Refine
Document Type
- Article (10)
- In Proceedings (5)
- ZIB-Report (5)
- Research data (3)
- Doctoral Thesis (1)
- Master's Thesis (1)
- Poster (1)
Is part of the Bibliography
- no (26)
Keywords
- Sparse Geometry Reconstruction (2)
- Statistical Shape Models (2)
- 4D-CT (1)
- Deep Learning (1)
- Deep learning (1)
- Gruppenweise Registrierung (1)
- Knee Arthroplasty (1)
- Total Knee Arthoplasty (1)
- cartilage (1)
- cartilage morphometry (1)
Institute
In der Strahlentherapie von Lungentumoren kann mittels Dosisakkumulation der Einfluss von Atembewegungen auf statisch geplante Dosisverteilungen abgeschätzt werden. Grundlage sind 4D-CT-Daten des Patienten, aus denen mittels nicht-linearer Bildregistrierung eine Sequenz von Bewegungsfeldern berechnet wird. Typischerweise werden Methoden der paarweisen Bildregistrierung eingesetzt, d.h. konsekutiv zwei Atemphasen aufeinander registriert. Hierbei erfolgt i.d.R. eine physiologisch nicht plausible Anpassung der Felder an CT-Bewegungsartefakte. Gruppenweise Registrierungsansätze berücksichtigen hingegen gleichzeitig sämtliche Bilddaten des 4D-CT-Scans und ermöglichen die Integration von zeitlichen Konsistenzbetrachtungen. In diesem Beitrag wird der potentielle Vorteil der gruppen- im Vergleich zur paarweisen Registrierung in artefaktbehafteten 4D-CT-Daten untersucht.
Precise voxel trajectory estimation in 4D CT images is a prerequisite for reliable dose accumulation during 4D treatment planning. 4D CT image data is, however, often affected by motion artifacts and applying standard pairwise registration to such data sets bears the risk of aligning anatomical structures to artifacts – with physiologically unrealistic trajectories being the consequence. In this work, the potential of a novel non-linear hybrid intensity- and feature-based groupwise registration method for robust motion field estimation in artifact-affected 4D CT image data is investigated. The overall registration performance is evaluated on the DIR-lab datasets; Its robustness if applied to artifact-affected data sets is analyzed using clinically acquired data sets with and without artifacts. The proposed registration approach achieves an accuracy comparable to the state-of-the-art (subvoxel accuracy), but smoother voxel trajectories compared to pairwise registration. Even more important: it maintained accuracy and trajectory smoothness in the presence of image artifacts – in contrast to standard pairwise registration, which yields higher landmark-based registration errors and a loss of trajectory smoothness when applied to artifact-affected data sets.
Vertebral labelling and segmentation are two fundamental tasks in an automated spine processing pipeline. Reliable and accurate processing of spine images is expected to benefit clinical decision support systems for diagnosis, surgery planning, and population-based analysis of spine and bone health. However, designing automated algorithms for spine processing is challenging predominantly due to considerable variations in anatomy and acquisition protocols and due to a severe shortage of publicly available data. Addressing these limitations, the Large Scale Vertebrae Segmentation Challenge (VerSe) was organised in conjunction with the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in 2019 and 2020, with a call for algorithms tackling the labelling and segmentation of vertebrae. Two datasets containing a total of 374 multi-detector CT scans from 355 patients were prepared and 4505 vertebrae have individually been annotated at voxel level by a human-machine hybrid algorithm (https://osf.io/nqjyw/, https://osf.io/t98fz/). A total of 25 algorithms were benchmarked on these datasets. In this work, we present the results of this evaluation and further investigate the performance variation at the vertebra level, scan level, and different fields of view. We also evaluate the generalisability of the approaches to an implicit domain shift in data by evaluating the top-performing algorithms of one challenge iteration on data from the other iteration. The principal takeaway from VerSe: the performance of an algorithm in labelling and segmenting a spine scan hinges on its ability to correctly identify vertebrae in cases of rare anatomical variations. The VerSe content and code can be accessed at: https://github.com/anjany/verse.
Gruppenweise Registrierung zur robusten Bewegungsfeldschätzung in artefaktbehafteten 4D-CT-Bilddaten
(2015)
Das Ziel der Strahlentherapie ist, eine möglichst hohe Dosis in den Tumor zu applizieren und zeitgleich die Strahlenexposition des Normalgewebes zu minimieren. Insbesondere bei thorakalen und abdominalen Tumoren treten aufgrund der Atmung während der Bestrahlung große, komplexe und patientenspezifisch unterschiedliche Bewegungen der Gewebe auf. Um den Einfluss dieser Bewegung auf die i.d.R. statisch geplante Dosisverteilung abzuschätzen, können unter Verwendung der nicht-linearen Bildregistrierung anhand von 3D-CT-Aufnahmen eines Atmungszyklus - also 4D-CT-Daten - zunächst die Bewegungsfelder für die strahlentherapeutisch relevanten Strukturen, beispielsweise für die Lunge, berechnet werden. Diese Informationen bilden die Grundlage für sogenannte 4D-Dosisberechnungs- oder Dosisakkumulationsverfahren. Deren Genauigkeit hängt aber wesentlich von der Genauigkeit der Bewegungsfeldschätzung ab.
