Search

Colorectal cancer is one of the most common cancers in the world. While colonoscopy is an effective screening technique, navigating an endoscope through the colon to detect polyps is challenging. A 3D map of the observed surfaces could enhance the identification of unscreened colon tissue and serve as a training platform. However, reconstructing the colon from video footage remains unsolved due to numerous factors such as self-occlusion, reflective surfaces, lack of texture, and tissue deformation that limit feature-based methods. Learning-based approaches hold promise as robust alternatives, but necessitate extensive datasets. By establishing a benchmark, the 2022 EndoVis sub-challenge SimCol3D aimed to facilitate data-driven depth and pose prediction during colonoscopy. The challenge was hosted as part of MICCAI 2022 in Singapore. Six teams from around the world and representatives from academia and industry participated in the three sub-challenges: synthetic depth prediction, synthetic pose prediction, and real pose prediction. This paper describes the challenge, the submitted methods, and their results. We show that depth prediction in virtual colonoscopy is robustly solvable, while pose estimation remains an open research question.

Colonoscopy is widely recognised as the gold standard procedure for the early detection of colorectal cancer (CRC). Segmentation is valuable for two significant clinical applications, namely lesion detection and classification, providing means to improve accuracy and robustness. The manual segmentation of polyps in colonoscopy images is time-consuming. As a result, the use of deep learning (DL) for automation of polyp segmentation has become important. However, DL-based solutions can be vulnerable to overfitting and the resulting inability to generalise to images captured by different colonoscopes. Recent transformer-based architectures for semantic segmentation both achieve higher performance and generalise better than alternatives, however typically predict a segmentation map of $\frac{h}{4}\times\frac{w}{4}$ spatial dimensions for a $h\times w$ input image. To this end, we propose a new architecture for full-size segmentation which leverages the strengths of a transformer in extracting the most important features for segmentation in a primary branch, while compensating for its limitations in full-size prediction with a secondary fully convolutional branch. The resulting features from both branches are then fused for final prediction of a $h\times w$ segmentation map. We demonstrate our method's state-of-the-art performance with respect to the mDice, mIoU, mPrecision, and mRecall metrics, on both the Kvasir-SEG and CVC-ClinicDB dataset benchmarks. Additionally, we train the model on each of these datasets and evaluate on the other to demonstrate its superior generalisation performance., 16 pages, 4 figures

In this paper, we introduce mREAL-GAN, a generative adversarial network (GAN) for the parallel generation of multiple residential electrical appliance load (mREAL) profiles. mREAL-GAN is intended for use in community-scale low-voltage network analysis, and represents a departure from previous methods for this purpose, which break the generation of appliance load profiles into several steps and largely model each appliance independently. Instead, mREAL-GAN models appliance load profiles in an end-to-end manner, and generates multiple appliance load profiles in parallel in a way that captures inter-dependencies. We show that mREAL-GAN generates load profiles for individual appliance-types with greater fidelity than a popular example of previous methods, and demonstrate its ability to capture inter-dependencies between appliances.

This paper describes preliminary work in the recent promising approach of generating synthetic training data for facilitating the learning procedure of deep learning (DL) models, with a focus on aerial photos produced by unmanned aerial vehicles (UAV). The general concept and methodology are described, and preliminary results are presented, based on a classification problem of fire identification in forests as well as a counting problem of estimating number of houses in urban areas. The proposed technique constitutes a new possibility for the DL community, especially related to UAV-based imagery analysis, with much potential, promising results, and unexplored ground for further research., Workshop on Deep-learning based computer vision for UAV in conjunction with CAIP 2019, Salerno, italy, September 2019

Colorectal adenocarcinoma originating in intestinal glandular structures is the most common form of colon cancer. In clinical practice, the morphology of intestinal glands, including architectural appearance and glandular formation, is used by pathologists to inform prognosis and plan the treatment of individual patients. However, achieving good inter-observer as well as intra-observer reproducibility of cancer grading is still a major challenge in modern pathology. An automated approach which quantifies the morphology of glands is a solution to the problem. This paper provides an overview to the Gland Segmentation in Colon Histology Images Challenge Contest (GlaS) held at MICCAI'2015. Details of the challenge, including organization, dataset and evaluation criteria, are presented, along with the method descriptions and evaluation results from the top performing methods.

