Search

NeSy4VRD is a multifaceted resource designed to support the development of neurosymbolic AI (NeSy) research. NeSy4VRD re-establishes public access to the images of the VRD dataset and couples them with an extensively revised, quality-improved version of the VRD visual relationship annotations. Crucially, NeSy4VRD provides a well-aligned, companion OWL ontology that describes the dataset domain.It comes with open source infrastructure that provides comprehensive support for extensibility of the annotations (which, in turn, facilitates extensibility of the ontology), and open source code for loading the annotations to/from a knowledge graph. We are contributing NeSy4VRD to the computer vision, NeSy and Semantic Web communities to help foster more NeSy research using OWL-based knowledge graphs.

U-Net has been the go-to architecture for medical image segmentation tasks, however computational challenges arise when extending the U-Net architecture to 3D images. We propose the Implicit U-Net architecture that adapts the efficient Implicit Representation paradigm to supervised image segmentation tasks. By combining a convolutional feature extractor with an implicit localization network, our implicit U-Net has 40% less parameters than the equivalent U-Net. Moreover, we propose training and inference procedures to capitalize sparse predictions. When comparing to an equivalent fully convolutional U-Net, Implicit U-Net reduces by approximately 30% inference and training time as well as training memory footprint while achieving comparable results in our experiments with two different abdominal CT scan datasets., 11 pages, 4 figures, Accepted MIUA 2022

Image demonstrating visual assessment and comparison between automated segmentation and manual segmentation. (PDF 210 kb)

The availability of training data for supervision is a frequently encountered bottleneck of medical image analysis methods. While typically established by a clinical expert rater, the increase in acquired imaging data renders traditional pixel-wise segmentations less feasible. In this paper, we examine the use of a crowdsourcing platform for the distribution of super-pixel weak annotation tasks and collect such annotations from a crowd of non-expert raters. The crowd annotations are subsequently used for training a fully convolutional neural network to address the problem of fetal brain segmentation in T2-weighted MR images. Using this approach we report encouraging results compared to highly targeted, fully supervised methods and potentially address a frequent problem impeding image analysis research.

We present an automated technique using noise-based level-set methods and non-rigid registration for myocardial detection as a basis for perfusion quantification from cardiac magnetic resonance (CMR) images. We studied 27 patients undergoing contrast-enhanced CMR imaging (1.5T) at rest and during adenosine stress, and validated our technique against conventional manual analysis both directly and using quantitative coronary angiography (QCA) as reference. Contrast enhancement time-curves were constructed and used to calculate a number of perfusion indices. Measured segmental pixel intensities in each frame correlated highly with manual analysis (r=0.95). Bland-Altman analysis showed small biases (1.3 at rest; 0.0 at stress) and narrow limits of agreement (±13 at rest; ±14 at stress). The derived perfusion indices showed the same diagnostic accuracy as manual analysis (AUC up to 0.71 vs. 0.70). These results indicate that our automated technique allows fast detection of myocardial ROIs and quantification of perfusion abnormalities as accurately as manual analysis.

Myocardial perfusion quantification by means of Contrast-Enhanced Cardiac Magnetic Resonance images relies on time consuming frame-by-frame manual tracing of regions of interest. In this Thesis, a novel automated technique for myocardial segmentation and non-rigid registration as a basis for perfusion quantification is presented. The proposed technique is based on three steps: reference frame selection, myocardial segmentation and non-rigid registration. In the first step, the reference frame in which both endo- and epicardial segmentation will be performed is chosen. Endocardial segmentation is achieved by means of a statistical region-based level-set technique followed by a curvature-based regularization motion. Epicardial segmentation is achieved by means of an edge-based level-set technique followed again by a regularization motion. To take into account the changes in position, size and shape of myocardium throughout the sequence due to out of plane respiratory motion, a non-rigid registration algorithm is required. The proposed non-rigid registration scheme consists in a novel multiscale extension of the normalized cross-correlation algorithm in combination with level-set methods. The myocardium is then divided into standard segments. Contrast enhancement curves are computed measuring the mean pixel intensity of each segment over time, and perfusion indices are extracted from each curve. The overall approach has been tested on synthetic and real datasets. For validation purposes, the sequences have been manually traced by an experienced interpreter, and contrast enhancement curves as well as perfusion indices have been computed. Comparisons between automatically extracted and manually obtained contours and enhancement curves showed high inter-technique agreement. Comparisons of perfusion indices computed using both approaches against quantitative coronary angiography and visual interpretation demonstrated that the two technique have similar diagnostic accuracy. In conclusion, the proposed technique allows fast, automated and accurate measurement of intra-myocardial contrast dynamics, and may thus address the strong clinical need for quantitative evaluation of myocardial perfusion.

