. La PCA differenziale ha permesso di calcolare la regola di stilizzazione ed applicarla alle forme medie dei positivi grezzi ottenendo i possibili stilizzati. Anche in questo caso sono state studiate le variazioni di forma indotte da ogni componente principale. Le stilizzazioni medie calcolate sono state applicate a tutti i positivi grezzi ottenendo così i rispettivi stilizzati che sono stati confrontati con quelli realizzati da ogni tecnico ortopedico. I valori di errore medio relativi a RE-INAIL e SNAE mostrano scostamenti compresi tra -0,47 ÷ 0,59 mm e 3 ÷ 6,41° per i transtibiali e -0,77 ÷ 0,87 mm e 4,25 ÷ 6,97° per i transfemorali. A titolo di esempio, la regola di stilizzazione calcolata è poi stata applicata al positivo grezzo Aqua System del tredicesimo soggetto TT INAIL, per ottenere uno stilizzato sulla base di cui è stata realizzata l’invasatura. Dallo stilizzato così ottenuto sono state ricavate le sezioni e costruiti i grafici di circonferenze e volumi per confrontarlo con gli stilizzati realizzati dal tecnico leader. DISCUSSIONI Scanner Einscan Pro 2X Plus Gli MRE RE-INAIL e MRE SNAE massimi risultano molto inferiori rispetto ai range clinici di accuratezza individuati in letteratura da Sanders et al. 1, che risultano essere rispettivamente pari a 0,25 mm e 4°. Il bias relativo all’accuratezza sul volume è pari a 0,05% e risulta essere inferiore ai dati presenti in letteratura di Artec Eva2 e VIUScan3. Inoltre, lo scanner Einscan Pro 2X Plus ha mostrato alti valori di ripetibilità e riproducibilità per il volume con coefficienti di ripetibilità rispettivamente pari al 0,49% e 0,21%. Scanner Echo 3D Per lo scanner Echo 3D gli errori riscontrati nelle 2 configurazioni del cilindro a 90° sono minori rispetto a quelli a 60°. I valori ottenuti per la semi-sfera in entrambe le configurazioni di inclinazione della testa dello scanner risultano maggiori rispetto al cilindro. Tutte le configurazioni per il cilindro, sia a 60° che a 90°, mostrano una buona accuratezza in quanto permettono di scansionare con errori inferiori a quelli dichiarati dal produttore, pari ad 1 mm sul diametro e quindi a 3,14 mm sulla circonferenza. Tuttavia, la configurazione a 90° per il cilindro, sia alta che bassa, mostra la miglior accuratezza e per questo è stata utilizzata nel progetto. Analisi dei dati Le discussioni sono riportate in tabella nella sezione RISULTATI - Analisi dei dati, sempre suddivise per amputazioni transtibiali e transfemorali considerando sia i dati INAIL che NU. Morphing e PCA Dall’analisi dei risultati ottenuti dalla PCA anatomica si sono potute studiare le componenti principali che maggiormente influenzano la variazione di forma e quali sono gli effetti principali che hanno sulla forma. Da essi si evince come la prima componente principale influenzi soprattutto la lunghezza del positivo grezzo mentre le altre il suo profilo. Con la PCA differenziale si è ottenuta la regola di stilizzazione e anche in questo caso sono state studiate le variazioni di forma indotte da ogni componente principale sulle forme medie. Si evince come la prima componente mostri le principali modifiche effettuate dal tecnico durante la stilizzazione. I valori di errore medio relativi a RE-INAIL e SNAE per il confronto tra gli stilizzati ottenuti con il software e quelli costruiti dal tecnico ortopedico, indicano che vi sono delle differenze che richiederebbero di indagare ulteriormente tale risultato effettuando anche prove sui soggetti. Infine, l’applicazione della regola di stilizzazione sul positivo grezzo Aqua System del tredicesimo soggetto TT INAIL mostra come lo stilizzato ottenuto dal morphing abbia una forma più larga e un volume maggiore ma comunque sia quello che ha portato alla realizzazione dell’invasatura a cui il soggetto ha associato un comfort maggiore (voto associato 9,5). CONCLUSIONI L’analisi condotta per la validazione della strumentazione utilizzata mostra come sia l’Einscan Pro 2X Plus che l’Echo 3D abbiano una buona accuratezza. Le valutazioni sulle ipotesi riportate nelle tabelle della sezione RISULTATI - Analisi dei dati, tuttavia, dovranno essere ampliate su tutti i soggetti partecipanti allo studio per poter rispondere agli obiettivi di consistenza, efficienza ed outcomes. Infine, per quanto riguarda il morphing e la PCA anatomica e differenziale, gli esiti ottenuti sono incoraggianti, ma è necessario ampliare il dataset da cui si ottiene la regola di stilizzazione.
