Search

Bruch's membrane (BM) segmentation on optical coherence tomography (OCT) is a pivotal step for the diagnosis and follow-up of age-related macular degeneration (AMD), one of the leading causes of blindness in the developed world. Automated BM segmentation methods exist, but they usually do not account for the anatomical coherence of the results, neither provide feedback on the confidence of the prediction. These factors limit the applicability of these systems in real-world scenarios. With this in mind, we propose an end-to-end deep learning method for automated BM segmentation in AMD patients. An Attention U-Net is trained to output a probability density function of the BM position, while taking into account the natural curvature of the surface. Besides the surface position, the method also estimates an A-scan wise uncertainty measure of the segmentation output. Subsequently, the A-scans with high uncertainty are interpolated using thin plate splines (TPS). We tested our method with ablation studies on an internal dataset with 138 patients covering all three AMD stages, and achieved a mean absolute localization error of 4.10 [Formula: see text]m. In addition, the proposed segmentation method was compared against the state-of-the-art methods and showed a superior performance on an external publicly available dataset from a different patient cohort and OCT device, demonstrating strong generalization ability.

To investigate the therapeutic effect of intravitreal pegcetacoplan on the inhibition of photoreceptor (PR) loss and thinning in geographic atrophy (GA) on conventional spectral-domain OCT (SD-OCT) imaging by deep learning-based automated PR quantification.Post hoc analysis of a prospective, multicenter, randomized, sham (SM)-controlled, masked phase II trial investigating the safety and efficacy of pegcetacoplan for the treatment of GA because of age-related macular degeneration.Study eyes of 246 patients, randomized 1:1:1 to monthly (AM), bimonthly (AEOM), and SM treatment.We performed fully automated, deep learning-based segmentation of retinal pigment epithelium (RPE) loss and PR thickness on SD-OCT volumes acquired at baseline and months 2, 6, and 12. The difference in the change of PR loss area was compared among the treatment arms. Change in PR thickness adjacent to the GA borders and the entire 20° scanning area was compared between treatment arms.Square-root transformed PR loss area in μm or mm, PR thickness in μm, and PR loss/RPE loss ratio.A total of 31 556 B-scans of 644 SD-OCT volumes of 161 study eyes (AM 52, AEOM 54, SM 56) were evaluated from baseline to month 12. Comparison of the mean change in PR loss area revealed statistically significantly less growth in the AM group at months 2, 6, and 12 than in the SM group (-41 μm ± 219 vs. 77 μm ± 126; P = 0.0004; -5 μm ± 221 vs. 156 μm ± 139; P0.0001; 106 μm ± 400 vs. 283 μm ± 226; P = 0.0014). Photoreceptor thinning was significantly reduced under AM treatment compared with SM within the GA junctional zone, as well as throughout the 20° area. A trend toward greater inhibition of PR loss than RPE loss was observed under therapy.Distinct and reliable quantification of PR loss using deep learning-based algorithms offers an essential tool to evaluate therapeutic efficacy in slowing disease progression. Photoreceptor loss and thinning are reduced by intravitreal complement C3 inhibition. Automated quantification of PR loss/maintenance based on OCT images is an ideal approach to reliably monitor disease activity and therapeutic efficacy in GA management in clinical routine and regulatory trials.

To identify disease activity and effects of intravitreal pegcetacoplan treatment on the topographic progression of geographic atrophy (GA) secondary to age-related macular degeneration quantified in spectral-domain OCT (SD-OCT) by automated deep learning assessment.Retrospective analysis of a phase II clinical trial study evaluating pegcetacoplan in GA patients (FILLY, NCT02503332).SD-OCT scans of 57 eyes with monthly treatment, 46 eyes with every-other-month (EOM) treatment, and 53 eyes with sham injection from baseline and 12-month follow-ups were included, in a total of 312 scans.Retinal pigment epithelium loss, photoreceptor (PR) integrity, and hyperreflective foci (HRF) were automatically segmented using validated deep learning algorithms. Local progression rate (LPR) was determined from a growth model measuring the local expansion of GA margins between baseline and 1 year. For each individual margin point, the eccentricity to the foveal center, the progression direction, mean PR thickness, and HRF concentration in the junctional zone were computed. Mean LPR in disease activity and treatment effect conditioned on these properties were estimated by spatial generalized additive mixed-effect models.LPR of GA, PR thickness, and HRF concentration in μm.A total of 31,527 local GA margin locations were analyzed. LPR was higher for areas with low eccentricity to the fovea, thinner PR layer thickness, or higher HRF concentration in the GA junctional zone. When controlling for topographic and structural risk factors, we report on average a significantly lower LPR by -28.0% (95% confidence interval [CI], -42.8 to -9.4; P = 0.0051) and -23.9% (95% CI, -40.2 to -3.0; P = 0.027) for monthly and EOM-treated eyes, respectively, compared with sham.Assessing GA progression on a topographic level is essential to capture the pathognomonic heterogeneity in individual lesion growth and therapeutic response. Pegcetacoplan-treated eyes showed a significantly slower GA lesion progression rate compared with sham, and an even slower growth rate toward the fovea. This study may help to identify patient cohorts with faster progressing lesions, in which pegcetacoplan treatment would be particularly beneficial. Automated artificial intelligence-based tools will provide reliable guidance for the management of GA in clinical practice.

