Prediction of Stroke Infarct Growth Rates by Baseline Perfusion Imaging.

BACKGROUND AND PURPOSE: Computed tomography perfusion imaging allows estimation of tissue status in patients with acute ischemic stroke. We aimed to improve prediction of the final infarct and individual infarct growth rates using a deep learning approach. METHODS: We trained a deep neural network to predict the final infarct volume in patients with acute stroke presenting with large vessel occlusions based on the native computed tomography perfusion images, time to reperfusion and reperfusion status in a derivation cohort (MR CLEAN trial [Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands]). The model was internally validated in a 5-fold cross-validation and externally in an independent dataset (CRISP study [CT Perfusion to Predict Response to Recanalization in Ischemic Stroke Project]). We calculated the mean absolute difference between the predictions of the deep learning model and the final infarct volume versus the mean absolute difference between computed tomography perfusion imaging processing by RAPID software (iSchemaView, Menlo Park, CA) and the final infarct volume. Next, we determined infarct growth rates for every patient. RESULTS: We included 127 patients from the MR CLEAN (derivation) and 101 patients of the CRISP study (validation). The deep learning model improved final infarct volume prediction compared with the RAPID software in both the derivation, mean absolute difference 34.5 versus 52.4 mL, and validation cohort, 41.2 versus 52.4 mL (P<0.01). We obtained individual infarct growth rates enabling the estimation of final infarct volume based on time and grade of reperfusion. CONCLUSIONS: We validated a deep learning-based method which improved final infarct volume estimations compared with classic computed tomography perfusion imaging processing. In addition, the deep learning model predicted individual infarct growth rates which could enable the introduction of tissue clocks during the management of acute stroke.

3D Tendon Strain Estimation Using High-frequency Volumetric Ultrasound Images: A Feasibility Study.

Estimation of strain in tendons for tendinopathy assessment is a hot topic within the sports medicine community. It is believed that, if accurately estimated, existing treatment and rehabilitation protocols can be improved and presymptomatic abnormalities can be detected earlier. State-of-the-art studies present inaccurate and highly variable strain estimates, leaving this problem without solution. Out-of-plane motion, present when acquiring two-dimensional (2D) ultrasound (US) images, is a known problem and may be responsible for such errors. This work investigates the benefit of high-frequency, three-dimensional (3D) US imaging to reduce errors in tendon strain estimation. Volumetric US images were acquired in silico, in vitro, and ex vivo using an innovative acquisition approach that combines the acquisition of 2D high-frequency US images with a mechanical guided system. An affine image registration method was used to estimate global strain. 3D strain estimates were then compared with ground-truth values and with 2D strain estimates. The obtained results for in silico data showed a mean absolute error (MAE) of 0.07%, 0.05%, and 0.27% for 3D estimates along axial, lateral direction, and elevation direction and a respective MAE of 0.21% and 0.29% for 2D strain estimates. Although 3D could outperform 2D, this does not occur in in vitro and ex vivo settings, likely due to 3D acquisition artifacts. Comparison against the state-of-the-art methods showed competitive results. The proposed work shows that 3D strain estimates are more accurate than 2D estimates but acquisition of appropriate 3D US images remains a challenge.

Convexity-constrained and nonnegativity-constrained spherical factorization in diffusion-weighted imaging.

Diffusion-weighted imaging (DWI) facilitates probing neural tissue structure non-invasively by measuring its hindrance to water diffusion. Analysis of DWI is typically based on generative signal models for given tissue geometry and microstructural properties. In this work, we generalize multi-tissue spherical deconvolution to a blind source separation problem under convexity and nonnegativity constraints. This spherical factorization approach decomposes multi-shell DWI data, represented in the basis of spherical harmonics, into tissue-specific orientation distribution functions and corresponding response functions, without assuming the latter as known thus fully unsupervised. In healthy human brain data, the resulting components are associated with white matter fibres, grey matter, and cerebrospinal fluid. The factorization results are on par with state-of-the-art supervised methods, as demonstrated also in Monte-Carlo simulations evaluating accuracy and precision of the estimated response functions and orientation distribution functions of each component. In animal data and in the presence of oedema, the proposed factorization is able to recover unseen tissue structure, solely relying on DWI. As such, our method broadens the applicability of spherical deconvolution techniques to exploratory analysis of tissue structure in data where priors are uncertain or hard to define.

