Matthias Wilms

Extra tree forests for sub-acute ischemic stroke lesion segmentation in MR sequences.

BACKGROUND To analyse the relationship between structure and (dys-)function of the brain after stroke, accurate and repeatable segmentation of the lesion area in magnetic resonance (MR) images is required. Manual delineation, the current gold standard, is time consuming and suffers from high intra- and inter-observer differences. NEW METHOD A new approach is presented for the automatic and reproducible segmentation of sub-acute ischemic stroke lesions in MR images in the presence of other pathologies. The proposition is based on an Extra Tree forest framework for voxel-wise classification and mainly intensity derived image features are employed. RESULTS A thorough investigation of multi-spectral variants, which combine the information from multiple MR sequences, finds the fluid attenuated inversion recovery sequence to be both required and sufficient for a good segmentation result. The accuracy can be further improved by adding features extracted from the T1-weighted and the diffusion weighted sequences. The use of other sequences is discouraged, as they impact negatively on the results. COMPARISON WITH EXISTING METHODS Quantitative evaluation was carried out on 37 clinical cases. With a Dice coefficient of 0.65, the method outperforms earlier published methods. CONCLUSIONS The approach proves especially suitable to differentiate between new stroke and other white matter lesions based on the FLAIR sequence alone. This, and the high overlap, renders it suitable for automatic screening of large databases of MR scans, e.g. for a subsequent neuropsychological investigation. Finally, each features importance is assessed in detail and the approachs statistical dependency on clinical and image characteristics is investigated.

ISLES 2015 - A public evaluation benchmark for ischemic stroke lesion segmentation from multispectral MRI

&NA; Ischemic stroke is the most common cerebrovascular disease, and its diagnosis, treatment, and study relies on non‐invasive imaging. Algorithms for stroke lesion segmentation from magnetic resonance imaging (MRI) volumes are intensely researched, but the reported results are largely incomparable due to different datasets and evaluation schemes. We approached this urgent problem of comparability with the Ischemic Stroke Lesion Segmentation (ISLES) challenge organized in conjunction with the MICCAI 2015 conference. In this paper we propose a common evaluation framework, describe the publicly available datasets, and present the results of the two sub‐challenges: Sub‐Acute Stroke Lesion Segmentation (SISS) and Stroke Perfusion Estimation (SPES). A total of 16 research groups participated with a wide range of state‐of‐the‐art automatic segmentation algorithms. A thorough analysis of the obtained data enables a critical evaluation of the current state‐of‐the‐art, recommendations for further developments, and the identification of remaining challenges. The segmentation of acute perfusion lesions addressed in SPES was found to be feasible. However, algorithms applied to sub‐acute lesion segmentation in SISS still lack accuracy. Overall, no algorithmic characteristic of any method was found to perform superior to the others. Instead, the characteristics of stroke lesion appearances, their evolution, and the observed challenges should be studied in detail. The annotated ISLES image datasets continue to be publicly available through an online evaluation system to serve as an ongoing benchmarking resource (www.isles‐challenge.org). HighlightsEvaluation framework for automatic stroke lesion segmentation from MRIPublic multi‐center, multi‐vendor, multi‐protocol databases releasedOngoing fair and automated benchmark with expert created ground truth setsComparison of 14+7 groups who responded to an open challenge in MICCAISegmentation feasible in acute and unsolved in sub‐acute cases Graphical abstract Figure. No caption available.