To perform an optical coherence tomography (OCT)-based analysis of geographic atrophy (GA) progression in patients treated with pegcetacoplan.Post hoc analysis of a phase 2 multicenter, randomized, sham-controlled trial.Manual annotation of retinal pigment epithelium (RPE), ellipsoid zone (EZ), and external limiting membrane (ELM) loss was performed on OCT volumes from baseline and month 12 from the phase 2 FILLY trial of intravitreal pegcetacoplan for the treatment of GA secondary to age-related macular degeneration.Correlation of GA areas measured on fundus autofluorescence and OCT. Difference in square root transformed growth rates of RPE, EZ, and ELM loss between treatment groups (monthly injection [AM], injection every other month [AEOM], and sham [SM]).OCT volumes from 113 eyes of 113 patients (38 AM, 36 AEOM, and 39 SM) were included, resulting in 11 074 B-scans. The median growth of RPE loss was significantly slower in the AM group (0.158 [0.057-0.296]) than the SM group (0.255 [0.188-0.359], P = .014). Importantly, the growth of EZ loss was also significantly slower in the AM group (0.127 [0.041-0.247]) than the SM group (0.232 [0.130-0.349], P = .017). There was no significant difference in the growth of ELM loss between the treatment groups (P = .114).OCT imaging provided consistent results for GA growth compared with fundus autofluorescence. In addition to slower RPE atrophy progression in patients treated with pegcetacoplan, a significant reduction in EZ impairment was also identified by OCT, suggesting the use of OCT as a potentially more sensitive monitoring tool in GA therapy.

In this paper, we address the problem of automatic three-dimensional cephalometric analysis. Cephalometric analysis performed on lateral radiographs doesn’t fully exploit the structure of 3D objects due to projection onto the lateral plane. With the development of three-dimensional imaging techniques such as CT, several analysis methods have been proposed that extend to the 3D case. The analysis based on these methods is invariant to rotations and translations and can describe difficult skull deformation, where 2D cephalometry has no use. In this paper, we provide a wide overview of existing approaches for cephalometric landmark regression. Moreover, we perform a series of experiments with state of the art 3D convolutional neural network (CNN) based methods for keypoint regression: direct regression with CNN, heatmap regression and Softargmax regression. For the first time, we extensively evaluate the described methods and demonstrate their effectiveness in the estimation of Frankfort Horizontal and cephalometric points locations for patients with severe skull deformations. We demonstrate that Heatmap and Softargmax regression models provide sufficient regression error for medical applications (less than 4 mm). Moreover, the Softargmax model achieves 1.15° inclination error for the Frankfort horizontal. For the fair comparison with the prior art, we also report results projected on the lateral plane.

