Search

In traditional speech denoising tasks, clean audio signals are often used as the training target, but absolutely clean signals are collected from expensive recording equipment or in studios with the strict environments. To overcome this drawback, we propose an end-to-end self-supervised speech denoising training scheme using only noisy audio signals, named Only-Noisy Training (ONT), without extra training conditions. The proposed ONT strategy constructs training pairs only from each single noisy audio, and it contains two modules: training audio pairs generated module and speech denoising module. The first module adopts a random audio sub-sampler on each noisy audio to generate training pairs. The sub-sampled pairs are then fed into a novel complex-valued speech denoising module. Experimental results show that the proposed method not only eliminates the high dependence on clean targets of traditional audio denoising tasks, but also achieves on-par or better performance than other training strategies. Availability-ONT is available at https://github.com/liqingchunnnn/Only-Noisy-Training, Comment: 11 pages, 4 figures, 6 tables

International audience; One-to-multiple medical image segmentation aims to directly test a segmentation model trained with the medical images of a one-domain site on those of a multiple-domain site, suffering from segmentation performance degradation on multiple domains. This process avoids additional annotations and helps improve the application value of the model. However, no successful one-to-multiple unsupervised domain adaptation (O2M-UDA) work has been reported in one-to-multiple medical image segmentation due to its inherent challenges: distribution differences among multiple target domains (among-target differences) caused by different scanning equipment and distribution differences between one source domain and multiple target domains (source–target differences). In this paper, we propose an O2M-UDA framework called dynamic domain adaptation (DyDA), for one-to-multiple medical image segmentation, which has two innovations: (1) dynamic credible sample strategy (DCSS) dynamically extracts credible samples from the target site and iteratively updates their number, thus iteratively expanding the generalization boundary of the model and minimizing the among-target differences; (2) hybrid uncertainty learning (HUL) reduces the voxel-level and domain-level uncertainty simultaneously, thus minimizing the source–target differences from the detail and entire perspective concurrently. Experiments on two one-to-multiple medical image segmentation tasks have been conducted to demonstrate the performance of the proposed DyDA. The proposed DyDA achieved competitive segmentation results and high adaptation with an average of 83.8% and 48.1% dice for the two tasks, respectively, which has improved by 21.7% and 9.2% compared with no adaptation, respectively. The code developed in this study code can be downloaded at https://github.com/ZoeyJiang/DyDA. © 2023 Elsevier B.V.

Objective. Plan-of-the-day (PoD) adaptive radiation therapy (ART) is based on a library of treatment plans, among which, at each treatment fraction, the PoD is selected using daily images. However, this strategy is limited by PoD selection uncertainties. This work aimed to propose and evaluate a workflow to automatically and quantitatively identify the PoD for cervix cancer ART based on daily CBCT images. Approach. The quantification was based on the segmentation of the main structures of interest in the CBCT images (clinical target volume [CTV], rectum, bladder, and bowel bag) using a deep learning model. Then, the PoD was selected from the treatment plan library according to the geometrical coverage of the CTV. For the evaluation, the resulting PoD was compared to the one obtained considering reference CBCT delineations. Main results. In experiments on a database of 23 patients with 272 CBCT images, the proposed method obtained an agreement between the reference PoD and the automatically identified PoD for 91.5% of treatment fractions (99.6% when considering a 5% margin on CTV coverage). Significance. The proposed automatic workflow automatically selected PoD for ART using deep-learning methods. The results showed the ability of the proposed process to identify the optimal PoD in a treatment plan library.

International audience; 4D cone-beam computed tomography (CBCT) is an important imaging modality in image-guided radiation therapy to address the motion-induced artifacts caused by organ movements during the respiratory process. However, due to the extremely sparse projection data for each temporal phase, 4D CBCT reconstructions will suffer from severe streaking artifacts. Therefore, to tackle the streak artifacts and provide high-quality images, we proposed a framework termed Prior-Regularized Iterative Optimization Reconstruction (PRIOR) for 4D CBCT. The PRIOR framework combines the physics-based model and data-driven method simultaneously, with powerful feature extracting capacity, significantly promoting the image quality compared to single model-based or deep learning-based methods. Besides, we designed a specialized deep learning model named PRIOR-Net, which can effectively excavate the static information in the prior image reconstructed from the fully-sampled projections at the encoding stage to improve the reconstruction performance for individual phase-resolved images. Both the simulated and clinical 4D CBCT datasets were performed to evaluate the performance of the PRIOR-Net and the PRIOR framework. Compared with the advanced 4D CBCT reconstruction methods, the proposed methods achieve promising results quantitatively and qualitatively in streak artifact suppression, soft tissue restoration, and tiny detail preservation.