Modeling 3D facial shape from DNA.

Human facial diversity is substantial, complex, and largely scientifically unexplained. We used spatially dense quasi-landmarks to measure face shape in population samples with mixed West African and European ancestry from three locations (United States, Brazil, and Cape Verde). Using bootstrapped response-based imputation modeling (BRIM), we uncover the relationships between facial variation and the effects of sex, genomic ancestry, and a subset of craniofacial candidate genes. The facial effects of these variables are summarized as response-based imputed predictor (RIP) variables, which are validated using self-reported sex, genomic ancestry, and observer-based facial ratings (femininity and proportional ancestry) and judgments (sex and population group). By jointly modeling sex, genomic ancestry, and genotype, the independent effects of particular alleles on facial features can be uncovered. Results on a set of 20 genes showing significant effects on facial features provide support for this approach as a novel means to identify genes affecting normal-range facial features and for approximating the appearance of a face from genetic markers.

Global tractography of multi-shell diffusion-weighted imaging data using a multi-tissue model.

Diffusion-weighted imaging and tractography provide a unique, non-invasive technique to study the macroscopic structure and connectivity of brain white matter in vivo. Global tractography methods aim at reconstructing the full-brain fiber configuration that best explains the measured data, based on a generative signal model. In this work, we incorporate a multi-shell multi-tissue model based on spherical convolution, into a global tractography framework, which allows to deal with partial volume effects. The required tissue response functions can be estimated from and hence calibrated to the data. The resulting track reconstruction is quantitatively related to the apparent fiber density in the data. In addition, the fiber orientation distribution for white matter and the volume fractions of gray matter and cerebrospinal fluid are produced as ancillary results. Validation results on simulated data demonstrate that this data-driven approach improves over state-of-the-art streamline and global tracking methods, particularly in the valid connection rate. Results in human brain data correspond to known white matter anatomy and show improved modeling of partial voluming. This work is an important step toward detecting and quantifying white matter changes and connectivity in healthy subjects and patients.

An investigation of matching symmetry in the human pinnae with possible implications for 3D ear recognition and sound localization.

The human external ears, or pinnae, have an intriguing shape and, like most parts of the human external body, bilateral symmetry is observed between left and right. It is a well-known part of our auditory sensory system and mediates the spatial localization of incoming sounds in 3D from monaural cues due to its shape-specific filtering as well as binaural cues due to the paired bilateral locations of the left and right ears. Another less broadly appreciated aspect of the human pinna shape is its uniqueness from one individual to another, which is on the level of what is seen in fingerprints and facial features. This makes pinnae very useful in human identification, which is of great interest in biometrics and forensics. Anatomically, the type of symmetry observed is known as matching symmetry, with structures present as separate mirror copies on both sides of the body, and in this work we report the first such investigation of the human pinna in 3D. Within the framework of geometric morphometrics, we started by partitioning ear shape, represented in a spatially dense way, into patterns of symmetry and asymmetry, following a two-factor anova design. Matching symmetry was measured in all substructures of the pinna anatomy. However, substructures that 'stick out' such as the helix, tragus, and lobule also contained a fair degree of asymmetry. In contrast, substructures such as the conchae, antitragus, and antihelix expressed relatively stronger degrees of symmetric variation in relation to their levels of asymmetry. Insights gained from this study were injected into an accompanying identification setup exploiting matching symmetry where improved performance is demonstrated. Finally, possible implications of the results in the context of ear recognition as well as sound localization are discussed.

Track orientation density imaging (TODI) and track orientation distribution (TOD) based tractography.