Segmentation of abdominal organs has been a comprehensive, yet unresolved, research field for many years. In the last decade, intensive developments in deep learning (DL) introduced new state-of-the-art segmentation systems. Despite outperforming the overall accuracy of existing systems, the effects of DL model properties and parameters on the performance are hard to interpret. This makes comparative analysis a necessary tool towards interpretable studies and systems. Moreover, the performance of DL for emerging learning approaches such as cross-modality and multi-modal semantic segmentation tasks has been rarely discussed. In order to expand the knowledge on these topics, the CHAOS – Combined (CT-MR) Healthy Abdominal Organ Segmentation challenge was organized in conjunction with the IEEE International Symposium on Biomedical Imaging (ISBI), 2019, in Venice, Italy. Abdominal organ segmentation from routine acquisitions plays an important role in several clinical applications, such as pre-surgical planning or morphological and volumetric follow-ups for various diseases. These applications require a certain level of performance on a diverse set of metrics such as maximum symmetric surface distance (MSSD) to determine surgical error-margin or overlap errors for tracking size and shape differences. Previous abdomen related challenges are mainly focused on tumor/lesion detection and/or classification with a single modality. Conversely, CHAOS provides both abdominal CT and MR data from healthy subjects for single and multiple abdominal organ segmentation. Five different but complementary tasks were designed to analyze the capabilities of participating approaches from multiple perspectives. The results were investigated thoroughly, compared with manual annotations and interactive methods. The analysis shows that the performance of DL models for single modality (CT / MR) can show reliable volumetric analysis performance (DICE: 0.98 ± 0.00 / 0.95 ± 0.01), but the best MSSD performance remains limited (21.89 ± 13.94 / 20.85 ± 10.63 mm). The performances of participating models decrease dramatically for cross-modality tasks both for the liver (DICE: 0.88 ± 0.15 MSSD: 36.33 ± 21.97 mm). Despite contrary examples on different applications, multi-tasking DL models designed to segment all organs are observed to perform worse compared to organ-specific ones (performance drop around 5%). Nevertheless, some of the successful models show better performance with their multi-organ versions. We conclude that the exploration of those pros and cons in both single vs multi-organ and cross-modality segmentations is poised to have an impact on further research for developing effective algorithms that would support real-world clinical applications. Finally, having more than 1500 participants and receiving more than 550 submissions, another important contribution of this study is the analysis on shortcomings of challenge organizations such as the effects of multiple submissions and peeking phenomenon. © 2020 Elsevier B.V., 116E133, BIDEB-2214 College of Environmental Science and Forestry, State University of New York, ESF: 1059B191701102, BIDEB-2219, ZS/2016/08/80646 Türkiye Bilimsel ve Teknolojik Araştirma Kurumu, TÜBITAK, The organizers would like to thank Ivana Isgum and Tom Vercauteren in the challenge committee of ISBI 2019 for their guidance and support. We express our gratitude to supporting organizations of the grand-challenge.org platform. We thank Esranur Kazaz, Umut Baran Ekinci, Ece K?se, Fabian Isensee, David V?lgyes, and Javier Coronel for their contributions. Last but not least, our special thanks go to Ludmila I. Kuncheva for her valuable contributions. This work is supported by Scientific and Technological Research Council of Turkey (TUBITAK) ARDEB-EEEAG under grant number 116E133 and TUBITAK BIDEB-2214 International Doctoral Research Fellowship Programme. The work of P. Ernst, S. Chatterjee, O. Speck and, A. N?rnberger was conducted within the context of the International Graduate School MEMoRIAL at OvGU Magdeburg, Germany, supported by ESF (project no. ZS/2016/08/80646). The work of S. Aslan within the context of Ca? Foscari University of Venice is supported by under TUBITAK BIDEB-2219 grant no 1059B191701102., This work is supported by Scientific and Technological Research Council of Turkey (TUBITAK) ARDEB-EEEAG under grant number 116E133 and TUBITAK BIDEB-2214 International Doctoral Research Fellowship Programme. The work of P. Ernst, S. Chatterjee, O. Speck and, A. Nürnberger was conducted within the context of the International Graduate School MEMoRIAL at OvGU Magdeburg, Germany, supported by ESF (project no. ZS/2016/08/80646). The work of S. Aslan within the context of Ca’ Foscari University of Venice is supported by under TUBITAK BIDEB-2219 grant no 1059B191701102.

In medical imaging, there are clinically relevant segmentation tasks where the output mask is a projection to a subset of input image dimensions. In this work, we propose a novel convolutional neural network architecture that can effectively learn to produce a lower-dimensional segmentation mask than the input image. The network restores encoded representation only in a subset of input spatial dimensions and keeps the representation unchanged in the others. The newly proposed projective skip-connections allow linking the encoder and decoder in a UNet-like structure. We evaluated the proposed method on two clinically relevant tasks in retinal Optical Coherence Tomography (OCT): geographic atrophy and retinal blood vessel segmentation. The proposed method outperformed the current state-of-the-art approaches on all the OCT datasets used, consisting of 3D volumes and corresponding 2D en-face masks. The proposed architecture fills the methodological gap between image classification and ND image segmentation.

The segmentation of brain tumors in multimodal MRIs is one of the most challenging tasks in medical image analysis. The recent state of the art algorithms solving this task are based on machine learning approaches and deep learning in particular. The amount of data used for training such models and its variability is a keystone for building an algorithm with high representation power.