Automatic quantification of the left ventricle (LV) from cardiac magnetic resonance (CMR) images plays an important role in making the diagnosis procedure efficient, reliable, and alleviating the laborious reading work for physicians. Considerable efforts have been devoted to LV quantification using different strategies that include segmentation-based (SG) methods and the recent direct regression (DR) methods. Although both SG and DR methods have obtained great success for the task, a systematic platform to benchmark them remains absent because of differences in label information during model learning. In this paper, we conducted an unbiased evaluation and comparison of cardiac LV quantification methods that were submitted to the Left Ventricle Quantification (LVQuan) challenge, which was held in conjunction with the Statistical Atlases and Computational Modeling of the Heart (STACOM) workshop at the MICCAI 2018. The challenge was targeted at the quantification of 1) areas of LV cavity and myocardium, 2) dimensions of the LV cavity, 3) regional wall thicknesses (RWT), and 4) the cardiac phase, from mid-ventricle short-axis CMR images. First, we constructed a public quantification dataset Cardiac-DIG with ground truth labels for both the myocardium mask and these quantification targets across the entire cardiac cycle. Then, the key techniques employed by each submission were described. Next, quantitative validation of these submissions were conducted with the constructed dataset. The evaluation results revealed that both SG and DR methods can offer good LV quantification performance, even though DR methods do not require densely labeled masks for supervision. Among the 12 submissions, the DR method LDAMT offered the best performance, with a mean estimation error of 301 mm $^2$ for the two areas, 2.15 mm for the cavity dimensions, 2.03 mm for RWTs, and a 9.5% error rate for the cardiac phase classification. Three of the SG methods also delivered comparable performances. Finally, we discussed the advantages and disadvantages of SG and DR methods, as well as the unsolved problems in automatic cardiac quantification for clinical practice applications.

International audience; Renal tumor diagnosis on abdominal enhanced CT volumes is one of the most significant tasks in kidney disease diagnosis. It helps clinicians decide whether to perform the surgery (subtype classification), perform radical operations or minimally invasive treatment (grade classification). However, inherent challenges greatly limit the performance of the model: 1) Tumor appearance differences caused by non-tumor factors. 2) Small inter-class differences and large intra-class variations. In this paper, we propose a novel diagnosis framework for renal tumors, Bi-knowledge Contrastive Network (BKC-Net), which has two innovations: (1) Focus-perceive learning segments the tumors while perceiving the surrounding healthy tissues, thus adjusting the model's representation of tumor appearance, helping the BKC-Net represent the inherent features of tumors. (2) Bi-knowledge contrastive learning introduces prior radiomics features, makes the prior radiomics knowledge and latent deep knowledge complementary to each other from the intra-case level, and forces the high cohesion and low coupling embedding feature space from the inter-case levels, helping to discover subtle but essential differences among classes. Experiments demonstrate that our BKC-Net has the best performance in renal tumor diagnosis. Results reveal that our framework has great potential for renal tumor diagnosis in clinical use. Source codes will be released at https://github.com/DD0922/BKC-Net-pytorch. © 2022 Elsevier B.V.

International audience; ObjectivesThe accurate preoperative segmentation of the uterus and uterine fibroids from magnetic resonance images (MRI) is an essential step for diagnosis and real-time ultrasound guidance during high-intensity focused ultrasound (HIFU) surgery. Conventional supervised methods are effective techniques for image segmentation. Recently, semi-supervised segmentation approaches have been reported in the literature. One popular technique for semi-supervised methods is to use pseudo-labels to artificially annotate unlabeled data. However, many existing pseudo-label generations rely on a fixed threshold used to generate a confidence map, regardless of the proportion of unlabeled and labeled data.Materials and MethodsTo address this issue, we propose a novel semi-supervised framework called Confidence-based Threshold Adaptation Network (CTANet) to improve the quality of pseudo-labels. Specifically, we propose an online pseudo-labels method to automatically adjust the threshold, producing high-confident unlabeled annotations and boosting segmentation accuracy. To further improve the network's generalization to fit the diversity of different patients, we design a novel mixup strategy by regularizing the network on each layer in the decoder part and introducing a consistency regularization loss between the outputs of two sub-networks in CTANet.ResultsWe compare our method with several state-of-the-art semi-supervised segmentation methods on the same uterine fibroids dataset containing 297 patients. The performance is evaluated by the Dice similarity coefficient, the precision, and the recall. The results show that our method outperforms other semi-supervised learning methods. Moreover, for the same training set, our method approaches the segmentation performance of a fully supervised U-Net (100% annotated data) but using 4 times less annotated data (25% annotated data, 75% unannotated data).ConclusionExperimental results are provided to illustrate the effectiveness of the proposed semi-supervised approach. The proposed method can contribute to multi-class segmentation of uterine regions from MRI for HIFU treatment.