3-D face recognition research has received significant attention in the past two decades because of the rapid development in imaging technology and ever increasing security demand of modern society. One of its challenges is to cope with non-rigid deformation among faces, which is often caused by the changes of appearance and facial expression. Popular solutions to deal with this problem are to detect the deformable parts of the face and exclude them, or to represent a face in terms of sparse signature points, curves or patterns that are invariant to deformation. Such approaches, however, may lead to loss of information which is important for classification. In this paper, we propose a new geodesic-map representation with statistical shape modelling for handling the non-rigid deformation challenge in face recognition. The proposed representation captures all geometrical information from the entire 3-D face and provides a compact and expression-free map that preserves intrinsic geometrical information. As a result, the search for dense points correspondence in the face recognition task can be speeded up by using a simple image-based method instead of time-consuming, recursive closest distance search in 3-D space. An experimental investigation was conducted on 3-D face scans using publicly available databases and compared with the benchmark approaches. The experimental results demonstrate that the proposed scheme provides a highly competitive new solution for 3-D face recognition.

The face reveals the healthy status of an individual, through a combination of physical signs and facial expressions. The project SEMEOTICONS is translating the semeiotic code of the human face into computational descriptors and measures, automatically extracted from videos, images, and 3D scans of the face. SEMEOTICONS is developing a multisensory platform, in the form of a smart mirror, looking for signs related to cardio-metabolic risk. The goal is to enable users to self-monitor their well-being status over time and improve their life-style via tailored user guidance. Building the multisensory mirror requires addressing significant scientific and technological challenges, from touch-less data acquisition, to real-time processing and integration of multimodal data.

International audience; Segmentation of cellular structures is of primary interest in cell imaging for cell shape reconstruction and to provide crucial information about possible cell morphology changes during radiotherapy for instance. From the particular perspective of predictive oncology, this paper reports on a novel method for membrane segmentation from single channel actin tagged fluorescence confocal microscopy images, which remains a challenging task. Proposed method is based on the use of the Approximate Entropy formerly introduced by Pincus embedded within a Geodesic Active Contour approach. Approximate Entropy can be seen as an estimator of the regularity of a particular sequence of values and, consequently, can be used as an edge detector. In this prospective study, a preliminary study on Approximate Entropy as an edge detector function is first proposed with a particular focus on the robustness to noise, and some promising membrane segmentation results obtained on confocal microscopy images are also shown.

International audience; We propose an unsupervised statistical region based active contour approach integrating an original fractional entropy measure for image segmentation with a particular application to single channel actin tagged fluorescence confocal microscopy image segmentation. Following description of statistical based active contour segmentation and the mathematical definition of the proposed fractional entropy descriptor, we demonstrate comparative segmentation results between the proposed approach and standard Shannon's entropy on synthetic and natural images. We also show that the proposed unsupervised statistical based approach, integrating the fractional entropy measure, leads to very satisfactory segmentation of the cell nuclei from which shape characterization can be calculated.

We propose an unsupervised statistical region based active contour approach integrating an original fractional entropy measure for image segmentation with a particular application to single channel actin tagged fluorescence confocal microscopy image segmentation. Following description of statistical based active contour segmentation and the mathematical definition of the proposed fractional entropy descriptor, we demonstrate comparative segmentation results between the proposed approach and standard Shannon’s entropy on synthetic and natural images. We also show that the proposed unsupervised \ud statistical based approach, integrating the fractional entropy measure, leads to very satisfactory segmentation of the cell nuclei from which shape characterization can be calculated.

International audience; Segmentation of cellular structures is of primary interest in cell imaging for a 3D reconstruction of cell shape. Such an analysis provides crucial information about cell morphology and is instrumental in understanding of biological processes leading to development of a particular pathology. The work presented in this paper reports on a novel method for segmentation of cellular structures (nuclei and cell boundaries) from 3D single channel actin tagged fluorescence confocal microscopy images. The proposed segmentation method uses histogram-based image similarity measure in a level-set active-contour framework. The novelty of the method is in application of the alpha-divergence distance measure which can be seen as a generalization of classic Kullback-Leibler and Chi 2 measures. The resulting alpha-divergence level-set formulation leads to a single front evolution formula for both nuclei and cell boundaries segmentation, with no requirements for any enhancement or preprocessing of acquired cell images (a monolayer of human cells (PNT2) culture).