We developed a method for automated quantification of myocardial perfusion from cardiac magnetic resonance (CMR) images. Our approach uses region-based and edge-based level set techniques for endocardial and epicardial border detection combined with non-rigid registration achieved by a 2D multi-scale cross-correlation and contour adaptation. This method was tested on 66 short-axis image sequences (Philips 1.5T) obtained in 11 patients at rest and during vasodilator stress at 3 levels of the left ventricle during first pass of a Gadolinium-DTPA bolus. Myocardial ROIs were automatically defined and contrast enhancement curves were constructed throughout the image sequence. Analysis of one sequence required

INTRODUCTION Myocardial perfusion assessment from cardiac magnetic resonance (CMR) images has proven to be feasible and useful. Extraction of quantitative perfusion information relies on the definition of myocardial regions of interest (ROI). This is usually achieved by manually drawing ROIs in one frame and then adjusting their position on subsequent frames. Unfortunately this tecnique requires a large amount of time and is potentially inaccurate. We recently developed a method based on image segmentation and registration for automated dynamic endocardial and epicardial borders detection for quantification of left ventricular perfusion. METHODS Data Acquisition: LV short-axis images (Philips 1.5T) were obtained at three levels of the left ventricle during first pass of a Gadolinium-DTPA bolus (0.10mmol/kg @5ml/ sec). Images were acquired using a hybrid gradient echo/echo planar imaging sequence (3 slices, thickness 10mm, pixel size 2.5x2.5mm, nonselective 90° saturation pulse followed by 80 ms delay, flip angle 20°, TR=5.9 ms, TE=2.7 ms, EPI factor 5, SENSE factor 2). Image segmentation: For each slice, the only manual action required was to place a seed point inside the LV cavity on a single frame. Starting from that point, an automated algorithm was developed to detect the two best frames for endocardial and epicardial boundaries segmentation, respectively. These frames were compulsory chosen among the first part of the image sequence, in which no myocardium movement took place. Hence the endocardial boundary was automatically computed using a modified region-based model based on the noise probability distribution of gray levels in MRI. Then epicardial boundary was detected using a classical edge based level-set model with curvature, advection and expansion driving terms. Both of the boundaries underwent a slight regularization motion in order to get a more physiological shape. Image registration: this step was necessary to compensate for respiratory motion and consisted in the rigid movement of the boundaries computed during the segmentation step. The goal was achieved by the implementation of two-dimensional cross-correlation. This algorithm was modified in order to look for correlation between the gradient of the images, in order to seek shape rather than intensity correspondence. Contrast enhancement curves extraction: Six standard myocardial ROIs were automatically defined and contrast enhancement curves were constructed throughout the image sequence. This approach was tested on 24 slices during first pass perfusion by: (1) visually judging frame-byframe the accuracy of endo- and epicardial boundary positions, and (2) calculating the ratio between the amplitude of the contrast enhancement curve and the SD of the plateau phase (SNR), which should be a good performance indicator of the algorithm as a whole. All the code was written in MATLAB® (The MathWorks, Natick, MA, USA). RESULTS Time needed to complete the automated analysis of one perfusion slice was less than 1 minute on a personal computer. The outcome of endo- and epicardial boundaries segmentation and registration was judged accurate in all image sequences. The computed contrast enhancement curves clearly showed the typical pattern of first-pass perfusion, with its wash-in slope and subsequent plateau phases. SNR averaged over 24 image sequences was 32±10. CONCLUSION Detection of myocardial boundaries and quantification of intra-myocardial contrast can be done in a fast, automated and dynamic fashion. The final outcome consists in regional contrast enhancement curves with excellent noise levels, which can be used to extract parameters of clinical value. The described approach provides a user-friendly and potentially more accurate technique, which may fulfill the strong need for clinical quantitative evaluation of myocardial perfusion from contrastenhanced CMR images.

3D video-fluoroscopy can accurately estimate in-vivo kinematics of joint prosthesis. To this aim, for each of the hundreds of frames of an acquisition, a 3D surface model of the prosthesis is registered to the relevant contours on the 2D X-ray projections. Commercial software only provide simple edge detector followed by a time consuming manual procedure to delete the undesired contours. A fast and robust semi-automated prosthesis segmentation method, combining region growing and level set methods, is proposed to speed up the analysis and to reduce the human interaction.