International audience; PURPOSE: Clinical rib fracture diagnosis via computed tomography (CT) screening has attracted much attention in recent years. However, automated and accurate segmentation solutions remain a challenging task due to the large sets of 3D CT data to deal with. Down-sampling is often required to face computer constraints, but the performance of the segmentation may decrease in this case. METHODS: A new multi-angle projection network (MAPNet) method is proposed for accurately segmenting rib fractures by means of a deep learning approach. The proposed method incorporates multi-angle projection images to complementarily and comprehensively extract the rib characteristics using a rib extraction (RE) module and the fracture features using a fracture segmentation (FS) module. A multi-angle projection fusion (MPF) module is designed for fusing multi-angle spatial features. RESULTS: It is shown that MAPNet can capture more detailed rib fracture features than some commonly used segmentation networks. Our method achieves a better performance in accuracy (88.06 ± 6.97%), sensitivity (89.26 ± 5.69%), specificity (87.58% ± 7.66%) and in terms of classical criteria like dice (85.41 ± 3.35%), intersection over union (IoU, 80.37 ± 4.63%), and Hausdorff distance (HD, 4.34 ± 3.1). CONCLUSION: We propose a rib fracture segmentation technique to deal with the problem of automatic fracture diagnosis. The proposed method avoids the down-sampling of 3D CT data through a projection technique. Experimental results show that it has excellent potential for clinical applications.

Tumor control and survival of patients with metastatic pancreatic ductal adenocarcinoma (PDAC) has improved with more effective polychemotherapies. The identification of novel therapeutic targets is strongly needed in order to propose maintenance therapies that improve quality of life while maintaining tumor control.PDAC with mutations in homologous recombination repair genes such asThe demonstration of olaparib efficacy in patients with metastatic PDAC and BRCA germline mutation has paved the way for maintenance with a targeted therapy. Further studies are needed to assess; the potential role for PARPI in earlier forms of PDAC, those with somatic or more rare BRACness signatures, to overcome primary or secondary resistances to PARPi, and to combine them with other antitumoral agents.

An effective backbone network is important to deep learning-based Deformable Medical Image Registration (DMIR), because it extracts and matches the features between two images to discover the mutual correspondence for fine registration. However, the existing deep networks focus on single image situation and are limited in registration task which is performed on paired images. Therefore, we advance a novel backbone network, XMorpher, for the effective corresponding feature representation in DMIR. 1) It proposes a novel full transformer architecture including dual parallel feature extraction networks which exchange information through cross attention, thus discovering multi-level semantic correspondence while extracting respective features gradually for final effective registration. 2) It advances the Cross Attention Transformer (CAT) blocks to establish the attention mechanism between images which is able to find the correspondence automatically and prompts the features to fuse efficiently in the network. 3) It constrains the attention computation between base windows and searching windows with different sizes, and thus focuses on the local transformation of deformable registration and enhances the computing efficiency at the same time. Without any bells and whistles, our XMorpher gives Voxelmorph 2.8% improvement on DSC , demonstrating its effective representation of the features from the paired images in DMIR. We believe that our XMorpher has great application potential in more paired medical images. Our XMorpher is open on https://github.com/Solemoon/XMorpher, Comment: accepted by MICCAI 2022

International audience; Although there have been a lot of researches on convolutional neural networks (CNNs), still what hap-pens in this black box remains a mystery. In this paper, we establish the connection between CNNs and signal modulation. From a signal modulation point of view, the forward-propagation process of CNNs can be explained as a process of modulating the input signals to the vicinity of a special energy spectrum distribution, and the back-propagation process is searching for the appropriate distribution which is better for classification or other tasks. Several experiments have been carried out to verify the modulated explanation of CNNs. Furthermore, we verify that modulating the signal to the appropriate energy spectrum distribution in advance can effectively improve the classification and segmentation accuracy.(c) 2022 Elsevier B.V. All rights reserved.