A fast procedure for generation of regular tetrahedral finite element mesh for objects with complex shape cavities is proposed. The procedure like LBIE-Mesher can generate tetrahedral meshes for the volume interior to a polygonal surface, or for an interval volume between two surfaces having a complex shape and defined in STL-format. This procedure consists of several stages: generation of a regular tetrahedral mesh that fills the volume of the required object; generation of clipping for the uniform grid parts by a boundary surface; shifting vertices of the boundary layer to align onto the surface.We present a sequential and parallel implementation of the algorithm and compare their performance with existing generators of tetrahedral grids such as TetGen, NETGEN, and CGAL. The current version of the algorithm using the mobile GPU is about 5 times faster than NETGEN. The source code of the developed software is available on GitHub.

Personalizing the three-dimensional atlas model (template) of an organ under analysis based on the data of patient’s three-dimensional examination is an important task for modern digital medicine. The template should be represented by a set of points Y and marked with the key points for the model (landmarks). The template is personalized by the non-rigid registration of the set Y with the set of points X that represents patient’s tomogram. Presently, the coherent point drift (CPD) is the most popular method for solving the problem of non-rigid alignment. In this paper, we propose and explore an approach that substantially improves the CPD result in the problem of automatic registration of cephalometric points (CPs). The proposed algorithm remains robust in the presence of significant skull deformations. Traditionally, 3D cephalometry uses the geometric descriptor of a CP, which refines the position of the point on the bone surface. However, the result of applying the descriptor depends significantly on the accuracy of its anatomical basis reconstruction. The proposed approach solves this problem by clarifying the basis of geometric descriptors, the supporting elements of which are orbital planes and the Frankfurt (orbital-ear) horizontal. For this purpose, additionally marked points of the orbitals (YO) are included into the template Y. Once Y and X are aligned by the CPD method, the plane positions of the orbitals are refined by solving a linear regression problem on the subsets of YO for the left and right orbitals. Cases of refinement with and without use of Tikhonov regularization (ridge-regression) are analyzed. The effect of the increase in the cardinality of the set X relative to Y on the registration accuracy is investigated. It is found that the condition |X| < |Y| has a negative effect on the accuracy, which increases when the cardinality of X relative to Y decreases. The refinement of CPs by the geometric descriptor is carried out on tomogram data in the region around CPs that is found by the CPD method. The dimensions of this region along three coordinates are determined by the anatomical domain of a particular CP descriptor. The quality of the algorithm is measured by the Euclidean distance between hand-marked and automatically found points. The template Y for the algorithm is built on a trauma-deformed skull. The algorithm is quantitatively verified by registering the CPs of orbitals and cheekbones from the data of four tomograms for the deformed skull. A key feature and source of high accuracy of the approach is the use of linear regression with Tikhonov regularization to refine it. As a result, 81.25% of the points found fall within a radius of 2 mm from the hand-marked points and 100% of the points fall within a radius of 4 mm.

MRI analysis takes central position in brain tumor diagnosis and treatment, thus its precise evaluation is crucially important. However, its 3D nature imposes several challenges, so the analysis is often performed on 2D projections that reduces the complexity, but increases bias. On the other hand, time consuming 3D evaluation, like segmentation, is able to provide precise estimation of a number of valuable spatial characteristics, giving us understanding about the course of the disease.

In this paper, we evaluate the performance of the Intel Distribution of OpenVINO toolkit in practical solving of the problem of automatic three-dimensional Cephalometric analysis using deep learning methods. This year, the authors proposed an approach to the detection of cephalometric landmarks from CT-tomography data, which is resistant to skull deformities and use convolutional neural networks (CNN). Resistance to deformations is due to the initial detection of 4 points that are basic for the parameterization of the skull shape. The approach was explored on CNN for three architectures. A record regression accuracy in comparison with analogs was obtained. This paper evaluates the perfor- mance of decision making for the trained CNN-models at the inference stage. For a comparative study, the computing environments PyTorch and Intel Distribution of OpenVINO were selected, and 2 of 3 CNN architectures: based on VGG for regression of cephalometric landmarks and an Hourglass-based model, with the RexNext backbone for the land- marks heatmap regression. The experimental dataset was consist of 20 CT of patients with acquired craniomaxillofacial deformities and was in- clude pre- and post-operative CT scans whose format is 800x800x496 with voxel spacing of 0.2x0.2x0.2 mm. Using OpenVINO showed a great increase in performance over the PyTorch, with inference speedup from 13 to 16 times for a Direct Regression model and from 3.5 to 3.8 times for a more complex and precise Hourglass model.