Best Student Paper Award; International audience; In the framework of cell structure characterization for predictive oncology, we pro- pose in this paper an unsupervised statistical region based active contour approach in- tegrating an original fractional entropy measure for single channel actin tagged fluo- rescence confocal microscopy image segmentation. Following description of statistical based active contour segmentation and the mathematical definition of the proposed frac- tional entropy descriptor, we demonstrate comparative segmentation results between the proposed approach and standard Shannon's entropy obtained for nuclei segmentation. We show that the unsupervised proposed statistical based approach integrating the frac- tional entropy measure leads to very satisfactory segmentation of the cell nuclei from which shape characterization can be subsequently used for the therapy progress assess- ment.

This paper presents a novel technique for a symmetric deformable image registration based on a new method for fast and accurate direct inversion of a large motion model deformation field. The proposed image registration algorithm maintain a one-to-one mapping between registered images by symmetrically warping them to each other, and by ensuring the inverse consistency criterion at each iteration. This makes the final estimation of forward and backward deformation fields anatomically plausible. The quantitative validation of the method has been performed on magnetic resonance data obtained for a pelvis area demonstrating applicability of the method to adaptive prostate radiotherapy. The experiments demonstrate the improved robustness in terms of inverse consistency error when compared to previously proposed methods for symmetric image registration.

International audience; This paper describes a novel method for active contour segmentation based on foreground/background alpha-divergence histogram distance measure. In recent years a number of variational segmentation techniques have been proposed for a region based active contour segmentation utilising different distance measures between probability density functions (pdf) describing foreground and background regions. The most common techniques use Ki2, Hellinger/Bhattacharya distances or Kullback-Leibler divergence. In this paper, it is proposed to generalize these methods by using the alpha-divergences distance function. This distance function depending on the selected value of its parameter encompasses mentioned above classical distances. The paper defines a partial differential equation, associated with alpha-divergence variational criterion, that governs the iterative deformations of the active contour. The experimental results on a synthetic data demonstrate that the proposed method outperforms previously proposed histogram based methods in terms of segmentation accuracy and robustness with respect to type and level of noise. The potential of the proposed technique for segmentation of cellular structures in fluorescence confocal microscopy data is also illustrated.

Facial expressions are visible signs of a person’s affective state, cognitive activity and personality. Humans can perform expression recognition with a remarkable robustness without conscious effort even under a variety of adverse conditions such as partially occluded faces, different appearances and poor illumination. Over the last two decades, the advances in imaging technology and ever increasing computing power have opened up a possibility of automatic facial expression recognition and this has led to significant research efforts from the computer vision and pattern recognition communities. One reason for this growing interest is due to a wide spectrum of possible applications in diverse areas, such as more engaging human-computer interaction (HCI) systems, video conferencing, augmented reality. Additionally from the biometric perspective, automatic recognition of facial expressions has been investigated in the context of monitoring patients in the intensive care and neonatal units for signs of pain and anxiety, behavioural research, identifying level of concentration, and improving face recognition. Automatic facial expression recognition is a difficult task due to its inherent subjective nature, which is additionally hampered by usual difficulties encountered in pattern recognition and computer vision research. The vast majority of the current state-of-the-art facial expression recognition systems are based on 2-D facial images or videos, which offer good performance only for the data captured under controlled conditions. As a result, there is currently a shift towards the use of 3-D facial data to yield better recognition performance. However, it requires more expensive data acquisition systems and sophisticated processing algorithms. The aim of this chapter is to provide an overview of the existing methodologies and recent advances in the facial expression recognition, as well as present a systematic description of the authors’ work on the use of 3-D facial data for automatic recognition of facial expressions, starting from data acquisition and database creation to data processing algorithms and performance evaluation.

This paper presents a novel technique for a consistent symmetric deformable image registration based on an accurate method for a direct inversion of a large motion model deformation field. The proposed image registration algorithm maintains one-to-one mapping between registered images by symmetrically warping them to another image. This makes the final estimation of forward and backward deformation fields anatomically plausible and applicable to adaptive prostate radiotherapy. The quantitative validation of the method is performed on magnetic resonance data obtained for pelvis area. The experiments demonstrate the improved robustness in terms of inverse consistency error and estimation accuracy of prostate position in comparison to the previously proposed methods.