International audience; BACKGROUND AND OBJECTIVE: Recently, spectral Dynamic Causal Modelling (DCM) has been used increasingly to infer effective connectivity from epileptic intracranial electroencephalographic (iEEG) signals. In this context, the Physiology-Based Model (PBM), a neural mass model, is used as a generative model. However, previous studies have highlighted out the inability of PBM to properly describe iEEG signals with specific power spectral densities (PSDs). More precisely, PSDs that have multiple peaks around β and γ rhythms (i.e. spectral characteristics at seizure onset) are concerned. METHODS: To cope with this limitation, an alternative neural mass model, called the complete PBM (cPBM), is investigated. The spectral DCM and two recent variants are used to evaluate the relevance of cPBM over PBM. RESULTS: The study is conducted on both simulated signals and real epileptic iEEG recordings. Our results confirm that, compared to PBM, cPBM shows (i) more ability to model the desired PSDs and (ii) lower numerical complexity whatever the method. CONCLUSIONS: Thanks to its intrinsic and extrinsic connectivity parameters as well as the input coming into the fast inhibitory subpopulation, the cPBM provides a more expressive model of PSDs, leading to a better understanding of epileptic patterns and DCM-based effective connectivity inference.

International audience; Automatic detection of arrhythmia through an electrocardiogram (ECG) is of great significance for the prevention and treatment of cardiovascular diseases. In Convolutional neural network, the ECG signal is converted into multiple feature channels with equal weights through the convolution operation. Multiple feature channels can provide richer and more comprehensive information, but also contain redundant information, which will affect the diagnosis of arrhythmia, so feature channels that contain arrhythmia information should be paid attention to and given larger weight. In this paper, we introduced the Squeeze-and-Excitation (SE) block for the first time for the automatic detection of multiple types of arrhythmias with ECG. Our algorithm combines the residual convolutional module and the SE block to extract features from the original ECG signal. The SE block adaptively enhances the discriminative features and suppresses noise by explicitly modeling the interdependence between the channels, which can adaptively integrate information from different feature channels of ECG. The one-dimensional convolution operation over the time dimension is used to extract temporal information and the shortcut connection of the Se-Residual convolutional module in the proposed model makes the network easier to optimize. Thanks to the powerful feature extraction capabilities of the network, which can effectively extract discriminative arrhythmia features in multiple feature channels, so that no extra data preprocessing including denoising in other methods are need for our framework. It thus improves the working efficiency and keeps the collected biological information without loss. Experiments conducted with the 12-lead ECG dataset of the China Physiological Signal Challenge (CPSC) 2018 and the dataset of PhysioNet/Computing in Cardiology (CinC) Challenge 2017. The experiment results show that our model gains great performance and has great potential in clinical.

International audience; Generative adversarial networks and variational autoencoders (VAEs) provide impressive image generation from Gaussian white noise, but both are difficult to train, since they need a generator (or encoder) and a discriminator (or decoder) to be trained simultaneously, which can easily lead to unstable training. To solve or alleviate these synchronous training problems of generative adversarial networks (GANs) and VAEs, researchers recently proposed generative scattering networks (GSNs), which use wavelet scattering networks (ScatNets) as the encoder to obtain features (or ScatNet embeddings) and convolutional neural networks (CNNs) as the decoder to generate an image. The advantage of GSNs is that the parameters of ScatNets do not need to be learned, while the disadvantage of GSNs is that their ability to obtain representations of ScatNets is slightly weaker than that of CNNs. In addition, the dimensionality reduction method of principal component analysis (PCA) can easily lead to overfitting in the training of GSNs and, therefore, affect the quality of generated images in the testing process. To further improve the quality of generated images while keeping the advantages of GSNs, this study proposes generative fractional scattering networks (GFRSNs), which use more expressive fractional wavelet scattering networks (FrScatNets), instead of ScatNets as the encoder to obtain features (or FrScatNet embeddings) and use similar CNNs of GSNs as the decoder to generate an image. Additionally, this study develops a new dimensionality reduction method named feature-map fusion (FMF) instead of performing PCA to better retain the information of FrScatNets,; it also discusses the effect of image fusion on the quality of the generated image. The experimental results obtained on the CIFAR-10 and CelebA datasets show that the proposed GFRSNs can lead to better generated images than the original GSNs on testing datasets. The experimental results of the proposed GFRSNs with deep convolutional GAN (DCGAN), progressive GAN (PGAN), and CycleGAN are also given.