Multivariate regression approaches for surrogate-based diffeomorphic estimation of respiratory motion in radiation therapy

Breathing-induced location uncertainties of internal structures are still a relevant issue in the radiation therapy of thoracic and abdominal tumours. Motion compensation approaches like gating or tumour tracking are usually driven by low-dimensional breathing signals, which are acquired in real-time during the treatment. These signals are only surrogates of the internal motion of target structures and organs at risk, and, consequently, appropriate models are needed to establish correspondence between the acquired signals and the sought internal motion patterns. In this work, we present a diffeomorphic framework for correspondence modelling based on the Log-Euclidean framework and multivariate regression. Within the framework, we systematically compare standard and subspace regression approaches (principal component regression, partial least squares, canonical correlation analysis) for different types of common breathing signals (1D: spirometry, abdominal belt, diaphragm tracking; multi-dimensional: skin surface tracking). Experiments are based on 4D CT and 4D MRI data sets and cover intra- and inter-cycle as well as intra- and inter-session motion variations. Only small differences in internal motion estimation accuracy are observed between the 1D surrogates. Increasing the surrogate dimensionality, however, improved the accuracy significantly; this is shown for both 2D signals, which consist of a common 1D signal and its time derivative, and high-dimensional signals containing the motion of many skin surface points. Eventually, comparing the standard and subspace regression variants when applied to the high-dimensional breathing signals, only small differences in terms of motion estimation accuracy are found.

A 4d statistical shape model for automated segmentation of lungs with large tumors

Segmentation of lungs with large tumors is a challenging and time-consuming task, especially for 4D CT data sets used in radiation therapy. Existing lung segmentation methods are ineffective in these cases, because they are either not able to deal with large tumors and/or process every 3D image independently neglecting temporal information. In this paper, we present a approach for model-based 4D segmentation of lungs with large tumors in 4D CT data sets. In our approach, a 4D statistical shape model that accounts for inter- and intra-patient variability is fitted to the 4D image sequence, and the segmentation result is refined by a 4D graph-based optimal surface finding. The approach is evaluated using 10 4D CT data sets of lung tumor patients. The segmentation results are compared with a standard intensity-based approach and a 3D version of the presented model-based segmentation method. The intensity-based approach shows a better performance for normal lungs, however, fails in presence of large lung tumors. Although overall performance of 3D and 4D model-based segmentation is similar, the results indicate improved temporal coherence and improved robustness with respect to the segmentation parameters for the 4D model-based segmentation.

Statistical Shape Modeling of the Left Ventricle: Myocardial Infarct Classification Challenge

Statistical shape modeling is a powerful tool for visualizing and quantifying geometric and functional patterns of the heart. After myocardial infarction (MI), the left ventricle typically remodels in response to physiological challenges. Several methods have been proposed in the literature to describe statistical shape changes. Which method best characterizes the left ventricular remodeling after MI is an open research question. A better descriptor of remodeling is expected to provide a more accurate evaluation of disease status in MI patients. We therefore designed a challenge to test shape characterization in MI given a set of three-dimensional left ventricular surface points. The training set comprised 100 MI patients, and 100 asymptomatic volunteers (AV). The challenge was initiated in 2015 at the Statistical Atlases and Computational Models of the Heart workshop, in conjunction with the MICCAI conference. The training set with labels was provided to participants, who were asked to submit the likelihood of MI from a different (validation) set of 200 cases (100 AV and 100 MI). Sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve were used as the outcome measures. The goals of this challenge were to 1) establish a common dataset for evaluating statistical shape modeling algorithms in MI, and 2) test whether statistical shape modeling provides additional information characterizing MI patients over standard clinical measures. Eleven groups with a wide variety of classification and feature extraction approaches participated in this challenge. All methods achieved excellent classification results with accuracy ranges from 0.83 to 0.98. The areas under the receiver operating characteristic curves were all above 0.90. Four methods showed significantly higher performance than standard clinical measures. The dataset and software for evaluation are available from the Cardiac Atlas Project website.11 http://www.cardiacatlas.org.

Ischemic stroke lesion segmentation in multi-spectral MR images with support vector machine classifiers

Automatic segmentation of ischemic stroke lesions in magnetic resonance (MR) images is important in clinical practice and for neuroscientific trials. The key problem is to detect largely inhomogeneous regions of varying sizes, shapes and locations. We present a stroke lesion segmentation method based on local features extracted from multi-spectral MR data that are selected to model a human observer’s discrimination criteria. A support vector machine classifier is trained on expert-segmented examples and then used to classify formerly unseen images. Leave-one-out cross validation on eight datasets with lesions of varying appearances is performed, showing our method to compare favourably with other published approaches in terms of accuracy and robustness. Furthermore, we compare a number of feature selectors and closely examine each feature’s and MR sequence’s contribution.