Klassisch erfolgt die Berechnung der Bewegungsfelder mittels paarweiser Bildregistrierung, womit für die Berechnung des Bewegungsfeldes zwischen zwei Bildern im Allgemeinen eine sehr hohe Genauigkeit erreicht wird. Auch für CT-Bilder, die Bewegungsartefakte, wie beispielsweise doppelte oder unvollständige Strukturen, enthalten, wird unter Verwendung der paarweisen Bildregistrierung im Kontext der Registrierung eine exakte Abbildung der anatomischen Strukturen zwischen den beiden Bildern erreicht. Dabei erfolgt aber eine physiologisch unplausible Anpassung der Felder an die Artefakte. Bei Verwendung der paarweisen Bildregistrierung müssen weiterhin für einen Atemzyklus die Voxel-Trajektorien aus Bewegungsfeldern zwischen mehreren dreidimensionalen Bildern zusammengesetzt werden. Durch Bewegungsartefakte entsprechen diese Trajektorien dann teilweise keiner natürlichen Bewegung der anatomischen Strukturen. Diese Ungenauigkeit stellt in der klinischen Anwendung ein Problem dar; dies gilt umso mehr, wenn Bewegungsartefakte im Bereich eines Tumors vorliegen.
Im Gegensatz zu der paarweisen Registrierung kann mit der gruppenweisen Registrierung das Problem der durch Bewegungsartefakte hervorgerufenen ungenauen Abbildung der physiologischen Gegebenheiten dadurch reduziert werden, dass im Registrierungsprozess Bildinformationen aller Bilder, also in diesem Kontext der CT-Daten zu unterschiedlichen Atemphasen, gleichzeitig genutzt werden. Es kann bereits im Registrierungsprozess eine zeitliche Glattheit der Voxel-Trajektorien gefordert werden.
In dieser Arbeit wird eine Methode zur B-Spline-basierten zeitlich regularisierten gruppenweisen Registrierung entwickelt. Die Genauigkeit der entwickelten Methode wird für frei zugängliche klinische Datensätze landmarkenbasiert evaluiert. Dabei wird mit dem Target Registration Error (TRE) die durchschnittliche dreidimensionale euklidische Distanz zwischen den korrespondierenden Landmarken nach Transformation der Landmarken bezeichnet. Eine Genauigkeit in der Größenordnung von aktuellen paarweisen Registrierungen verdeutlicht die Qualität des vorgestellten Registrierungs-Algorithmus. Anschließend werden die Vorteile der gruppenweisen Registrierung durch Experimente an einem Lungenphantom und an manipulierten, artefaktbehafteten klinischen 4D-CT-Bilddaten demonstriert. Dabei werden unter Verwendung der gruppenweisen Registrierung im Vergleich zu der paarweisen Registrierung glattere Trajektorien berechnet, die der realen Bewegung der anatomischen Strukturen stärker entsprechen. Für die Patientendaten wird außerdem anhand von automatisch detektierten Landmarken der TRE ausgewertet. Der TRE verschlechterte sich für die paarweise Bildregistrierung unter Vorliegen von Bewegungsartefakten von durchschnittlich 1,30 mm auf 3,94 mm. Auch hier zeigte sich für die gruppenweise Registrierung die Robustheit gegenüber Bewegungsartefakten und der TRE verschlechterte sich nur geringfügig von 1,45 mm auf 1,71 mm.
Knee osteoarthritis (KOA) is a degenerative disease that leads to pain and loss of function. It is estimated to affect over 500 million humans world-wide and is one of the most common reasons for disability. KOA is usually diagnosed by radiologists or clinical experts by anamnesis, physical examination, and by assessing medical image data. The latter is typically acquired using X-Ray or magnetic resonance imaging. Since manual image reading is subjective, tedious and time-consuming, automated methods are required for a fast and objective decision support and for a better understanding of the pathogenesis of KOA.
This thesis sets a foundation towards automated computation of image-based KOA biomarkers for holistic assessment of the knee. This involves the assessment of multiple knee bones and soft tissues. An assessment of particular structures requires localization of these tissues. In order to automate a faithful localization of anatomical structures, deep learning-based methods are investigated and utilized. Additionally, convolutional neural networks (CNNs) are used for classification of medical image data, i.e., for a direct determination of the disease status and to detect anatomical structures and landmarks. The automatically computed anatomical volumes, locations, and other measurements are finally compared to values acquired by clinical experts and evaluated for clustering of KOA groups, classification of KOA severity, prediction of KOA progression, and prediction of total knee replacement.