International audience; The segmentation of the lung lobes is generally a prerequisite for some specific diagnostic tasks and even more so for the planning of the therapy of lung diseases. Recently this task has been tackled with the help of deep learning and more particularly by using U-Net networks. However, these networks have great difficulty in segmenting the lobes because the lung is extremely heterogeneous and the fissures between lobes are extremely fine and therefore blurred in the images. This leads to relatively inaccurate segmentation results, especially in terms of spatial continuity over the 2D lung. Inspired by the adversarial networks in medical image segmentation, we propose in this paper a new segmentation scheme: the lung lobes adversarial based segmentation network (LLASN). In this scheme U-Net is used to generate segmentation results and a discriminator network is used to discriminate the generated segmentation results from Ground Truth labels. The proposed method is evaluated on the LUNA16 dataset and the Tianchi Medical AI Competition dataset. Compared to the classical U-Net network, the proposed network improved Dice similarity coefficient from 0.82 to 0.84, 0.83 to 0.84, and 0.59 to 0.64 for left lower lobe, left upper lobe, and right middle lobe in LUNA16 dataset.

International audience; Background - During the COVID-19 pandemic people were asked to keep interpersonal distance, wash their hands and avoid gatherings of people. But, do people understand how much a change of the distance to a virus infected person means for the exposure to that person's virus? To answer this question, we studied how people perceive virus exposure from an infected person at different distances and lengths of a conversation. Method - An online questionnaire was distributed to 101 participants drawn from the general US population. Participants judged perceived virus exposure at different interpersonal distances to an infected person in a face to face conversation of different lengths of time. A model based on empirical and theoretical studies of dispersion of particles in the air was used to estimate a person's objective virus exposure during different times and distances from a virus source. The model and empirical data show that exposure changes with the square of the distance and linearly with time. Results - A majority (78%) of the participants underestimated the effects on virus exposure following a change of interpersonal distance. The dominating bias was assuming that exposure varies linearly with distance. To illustrate, an approach to a virus source from 6 to 2 feet was judged to give a 3 times higher exposure but, objectively it is 9 times. By way of contrast, perceptions of exposure as a function of the duration of a conversation were unbiased. The COVID-19 pandemic caused by the SARS-CoV2 virus is likely to be followed by other pandemics also caused by airborne Corona or other viruses. Therefore, the results are important for administrators when designing risk communications to the general public and workers in the health care sector about social distancing and infection risks. Conclusions - People quite drastically underestimate the increase in virus exposure following an approach to a virus infected person. They also overestimate exposure after a move away from an infected person. For public health reasons, the correct function connecting distance with virus exposure should be communicated to the general public to avoid deliberate violations of recommended interpersonal distances. Supplementary information - The online version contains supplementary material available at 10.1007/s44155-022-00027-9.

Renal compartment segmentation on CT images targets on extracting the 3D structure of renal compartments from abdominal CTA images and is of great significance to the diagnosis and treatment for kidney diseases. However, due to the unclear compartment boundary, thin compartment structure and large anatomy variation of 3D kidney CT images, deep-learning based renal compartment segmentation is a challenging task. We propose a novel weakly supervised learning framework, Cycle Prototype Network, for 3D renal compartment segmentation. It has three innovations: (1) A Cycle Prototype Learning (CPL) is proposed to learn consistency for generalization. It learns from pseudo labels through the forward process and learns consistency regularization through the reverse process. The two processes make the model robust to noise and label-efficient. (2) We propose a Bayes Weakly Supervised Module (BWSM) based on cross-period prior knowledge. It learns prior knowledge from cross-period unlabeled data and perform error correction automatically, thus generates accurate pseudo labels. (3) We present a Fine Decoding Feature Extractor (FDFE) for fine-grained feature extraction. It combines global morphology information and local detail information to obtain feature maps with sharp detail, so the model will achieve fine segmentation on thin structures. Our extensive experiments demonstrated our great performance. Our model achieves Dice of \(79.1\%\) and \(78.7\%\) with only four labeled images, achieving a significant improvement by about \(20\%\) than typical prototype model PANet [16].