International audience; Active contour methods are often methods of choice for de- manding segmentation problems, yet segmentation of medical images with complex intensity patterns still remains a challenge for these meth- ods. This paper proposes a method to incorporate interactively spec- ified foreground/background regions into the active model framework while keeping the user interaction to the minimum. To achieve that, the proposed functional to be minimized includes a term to encourage active contour to separate the points close to the specified foreground region from the points close to the specified background region in terms of geodesic distance. The experiments on multi-modal prostate images demonstrate that the proposed method not only can achieve robust and accurate results, but also provides an efficient way to interactively im- prove the results.

This short paper presents a deformable surface registration scheme which is based on the statistical shape\ud modelling technique. The method consists of two major processing stages, model building and model\ud fitting. A statistical shape model is first built using a set of training data. Then the model is deformed and\ud matched to the new data by a modified iterative closest point (ICP) registration process. The proposed\ud method is tested on real 3-D facial data from BU-3DFE database. It is shown that proposed method can\ud achieve a reasonable result on surface registration, and can be used for patient position monitoring in\ud radiation therapy and potentially can be used for monitoring of the radiation therapy progress for head and\ud neck patients by analysis of facial articulation.

Human faces play an important role in everyday life, including the expression of person identity, \ud emotion and intentionality, along with a range of biological functions. The human face has also become the \ud subject of considerable research effort, and there has been a shift towards understanding it using stimuli of \ud increasingly more realistic formats. In the current work, we outline progress made in the production of a \ud database of facial expressions in arguably the most realistic format, 3D dynamic. A suitable architecture for \ud capturing such 3D dynamic image sequences is described and then used to record seven expressions (fear, \ud disgust, anger, happiness, surprise, sadness and pain) by 10 actors at 3 levels of intensity (mild, normal and \ud extreme). We also present details of a psychological experiment that was used to formally evaluate the \ud accuracy of the expressions in a 2D dynamic format. The result is an initial, validated database for researchers \ud and practitioners. The goal is to scale up the work with more actors and expression types.

International audience; Purpose The non-invasive examination of the cardiac functions is an important diagnostic, and follow up, tool for detection and evaluation of cardiovascular dysfunctions. Magnetic resonance (MR) cardiac tagging is a technique imposing temporary spatial variation in longitudinal magnetization around the heart area, so that temporary tag patterns deforming with cardiac motion can be captured in a sequence of MR images. The tag patterns play the role of noninvasive fiducial markers, by tracking of which an accurate estimation of cardiac motion can be achieved in vivo. This motion pattern can be subsequently analyzed and abnormalities can be detected and assessed. Segmentation of myocardium is an essential step for most of the approaches used to estimate dynamic displacement field since only the tag lines inside the myocardium are of importance for the deformation estimation. However, in most of the literature, segmentation of myocardium is done manually, which involves a large amount of tedious, timeconsuming and error-prone work. In the work presented here a fully automatic segmentation method is proposed for delineation of myocardial boundaries in the tagged MR images. Methods A fully-automatic method has been developed by the authors to segment myocardium of left ventricle by identifying epicardial and endocardial contours using implicit active contour models driven by tag structures and constrained by shape priors. The method has been developed to work, and tested with a temporal short-axis sequences of tagged cardiac MR images acquired using the SPAMM sequence. Results The initial assessment of the proposed method was performed against manual segmentation carried out by clinicians. The evaluation data consisted of five data sets each containing five or six tagged MR images. The results obtained for the six images from the third data set are shown in Fig. 2, where manual delineation and automatic delineation are represented by thick and thin lines respectively. It can be seen that in general there is relatively good correlation between manual and automatic delineation of the myocardial boundaries. It has to be noted that for the reported initial validation results no information is available about intra- or inter-observer variability and therefore it is impossible to quantitatively assess quality of the ground truth. Indeed a closer look at the images in Fig. 2 reveals that on some occasions (e.g. for images 2 and 5) the automatically selected contours are closer to the endocardial boundaries than the manual ones.

International audience