Ever since the introduction of the concept of fiber tractography, methods to generate better and more plausible tractograms have become available. Many modern methods can handle complex fiber architecture and take on a probabilistic approach to account for different sources of uncertainty. The resulting tractogram from any such method typically represents a finite random sample from a complex distribution of possible tracks. Generating a higher amount of tracks allows for a more accurate depiction of the underlying distribution. The recently proposed method of track-density imaging (TDI) allows to capture the spatial distribution of a tractogram. In this work, we propose an extension of TDI towards the 5D spatio-angular domain, which we name track orientation density imaging (TODI). The proposed method aims to capture the full track orientation distribution (TOD). Just as the TDI map, the TOD is amenable to spatial super-resolution (or even sub-resolution), but in addition also to angular super-resolution. Through experiments on in vivo human subject data, an in silico numerical phantom and a challenging tractography phantom, we found that the TOD presents an increased amount of regional spatio-angular consistency, as compared to the fiber orientation distribution (FOD) from constrained spherical deconvolution (CSD). Furthermore, we explain how the amplitude of the TOD of a short-tracks distribution (i.e. where the track length is limited) can be interpreted as a measure of track-like local support (TLS). This in turn motivated us to explore the idea of TOD-based fiber tractography. In such a setting, the short-tracks TOD is able to guide a track along directions that are more likely to correspond to continuous structure over a longer distance. This powerful concept is shown to greatly robustify targeted as well as whole-brain tractography. We conclude that the TOD is a versatile tool that can be cast in many different roles and scenarios in the expanding domain of fiber tractography based methods and their applications.

Sexual dimorphism in multiple aspects of 3D facial symmetry and asymmetry defined by spatially dense geometric morphometrics.

Accurate measurement of facial sexual dimorphism is useful to understanding facial anatomy and specifically how faces influence, and have been influenced by, sexual selection. An important facial aspect is the display of bilateral symmetry, invoking the need to investigate aspects of symmetry and asymmetry separately when examining facial shape. Previous studies typically employed landmarks that provided only a sparse facial representation, where different landmark choices could lead to contrasting outcomes. Furthermore, sexual dimorphism is only tested as a difference of sample means, which is statistically the same as a difference in population location only. Within the framework of geometric morphometrics, we partition facial shape, represented in a spatially dense way, into patterns of symmetry and asymmetry, following a two-factor anova design. Subsequently, we investigate sexual dimorphism in symmetry and asymmetry patterns separately, and on multiple aspects, by examining (i) population location differences as well as differences in population variance-covariance; (ii) scale; and (iii) orientation. One important challenge in this approach is the proportionally high number of variables to observations necessitating the implementation of permutational and computationally feasible statistics. In a sample of gender-matched young adults (18-25 years) with self-reported European ancestry, we found greater variation in male faces than in women for all measurements. Statistically significant sexual dimorphism was found for the aspect of location in both symmetry and asymmetry (directional asymmetry), for the aspect of scale only in asymmetry (magnitude of fluctuating asymmetry) and, in contrast, for the aspect of orientation only in symmetry. Interesting interplays with hypotheses in evolutionary and developmental biology were observed, such as the selective nature of the force underpinning sexual dimorphism and the genetic independence of the structural patterns of fluctuating asymmetry. Additionally, insights into growth patterns of the soft tissue envelope of the face and underlying skull structure can also be obtained from the results.

Dysmorphometrics: the modelling of morphological abnormalities.

BACKGROUND: The study of typical morphological variations using quantitative, morphometric descriptors has always interested biologists in general. However, unusual examples of form, such as abnormalities are often encountered in biomedical sciences. Despite the long history of morphometrics, the means to identify and quantify such unusual form differences remains limited. METHODS: A theoretical concept, called dysmorphometrics, is introduced augmenting current geometric morphometrics with a focus on identifying and modelling form abnormalities. Dysmorphometrics applies the paradigm of detecting form differences as outliers compared to an appropriate norm. To achieve this, the likelihood formulation of landmark superimpositions is extended with outlier processes explicitly introducing a latent variable coding for abnormalities. A tractable solution to this augmented superimposition problem is obtained using Expectation-Maximization. The topography of detected abnormalities is encoded in a dysmorphogram. RESULTS: We demonstrate the use of dysmorphometrics to measure abrupt changes in time, asymmetry and discordancy in a set of human faces presenting with facial abnormalities. CONCLUSION: The results clearly illustrate the unique power to reveal unusual form differences given only normative data with clear applications in both biomedical practice & research.

MeVisLab-OpenVR prototyping platform for virtual reality medical applications.