International audience; BACKGROUND AND OBJECTIVE: Aortic dissection is a severe cardiovascular pathology in which an injury of the intimal layer of the aorta allows blood flowing into the aortic wall, forcing the wall layers apart. Such situation presents a high mortality rate and requires an in-depth understanding of the 3-D morphology of the dissected aorta to plan the right treatment. An accurate automatic segmentation algorithm is therefore needed. METHOD: In this paper, we propose a deep-learning-based algorithm to segment dissected aorta on computed tomography angiography (CTA) images. The algorithm consists of two steps. Firstly, a 3-D convolutional neural network (CNN) is applied to divide the 3-D volume into two anatomical portions. Secondly, two 2-D CNNs based on pyramid scene parsing network (PSPnet) segment each specific portion separately. An edge extraction branch was added to the 2-D model to get higher segmentation accuracy on intimal flap area. RESULTS: The experiments conducted and the comparisons made show that the proposed solution performs well with an average dice index over 92%. The combination of 3-D and 2-D models improves the aorta segmentation accuracy compared to 3-D only models and the segmentation robustness compared to 2-D only models. The edge extraction branch improves the DICE index near aorta boundaries from 73.41% to 81.39%. CONCLUSIONS: The proposed algorithm has satisfying performance for capturing the aorta structure while avoiding false positives on the intimal flaps.

International audience; Blind image quality assessment (BIQA) is one of the most challenging and difficult tasks in the field of IQA. Given that sparse representation through dictionary learning can learn the image feature well, this paper proposed a method termed Discriminative Feature Representation (DFR) from the perspective of feature learning for noise contaminated image quality assessment. DFR makes use of two sub-dictionaries composed of atoms featuring desirable image structures and undesirable noise, respectively. Noise is quantified via a joint evaluation of the sparse coefficients related to the atoms in the two sub-dictionaries. The method is validated using public databases with different types of noise, a comparison with other up-to-date methods is provided. The proposed method is also applied to CT images acquired at different-level doses and reconstructed by various well-known algorithms.

International audience; Sparse-view sampling scans reduce the patient's radiation dose by reducing the total exposure duration. CT reconstructions under such scan mode are often accompanied by severe artifacts due to the high ill-posedness of the problem. In this paper, we use a Non-Local means kernel as a regularization constraint to reconstruct image volumes from sparse-angle sampled cone-beam CT scans. To overcome the huge computational cost of the 3D reconstruction, we propose a sequential update scheme relying on ordered subsets in the image domain. It is shown through experiments on simulated and real data and comparisons with other methods that the proposed approach is robust enough to deal with the number of views reduced up to 1/10. When coupled with a CUDA parallel computing technique, the computation speed of the iterative reconstruction is greatly improved.

International audience; X-ray computed tomography (CT) is one of the most widely used tools in medical imaging, industrial nondestructive testing, lesion detection, and other applications. However, decreasing the projection number to lower the X-ray radiation dose usually leads to severe streak artifacts. To improve the quality of the images reconstructed from sparse-view projection data, we developed a hybrid-domain neural network (HDNet) processing for sparse-view CT (SVCT) reconstruction in this study. The HDNet decomposes the SVCT reconstruction problem into two stages and each stage focuses on one mission, which reduces the learning difficulty of the entire network. Experiments based on the simulated and clinical datasets are performed to demonstrate the performance of the proposed method. Compared with other competitive algorithms, quantitative and qualitative results show that the proposed method makes a great improvement on artifact suppression, tiny structure restoration, and contrast retention.

International audience; Accurate coronary lumen segmentation on coronary-computed tomography angiography (CCTA) images is crucial for quantification of coronary stenosis and the subsequent computation of fractional flow reserve. Many factors including difficulty in labeling coronary lumens, various morphologies in stenotic lesions, thin structures and small volume ratio with respect to the imaging field complicate the task. In this work, we fused the continuity topological information of centerlines which are easily accessible, and proposed a novel weakly supervised model, Examinee-Examiner Network (EE-Net), to overcome the challenges in automatic coronary lumen segmentation. First, the EE-Net was proposed to address the fracture in segmentation caused by stenoses by combining the semantic features of lumens and the geometric constraints of continuous topology obtained from the centerlines. Then, a Centerline Gaussian Mask Module was proposed to deal with the insensitiveness of the network to the centerlines. Subsequently, a weakly supervised learning strategy, Examinee-Examiner Learning, was proposed to handle the weakly supervised situation with few lumen labels by using our EE-Net to guide and constrain the segmentation with customized prior conditions. Finally, a general network layer, Drop Output Layer, was proposed to adapt to the class imbalance by dropping well-segmented regions and weights the classes dynamically. Extensive experiments on two different data sets demonstrated that our EE-Net has good continuity and generalization ability on coronary lumen segmentation task compared with several widely used CNNs such as 3D-UNet. The results revealed our EE-Net with great potential for achieving accurate coronary lumen segmentation in patients with coronary artery disease. Code at http://github.com/qiyaolei/Examinee-Examiner-Network.