Direct Visuo-Haptic 4D Volume Rendering Using Respiratory Motion Models

This article presents methods for direct visuo-haptic 4D volume rendering of virtual patient models under respiratory motion. Breathing models are computed based on patient-specific 4D CT image data sequences. Virtual patient models are visualized in real-time by ray casting based rendering of a reference CT image warped by a time-variant displacement field, which is computed using the motion models at run-time. Furthermore, haptic interaction with the animated virtual patient models is provided by using the displacements computed at high rendering rates to translate the position of the haptic device into the space of the reference CT image. This concept is applied to virtual palpation and the haptic simulation of insertion of a virtual bendable needle. To this aim, different motion models that are applicable in real-time are presented and the methods are integrated into a needle puncture training simulation framework, which can be used for simulated biopsy or vessel puncture in the liver. To confirm real-time applicability, a performance analysis of the resulting framework is given. It is shown that the presented methods achieve mean update rates around 2,000 Hz for haptic simulation and interactive frame rates for volume rendering and thus are well suited for visuo-haptic rendering of virtual patients under respiratory motion.

Pulmonary lobe segmentation with level sets

Automatic segmentation of the separate human lung lobes is a crucial task in computer aided diagnostics and intervention planning, and required for example for determination of disease spreading or pulmonary parenchyma quantification. In this work, a novel approach for lobe segmentation based on multi-region level sets is presented. In a first step, interlobular fissures are detected using a supervised enhancement filter. The fissures are then used to compute a cost image, which is incorporated in the level set approach. By this, the segmentation is drawn to the fissures at places where structure information is present in the image. In areas with incomplete fissures (e.g. due to insufficient image quality or anatomical conditions) the smoothing term of the level sets applies and a closed continuation of the fissures is provided. The approach is tested on nine pulmonary CT scans. It is shown that incorporating the additional force term improves the segmentation significantly. On average, 83% of the left fissure is traced correctly; the right oblique and horizontal fissures are properly segmented to 76% and 48%, respectively.

Training CNNs for Image Registration from Few Samples with Model-based Data Augmentation

Convolutional neural networks (CNNs) have been successfully used for fast and accurate estimation of dense correspondences between images in computer vision applications. However, much of their success is based on the availability of large training datasets with dense ground truth correspondences, which are only rarely available in medical applications. In this paper, we, therefore, address the problem of CNNs learning from few training data for medical image registration. Our contributions are threefold: (1) We present a novel approach for learning highly expressive appearance models from few training samples, (2) we show that this approach can be used to synthesize huge amounts of realistic ground truth training data for CNN-based medical image registration, and (3) we adapt the FlowNet architecture for CNN-based optical flow estimation to the medical image registration problem. This pipeline is applied to two medical data sets with less than 40 training images. We show that CNNs learned from the proposed generative model outperform those trained on random deformations or displacement fields estimated via classical image registration.

Image Features for Brain Lesion Segmentation Using Random Forests

From clinical practice as well as research methods arises the need for accurate, reproducible and reliable segmentation of pathological areas from brain MR scans. This paper describes a set of hand-selected, voxel-based image features highly suitable for the tissue discrimination task. Embedded in a random decision forest framework, the proposed method was applied to sub-acute ischemic stroke (ISLES 2015 - SISS), acute ischemic stroke (ISLES 2015 - SPES) and glioma (BRATS 2015) segmentation with only minor adaptation. For all of these three challenges, our generic approach received high ranks, among them a second place. The outcome underlines the robustness of our features for segmentation in brain MR, while simultaneously stressing the necessity for highly specialized solution to achieve state-of-the-art performance.