Valutazione comparativa di tecniche di presa misura in amputati di arto inferiore mediante strumenti di geometria computazionale
Mele, Claudia;FAZZINI, ANITA
2020/2021
Abstract
The study described in this master’s thesis is part of the three-year project "Comparative Effectiveness of Socket Casting Methods: Improving Form and Fit" funded by the U.S. Department of Defense (DoD) and was carried out at the INAIL Prosthesis Center in Vigorso di Budrio (BO). The aim of the project is to compare total surface bearing for subject with transfemoral and transtibial amputation realised with two different methods to capture the anatomy of the residual limb: a traditional manual one, and a one that uses a system called 'Aqua System'. Amputation involves the removal of the distal segments of a limb by surgery or trauma. The need for individuals with amputation to regain some or all of their independence has led to the need for continuous improvement of prostheses. The socket represents the core of a prosthesis: it constitutes the interface with the individual’s residual limb. Between residual limb and socket, there is an indissoluble relationship represented by a reciprocal exchange of forces and pressures. The socket has allowed optimal pressure distribution to ensure fit and function. For this reason, the first step in fabricating a socket involves capturing the residuum anatomy through a plaster cast that is properly modelled to guarantee proper pressure distribution. The cast can be obtained either through "manual" or "hydrostatic" techniques. In the first case, the negative cast is properly molded and modelled manually by a certified prosthetist. This technique is certainly the most widespread and is clearly influenced by the experience and knowledge of the prosthetist, through it is difficult to distribute pressure evenly around the residual limb. Therefore, new techniques have been developed that do not rely on manual plaster manipulation and are called "Hydrostatic pressure with patient in monopodal orthostatism". The technique uses a pressure cylinder into which the subject's stump is placed, which allows capturing of the anatomy of the residuum as well as the pressure situation in the negative cast under full weight-bearing load. The literature lacks comparative studies of the effectiveness, efficiency and consistency of these two techniques for both transtibial and transfemoral amputations. The project funded by the U.S. Department of Defense (DoD), in which this thesis is developed, was conducted precisely to fill these gaps. The project involves the collaboration of three centers: Northwestern University of Chicago (NU), Minnesota VA Health Care System (MVAHCS) and INAIL (National Institute for Insurance against Accidents at Work,) Prosthesis Center. In each center, 20 subjects with transtibial amputation and 10 with transfemoral amputation will be recruited for a total of 90 participants. This thesis aims to present the methodologies followed to analyse the plaster and socket scans made with the two different methods to capture the anatomy of the residual limb of the first 30 subjects enrolled at INAIL and NU. To do this, it was necessary to evaluate the accuracy of the scanner devices used. Finally, an analysis of the shapes of the scanned stumps was performed using Principal Component Analysis (PCA) and morphing with the aim of obtaining the average shapes and rectification rules applied by each leading prosthetist. MATERIALS AND METHODS Measurement and Analysis of Scanner Accuracy and Precision Since one of the aims of the study is to understand the consistency of the methods to capture the anatomy of the residual limb, it is necessary to be able to measure the shape of the stump during the phase of casting and the subsequent shapes made by the prosthetists to build the socket. Two three-dimensional scanners have been chosen to complete this objective, they are called "interior" and "exterior" scanner because they are able to scan the interior of a negative model and the external shape of a positive model, respectively. The Echo 3D laser scanner manufactured by Rodin4D was chosen to scan the interior, while the Einscan 2X Pro Plus scanner was chosen to scan the exterior. The ability to make accurate measurements in terms of shape and volume of the residual limb, with good consistency between measurements made by different technicians, is essential for the purposes of the study. Therefore, preliminary accuracy, repeatability and reproducibility analyses were performed on the scanners used, to confirm the quality of the data collected into the project. Two operators with different experience from each center scanned 4 types