PURPOSE: Virtual reality (VR) can provide an added value for diagnosis and/or intervention planning. Several VR software implementations have been proposed but they are often application dependent. Previous attempts for a more generic solution incorporating VR in medical prototyping software (MeVisLab) were still lacking functionality precluding easy and flexible development. METHODS: We propose an alternative solution that uses rendering to a graphical processing unit (GPU) texture to enable rendering arbitrary Open Inventor scenes in a VR context. It facilitates flexible development of user interaction and rendering of more complex scenes involving multiple objects. We tested the platform in planning a transcatheter cardiac stent placement procedure. RESULTS: This approach proved to enable development of a particular implementation that facilitates planning of percutaneous treatment of a sinus venosus atrial septal defect. The implementation showed it is intuitive to plan and verify the procedure using VR. CONCLUSION: An alternative implementation for linking OpenVR with MeVisLab is provided that offers more flexible development of VR prototypes which can facilitate further clinical validation of this technology in various medical disciplines.

Hierarchical spectral clustering reveals brain size and shape changes in asymptomatic carriers of C9orf72.

Traditional methods for detecting asymptomatic brain changes in neurodegenerative diseases such as Alzheimer's disease or frontotemporal degeneration typically evaluate changes in volume at a predefined level of granularity, e.g. voxel-wise or in a priori defined cortical volumes of interest. Here, we apply a method based on hierarchical spectral clustering, a graph-based partitioning technique. Our method uses multiple levels of segmentation for detecting changes in a data-driven, unbiased, comprehensive manner within a standard statistical framework. Furthermore, spectral clustering allows for detection of changes in shape along with changes in size. We performed tensor-based morphometry to detect changes in the Genetic Frontotemporal dementia Initiative asymptomatic and symptomatic frontotemporal degeneration mutation carriers using hierarchical spectral clustering and compared the outcome to that obtained with a more conventional voxel-wise tensor- and voxel-based morphometric analysis. In the symptomatic groups, the hierarchical spectral clustering-based method yielded results that were largely in line with those obtained with the voxel-wise approach. In asymptomatic C9orf72 expansion carriers, spectral clustering detected changes in size in medial temporal cortex that voxel-wise methods could only detect in the symptomatic phase. Furthermore, in the asymptomatic and the symptomatic phases, the spectral clustering approach detected changes in shape in the premotor cortex in C9orf72. In summary, the present study shows the merit of hierarchical spectral clustering for data-driven segmentation and detection of structural changes in the symptomatic and asymptomatic stages of monogenic frontotemporal degeneration.

Theoretical analysis and experimental validation of volume bias of soft Dice optimized segmentation maps in the context of inherent uncertainty.

The clinical interest is often to measure the volume of a structure, which is typically derived from a segmentation. In order to evaluate and compare segmentation methods, the similarity between a segmentation and a predefined ground truth is measured using popular discrete metrics, such as the Dice score. Recent segmentation methods use a differentiable surrogate metric, such as soft Dice, as part of the loss function during the learning phase. In this work, we first briefly describe how to derive volume estimates from a segmentation that is, potentially, inherently uncertain or ambiguous. This is followed by a theoretical analysis and an experimental validation linking the inherent uncertainty to common loss functions for training CNNs, namely cross-entropy and soft Dice. We find that, even though soft Dice optimization leads to an improved performance with respect to the Dice score and other measures, it may introduce a volume bias for tasks with high inherent uncertainty. These findings indicate some of the method's clinical limitations and suggest doing a closer ad-hoc volume analysis with an optional re-calibration step.

Prediction of final infarct volume from native CT perfusion and treatment parameters using deep learning.

CT Perfusion (CTP) imaging has gained importance in the diagnosis of acute stroke. Conventional perfusion analysis performs a deconvolution of the measurements and thresholds the perfusion parameters to determine the tissue status. We pursue a data-driven and deconvolution-free approach, where a deep neural network learns to predict the final infarct volume directly from the native CTP images and metadata such as the time parameters and treatment. This would allow clinicians to simulate various treatments and gain insight into predicted tissue status over time. We demonstrate on a multicenter dataset that our approach is able to predict the final infarct and effectively uses the metadata. An ablation study shows that using the native CTP measurements instead of the deconvolved measurements improves the prediction.

Robust motion correction for cardiac T1 and ECV mapping using a T1 relaxation model approach.