In various experiments it is shown that CNN-based methods are suitable for accurate medical image segmentation, object detection, landmark detection, and direct classification of disease stages from the image data. Computed features related to the menisci are found to be most expressive in terms of clustering of KOA groups and
predicting of future disease states, thus allowing diagnosis of current KOA conditions and prediction of future conditions.
The conclusion of this thesis is that machine learning-based, fully automated processing of medical image data shows potential for diagnosis and prediction of KOA grades. Future studies could investigate additional features in order to achieve an assessment of the whole knee or validate the findings of this work in clinical
studies.
In this study we investigate methods for fitting a Statistical Shape Model (SSM) to intraoperatively acquired point cloud data from a surgical navigation system. We validate the fitted models against the pre-operatively acquired Magnetic Resonance Imaging (MRI) data from the same patients.
We consider a cohort of 10 patients who underwent navigated total knee arthroplasty. As part of the surgical protocol the patients’ distal femurs were partially digitized. All patients had an MRI scan two months pre-operatively. The MRI data were manually segmented and the reconstructed bone surfaces used as ground truth against which the fit was compared. Two methods were used to fit the SSM to the data, based on (1) Iterative Closest Points (ICP) and (2) Gaussian Mixture Models (GMM).
For both approaches, the difference between model fit and ground truth surface averaged less than 1.7 mm and excellent correspondence with the distal femoral morphology can be demonstrated.
In this study we investigate methods for fitting a Statistical Shape Model (SSM) to intraoperatively acquired point cloud data from a surgical navigation system. We validate the fitted models against the pre-operatively acquired Magnetic Resonance Imaging (MRI) data from the same patients.
We consider a cohort of 10 patients who underwent navigated total knee arthroplasty. As part of the surgical protocol the patients’ distal femurs were partially digitized. All patients had an MRI scan two months pre-operatively. The MRI data were manually segmented and the reconstructed bone surfaces used as ground truth against which the fit was compared. Two methods were used to fit the SSM to the data, based on (1) Iterative Closest Points (ICP) and (2) Gaussian Mixture Models (GMM).
For both approaches, the difference between model fit and ground truth surface averaged less than 1.7 mm and excellent correspondence with the distal femoral morphology can be demonstrated.
The reconstruction of an object's shape or surface from a set of 3D points is a common topic in materials and life sciences, computationally handled in computer graphics. Such points usually stem from optical or tactile 3D coordinate measuring equipment. Surface reconstruction also appears in medical image analysis, e.g. in anatomy reconstruction from tomographic measurements or the alignment of intra-operative navigation and preoperative planning data. In contrast to mere 3D point clouds, medical imaging yields contextual information on the 3D point data that can be used to adopt prior information on the shape that is to be reconstructed from the measurements. In this work we propose to use a statistical shape model (SSM) as a prior for surface reconstruction. The prior knowledge is represented by a point distribution model (PDM) that is associated with a surface mesh. Using the shape distribution that is modelled by the PDM, we reformulate the problem of surface reconstruction from a probabilistic perspective based on a Gaussian Mixture Model (GMM). In order to do so, the given measurements are interpreted as samples of the GMM. By using mixture components with anisotropic covariances that are oriented according to the surface normals at the PDM points, a surface-based tting is accomplished. By estimating the parameters of the GMM in a maximum a posteriori manner, the reconstruction of the surface from the given measurements is achieved. Extensive experiments suggest that our proposed approach leads to superior surface reconstructions compared to Iterative Closest Point (ICP) methods.
We present a method for the automated segmentation of knee bones and cartilage from magnetic resonance imaging that combines a priori knowledge of anatomical shape with Convolutional Neural Networks (CNNs). The proposed approach incorporates 3D Statistical Shape Models (SSMs) as well as 2D and 3D CNNs to achieve a robust and accurate segmentation of even highly pathological knee structures. The shape models and neural networks employed are trained using data of the Osteoarthritis Initiative (OAI) and the MICCAI grand challenge "Segmentation of Knee Images 2010" (SKI10), respectively. We evaluate our method on 40 validation and 50 submission datasets of the SKI10 challenge. For the first time, an accuracy equivalent to the inter-observer variability of human readers has been achieved in this challenge. Moreover, the quality of the proposed method is thoroughly assessed using various measures for data from the OAI, i.e. 507 manual segmentations of bone and cartilage, and 88 additional manual segmentations of cartilage. Our method yields sub-voxel accuracy for both OAI datasets. We made the 507 manual segmentations as well as our experimental setup publicly available to further aid research in the field of medical image segmentation. In conclusion, combining statistical anatomical knowledge via SSMs with the localized classification via CNNs results in a state-of-the-art segmentation method for knee bones and cartilage from MRI data.