of sockets with the Einscan Pro 2X Plus scanner. These sockets were also scanned by an operator external to the centers with the Einscan FreeScan 5X Laser metrological scanner which has an accuracy of 0,02 mm and a precision of 0,03 mm. Accuracy analyses were conducted by comparing the scans obtained at the centers to those obtained with the metrological scanner. Repeatability and reproducibility analyses were conducted by comparing the scans obtained by the two operators in the centers. Distance maps were generated and two parameters, RE-INAIL and SNAE, were derived showing the deviation and offset between normals of homologous points belonging to two meshes. From these, the mean errors (MRE) and root mean square error (RMSE) were extracted. In addition, shape and volume were evaluated using methods specific to the measurement of "agreement" developed by Bland and Altman. The accuracy of the 3D Echo scanner (which has a plate value of 1 mm across the diameter) was evaluated using a hollow calibration cylinder provided by the manufacturer and a hemisphere printed with a 3D printer. The experimental set-up was scanned in all three centers and multiple scanner configurations were analysed to see the best one to use. In particular, the absolute errors of each scan with respect to the nominal diameter and nominal circumference of cylinder and semi-sphere were reported in a Bland-Altman plot. Data acquisition The individual, that is eligible with respect to the inclusion/exclusion criteria and provides informed consent, will undergo three visits, each with a specific purpose. Visit 1 - general information regarding the subject, cause, and time since amputation is collected. Functional mobility, residual limb length and tissue type, tenderness or pain areas of the residual limb, and current type of prosthetic socket are assessed. The subject is also assessed on the Ampute Mobility Predictor, a standardized rating scale to understand the subject's level of mobility. Visit 2 - For each individual, two prosthetists created two casts, one with the manual one with the hydrostatic technique. The order in which the prosthetists begin and the order in which they perform the two casts in succession is randomized. In addition, the time required to perform each cast is recorded, and so 4 times were collected in total. The stump was scanned at the beginning and at the end of the visit with the Einscan Pro 2X Plus scanner (scans that were referred to as Start and End). The 4 negative casts were filled with a mixture of water and plaster to obtain the corresponding unrectified positive casts. At this point, only one of the two prosthetists, referred to as the "leader", proceeds with the rectification of their unrectified positive casts obtaining 2 rectified models, one for each technique. This involves adding and removing material to obtain a model having the same circumferences as those measured on the subject's stump. The time taken for the rectification is recorded. From each positive model, a PETG test socket was obtained by thermoforming. Visit 3 - the amputee tries on the two sockets, developed by the lead prosthetist, in both a seated and orthostatic position. The subject, who is unaware of which socket they are trying on, rates their comfort by assigning a grade of on a scale 0-10 (Socket Confort Score - SCS). The perceived comfort level for each socket is the primary outcome of the study. In addition, the subject can express the need to modify the sockets to alleviate any discomfort. The time to complete any modifications is also recorded. Data analysis by extraction of distance maps and scalar parameters To trace the evolution of the stump shape from the negative plasters to the socket, the cast and socket were scanned using the Echo 3D and Einscan 3D at each stage of the process: the stump at the beginning and at the end of visit 1, the negative casts, unrectified positive casts, rectified positive casts and sockets. Because there are multiple possible comparisons, the data analysis was organized into a sequence of 28 hypotheses divided into 7 groups. In general, they relate to comparisons between Start and End, between unrectified positive casts, between rectified positive casts, between unrectified positive casts obtained with the Aqua System technique and rectified positive casts obtained with the manual technique, between Start and unrectified positives, and between Start and rectified positives. Of these, the last two will be reported only in the appendix B and therefore will not be discussed in this paper. In addition, statistical hypotheses were added on the time and on the votes associated with socket given by subjects during