T1 and ECV mapping are quantitative methods for myocardial tissue characterization using cardiac MRI, and are highly relevant for the diagnosis of diffuse myocardial diseases. Since the maps are calculated pixel-by-pixel from a set of MRI images with different T1-weighting, it is critical to assure exact spatial correspondence between these images. However, in practice, different sources of motion e.g. cardiac motion, respiratory motion or patient motion, hamper accurate T1 and ECV calculation such that retrospective motion correction is required. We propose a new robust non-rigid registration framework combining a data-driven initialization with a model-based registration approach, which uses a model for T1 relaxation to avoid direct registration of images with highly varying contrast. The registration between native T1 and enhanced T1 to obtain a motion free ECV map is also calculated using information from T1 model-fitting. The method was validated on three datasets recorded with two substantially different acquisition protocols (MOLLI (dataset 1 (n=15) and dataset 2 (n=29)) and STONE (dataset 3 (n = 210))), one in breath-hold condition and one free-breathing. The average Dice coefficient increased from 72.6 ±â€¯12.1% to 82.3 ±â€¯7.4% (P < 0.05) and mean boundary error decreased from 2.91 ±â€¯1.51mm to 1.62 ±â€¯0.80mm (P < 0.05) for motion correction in a single T1-weighted image sequence (3 datasets) while average Dice coefficient increased from 63.4 ±â€¯22.5% to 79.2 ±â€¯8.5% (P < 0.05) and mean boundary error decreased from 3.26 ±â€¯2.64mm to 1.77 ±â€¯0.86mm (P < 0.05) between native and enhanced sequences (dataset 1 and 2). Overall, the native T1 SD error decreased from 67.32 ±â€¯32.57ms to 58.11 ±â€¯21.59ms (P < 0.05), enhanced SD error from 30.15 ±â€¯25ms to 22.74 ±â€¯8.94ms (P < 0.05) and ECV SD error from 10.08 ±â€¯9.59% to 5.42 ±â€¯3.21% (P < 0.05) (dataset 1 and 2).

Facial recognition from DNA using face-to-DNA classifiers.

Facial recognition from DNA refers to the identification or verification of unidentified biological material against facial images with known identity. One approach to establish the identity of unidentified biological material is to predict the face from DNA, and subsequently to match against facial images. However, DNA phenotyping of the human face remains challenging. Here, another proof of concept to biometric authentication is established by using multiple face-to-DNA classifiers, each classifying given faces by a DNA-encoded aspect (sex, genomic background, individual genetic loci), or by a DNA-inferred aspect (BMI, age). Face-to-DNA classifiers on distinct DNA aspects are fused into one matching score for any given face against DNA. In a globally diverse, and subsequently in a homogeneous cohort, we demonstrate preliminary, but substantial true (83%, 80%) over false (17%, 20%) matching in verification mode. Consequences of future efforts include forensic applications, necessitating careful consideration of ethical and legal implications for privacy in genomic databases.

Calculated moment-arm and muscle-tendon lengths during gait differ substantially using MR based versus rescaled generic lower-limb musculoskeletal models.

Biomechanical analysis of gait relies on the use of lower-limb musculoskeletal models. Most models are based on a generic model which takes into account the subject's skeletal dimensions by isotropic or anisotropic rescaling. Alternatively, personalized models can be built based on information from magnetic resonance (MR) images. We have studied the effect of these approaches on muscle-tendon lengths (MTLs) and moment-arm lengths (MALs) for 16 major muscles of the lower limb of a normal adult during both normal and pathologic gait. For most muscles, the MTL and MAL calculated using the rescaled generic models showed high correlation values, but large offsets when compared to values calculated using personalized models. MTL and MAL differences with the personalized model are only slightly smaller for an anisotropic than for an isotropic rescaled model. Gait kinematics influenced the observed inter-model differences and correlations due to an altered range of joint angles in both gait patterns. In conclusion, both generic rescaling methods failed to accurately estimate absolute values for MTL and MAL calculated using the personalized model. However, the magnitude of MTL and MAL changes during normal and pathologic gait corresponded between all three models for most muscles. Since rescaling depends strongly on modelling assumptions and cannot fully take into account subject-specific musculoskeletal geometry, interpretation of MTL and MAL even in normal adult subjects requires extreme caution.

Personalized MR-based musculoskeletal models compared to rescaled generic models in the presence of increased femoral anteversion: effect on hip moment arm lengths.