the 3rd visit. For shape analysis and extraction of summary scalar parameters (circumferences on transverse sections, shape heights, and volume), the Socket Factory (SF) software developed at the INAIL Prosthesis Center was used and further developed. For the 3D comparison of the shapes, distance maps were generated from which two parameters RE-INAIL and SNAE were extracted. Bland- Altman plots were used for the analysis of summary scalar parameters. Bias and repeatability coefficient were extracted from these plots. Finally, descriptive statistical indicators were used first for the analysis of construction time and comfort ratings, and then t-test analysis and ANOVA were used. Data analysis by Morphing and PCA To extract information about the "typical" manual and Aqua System socket fabrication technique used at INAIL, an innovative technique has been used, based on two essential elements: 1. the geometric fitting of a reference model on all the scans performed, by means of the mesh morphing library named RBF-Morph, which we integrated into SocketFactory. 2. the use of a machine learning technique based on Principal Component Analysis (PCA), able to generate the average shape from a set of shapes (anatomical PCA) and to learn the difference between an unrectified positive and a rectified positive by the prosthetists (differential PCA). The ultimate purpose of this technique is to allow the software to learn the rectification technique of each leader prosthetist so that they can use this tool either as a support during manual rectification, or as a method to directly obtain the desired rectified positive from the unrectified positive. Since one of the fundamental aspects for the principal components analysis is that the meshes considered have the same topology, i.e. the same number of vertices and interconnections, it was necessary to use morphing. The morphing allows to project a reference mesh on a target mesh to obtain a mesh that has the reference mesh topology but target mesh morphology. It is necessary to identify two reference meshes, one for transtibial subjects and one for transfemoral ones. In addition, the best morphing method among those provided by the RBF-Morph library was identified. Once the rectification rule was learned, it was applied to each unrectified positive cast obtaining the possible rectified positive ones. These were then compared with those obtained by the prosthetists using RE-INAIL distance maps. Similarly, the rectification rule was also applied to the mean shape of unrectified positives obtained with anatomic PCA to obtain the mean shape of rectified positive. Finally, this rule was applied to a new unrectified positive Aqua System to obtain a possible rectified positive that was evaluated by the subject during the 3rd visit. RESULTS Einscan Pro 2X Plus scanner The MRE RE-INAIL data obtained from the distance maps between the scans of the 4 models made at each centre show a maximum mean error of 0,099 mm. Moreover, the MRE SNAE values show a maximum mean deviation between normals of 2,41°. Bland-Altman plots regarding circumferences show a bias for accuracy of -0,17 mm. Bland-Altman plots regarding volume show a consistent tendency to exceed the values obtained with the metrology scanner by 1,56 ml. Echo 3D Scanner For the 90°cylinder configuration, most of the differences between measurements are between limits of agreement of -0,26 mm and 0,3 0 mm. On the contrary, the 60° cylinder configuration shows higher differences than the nominal value, since 95% of the differences are between 0,57 mm and 1,93 mm. The values obtained for the semi-sphere in both configurations of the scanner head inclination are found to deviate from the nominal value by an average of 1,45 mm for the 90° configuration and by 0,94 for the 60° configuration. Data analysis The outcomes for each hypothesis have been reported in the table below, subdivided by transtibial (TT) and transfemoral (TF) amputations considering INAIL and NU data. Since the study is still in progress, for each statistical hypothesis (Hp. 2) we indicate whether it is true or false as a trend and specify whether or not the trend is supported by statistically significant differences based on the currently available sample. This allows for easier reading for the reader. Morphing and PCA In order to carry out the morphing the reference mesh was identified: TT reference has 31000 vertices and 62000 faces while TF reference has about 18000 vertices and 36000 faces. The best method of morphing to use is the implicit_vertex_gradient method in term of computational cost and a better accuracy. Anatomical PCA allowed to obtain the average shapes of unrectified and rectified positive casts using morphed mesh dataset. The principal components were then analysed: the first 3 components mostly influence shape variation for subjects with transtibial amputation while the first 4 mostly influence subjects with transfemoral amputation. The differential PCA allowed to calculate the rectification rule and apply it to the mean shape of the unrectified positive casts obtaining the possible rectified models. Also in this case, the shape variations induced by each principal component were studied. The calculated rectification rule was applied to all unrectified positive casts thus obtaining the respective rectified shapes that were compared with those made by each prostehtists. Mean error values related to RE-INAIL and SNAE parameters show deviations between -0,47 ÷ 0,59 mm and 3 ÷ 6,41° for transtibial and -0,77 ÷ 0,87 mm and 4,25 ÷ 6,97° for transfemoral. As an example, the calculated rectification rule was then applied to the Aqua System unrectified positive cast of INAIL thirteenth amputee TT to obtain a rectified positive cast on the basis of which the socket was made. This rectified positive cast was compared with the rectified positive cast made by the leader prosthetist in term of volume and circumferences on seven sections. DISCUSSIONS Einscan Pro 2X Plus scanner The maximum RE-INAIL MREs and SNAE MREs are lower than the clinical accuracy ranges found in the literature by da Sanders et al. , which are respectively 0,25 mm and 4°. The relative volume accuracy bias is 0,05%, which is lower than the data in the literature for Artec Eva and VIUScan. In addition, the Einscan Pro 2X Plus scanner showed high repeatability and reproducibility values for the volume with repeatability coefficients of 0,49% and 0,21%, respectively. Echo 3D Scanner The errors found in both cylinder configurations at 90° are smaller than those at 60°. The values obtained for the semi-sphere in both scanner head tilt configurations are greater than for the cylinder. All configurations for the cylinder, both at 60° and 90°, show good accuracy as they allow scanning with errors lower than those stated by the manufacturer, equal to 1 mm on the diameter and 3,14 mm on the circumference. However, the 90° configuration for the cylinder, both high and low, shows the best accuracy and was therefore used in this project. Data analysis Discussions are reported in the table in section RESULTS - Data analysis. They are divided between transtibial and transfemoral amputations considering both INAIL and NU data. Morphing and PCA PCA results allow to study the main components that mostly influence the shape variation and to determine the main effects that they have on the shape. The first main component mainly affects the length of the unrectified positive cast while the others affect its profile. The rectification rule was obtained thanks to differential PCA. The shape variations, induced by each principal component on the average shapes, were studied. The mean error values related to RE-INAIL and SNAE for the comparison between the rectified positive casts obtained with the software and those made by the prosthetists indicate that there are differences that would require further investigation of this result including tests on subjects. Finally, application of the rectification rule was tested to obtain the rectified positive cast and the socket of the thirteenth subject TT INAIL. Despite having a greater shape and volume than the thermoformed socket on the rectified positive made by the prosthetists, it was perceived as the most comfortable by the subject (associated rating 9,5). CONCLUSIONS The scanner accuracy analysis shows that both the Einscan Pro 2X Plus and the Echo 3D have good accuracy. The evaluations on the assumptions reported in the tables of the section RESULTS - Data analysis, however, will need to be expanded on all subjects participating in the study in order to evaluate consistency, efficiency, and outcomes. Finally, about morphing, anatomical and differential PCA, the outcomes obtained are hopeful but the dataset, from which the rectification rule is obtained, needs to be expanded.File | Dimensione | Formato | |
---|---|---|---|
Executive Summary_TESI FAZZINI-MELE.pdf
non accessibile
Descrizione: Executive Summary Tesi Fazzini-Mele
Dimensione
1.73 MB
Formato
Adobe PDF
|
1.73 MB | Adobe PDF | Visualizza/Apri |
TESI FAZZINI MELE.pdf
non accessibile
Descrizione: Tesi Fazzini-Mele: Valutazione comparativa di tecniche di presa misura in amputati di arto inferiore mediante strumenti di geometria computazionale
Dimensione
49.76 MB
Formato
Adobe PDF
|
49.76 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/10589/187885