Advanced biomechanical analysis of muscle function during gait relies on the use of a musculoskeletal model. In clinical practice, personalization of the model is usually limited to rescaling a generic model to approximate the patient's anthropometry, even in the presence of bony deformities, as in the case of cerebral palsy (CP). However, the current state of the art in biomechanics allows highly detailed subject-specific models to be built based on magnetic resonance (MR) images. We hypothesized that moment arm length (MAL) calculations from MR-based models would be more accurate than those from rescaled generic musculoskeletal models. Our study compared hip muscle MAL estimated by (1) a personalized model based on full-leg MR scans and (2) a rescaled generic model of both lower limbs in six children presenting with increased femoral anteversion. Personalized MR-based models were created using a custom-built workflow. Rescaled generic models were created based on three-dimensional positions of anatomical markers measured during a standing trial. For all 12 lower limb models, the hip flexion, adduction and rotation MAL of 13 major muscles were analyzed over a physiological range of hip motion using Software for interactive musculoskeletal modelling (SIMM) (Motion Analysis Corporation, USA). Our results showed that rescaled generic models, which do not take into account the subject's femoral geometry, overestimate MAL for hip flexion, extension, adduction, abduction and external rotation, but underestimate MAL for hip internal rotation. The differences in MAL introduced by taking the aberrant femoral geometry into account in the MR-based model were consistent with major gait characteristics presented in CP patients.

Skull reconstruction planning transfer to the operation room by thin metallic templates: clinical results.

INTRODUCTION: Craniofacial malformations implicate a risk of medical complications and a negative psychological impact on the patient. In order to correct functional and aesthetic aspects of these malformations, skull reconstruction is required. Because of the complexity of the surgery, pre-operative planning is unavoidable. Current and previously developed planning environments often lack the opportunity to transfer the simulated surgery to the operation room on a cheap but accurate, and easy to handle basis. MATERIALS AND METHODS: This study applies an automated filter procedure, implemented in Matlab, to generate a set of adapted contours from which a surface mesh can be directly deduced. Skull reconstruction planning is performed on the generated outer bone surface model. For each resected/osteotomized bone part, the presented semi-automatic Matlab procedure generates surface based bone cutting guides, also denoted bone segment templates. Autoclaved aluminium templates transfer the surgical plan to the operation room. RESULTS: The clinical feasibility is demonstrated by the successful pre-operative planning and surgical correction of three skull reconstruction cases in which the proposed procedure leads to considerable reduction in surgery time and good results. CONCLUSION: A cost-efficient and planning-environment-independent solution is generated for an accurate and fast transfer of a complex cranial surgery plan to the operation room.

Genome-wide mapping of global-to-local genetic effects on human facial shape.

Genome-wide association scans of complex multipartite traits like the human face typically use preselected phenotypic measures. Here we report a data-driven approach to phenotyping facial shape at multiple levels of organization, allowing for an open-ended description of facial variation while preserving statistical power. In a sample of 2,329 persons of European ancestry, we identified 38 loci, 15 of which replicated in an independent European sample (n = 1,719). Four loci were completely new. For the others, additional support (n = 9) or pleiotropic effects (n = 2) were found in the literature, but the results reported here were further refined. All 15 replicated loci highlighted distinctive patterns of global-to-local genetic effects on facial shape and showed enrichment for active chromatin elements in human cranial neural crest cells, suggesting an early developmental origin of the facial variation captured. These results have implications for studies of facial genetics and other complex morphological traits.

ISLES 2016 and 2017-Benchmarking Ischemic Stroke Lesion Outcome Prediction Based on Multispectral MRI.

Performance of models highly depend not only on the used algorithm but also the data set it was applied to. This makes the comparison of newly developed tools to previously published approaches difficult. Either researchers need to implement others' algorithms first, to establish an adequate benchmark on their data, or a direct comparison of new and old techniques is infeasible. The Ischemic Stroke Lesion Segmentation (ISLES) challenge, which has ran now consecutively for 3 years, aims to address this problem of comparability. ISLES 2016 and 2017 focused on lesion outcome prediction after ischemic stroke: By providing a uniformly pre-processed data set, researchers from all over the world could apply their algorithm directly. A total of nine teams participated in ISLES 2015, and 15 teams participated in ISLES 2016. Their performance was evaluated in a fair and transparent way to identify the state-of-the-art among all submissions. Top ranked teams almost always employed deep learning tools, which were predominately convolutional neural networks (CNNs). Despite the great efforts, lesion outcome prediction persists challenging. The annotated data set remains publicly available and new approaches can be compared directly via the online evaluation system, serving as a continuing benchmark (www.isles-challenge.org).