International audience; Tomographic perfusion imaging is a significant imaging modality for stroke diagnosis. However, the low rotational speed of the C-arm (6-8 s per circle) is a challenge for applying perfusion imaging in C-arm cone beam computed tomography (CBCT). Traditional reconstruction methods cannot remove the artifacts caused by the slow rotational speed or acquire enough sample points to restore the time density curve (TDC). This paper presents a dynamic rollback reconstruction method for CBCT. The proposed method can improve the temporal resolution by increasing the sample points used for calculating the TDC. Combined with existing techniques, the algorithm allows slow-rotating scanners to be used for perfusion imaging purposes. In the experiments, the proposed method was compared with other dynamic reconstruction algorithms based on standard reconstruction and the temporal interpolation approach. The presented algorithm could improve the temporal resolution without increasing the X-ray exposure time or contrast agent.

International audience; Background: Compared with conventional magnetic resonance imaging methods, the quantitative magnetic susceptibility mapping (QSM) technique can quantitatively measure the magnetic susceptibility distribution of tissues, which has an important clinical application value in the investigations of brain micro-bleeds, Parkinson's, and liver iron deposition, etc. However, the quantitative susceptibility mapping algorithm is an ill-posed inverse problem due to the near-zero value in the dipole kernel, and high-quality QSM reconstruction with effective streaking artifact suppression remains a challenge. In recent years, the performance of sparse representation has been well validated in improving magnetic resonance image (MRI) reconstruction. Methods: In this study, by incorporating feature learning into sparse representation, we propose an edge prior guided dictionary learning-based reconstruction method for the dipole inversion in quantitative susceptibility mapping reconstruction. The structure feature dictionary relies on magnitude images for susceptibility maps have similar structures with magnitude images, and this structure feature dictionary and edge prior information are used in the dipole inversion step. Results: The performance of the proposed algorithm is assessed through in vivo human brain clinical data, leading to high-quality susceptibility maps with improved streaking artifact suppression, structural recovery, and quantitative metrics. Conclusions: The proposed edge prior guided dictionary learning method for dipole inversion in QSM achieves improved performance in streaking artifacts suppression, structural recovery and deep gray matter reconstruction.

International audience; 3D complete renal structures (CRS) segmentation targets on segmenting the kidneys, tumors, renal arteries and veins in one inference. Once successful, it will provide preoperative plans and intraoperative guidance for laparoscopic partial nephrectomy(LPN), playing a key role in the renal cancer treatment. However, no success has been reported in 3D CRS segmentation due to the complex shapes of renal structures, low contrast and large anatomical variation. In this study, we utilize the adversarial ensemble learning and propose Ensemble Multi-condition GAN (EnMcGAN) for 3D CRS segmentation for the first time. Its contribution is three-fold. 1) Inspired by windowing [4], we propose the multi-windowing committee which divides CTA image into multiple narrow windows with different window centers and widths enhancing the contrast for salient boundaries and soft tissues. And then, it builds an ensemble segmentation model on these narrow windows to fuse the segmentation superiorities and improve whole segmentation quality. 2) We propose the multi-condition GAN which equips the segmentation model with multiple discriminators to encourage the segmented structures meeting their real shape conditions, thus improving the shape feature extraction ability. 3) We propose the adversarial weighted ensemble module which uses the trained discriminators to evaluate the quality of segmented structures, and normalizes these evaluation scores for the ensemble weights directed at the input image, thus enhancing the ensemble results. 122 patients are enrolled in this study and the mean Dice coefficient of the renal structures achieves 84.6%. Extensive experiments with promising results on renal structures reveal powerful segmentation accuracy and great clinical significance in renal cancer treatment. © 2021, Springer Nature Switzerland AG.

International audience; Three-dimensional (3D) integrated renal structures (IRS) segmentation targets segmenting the kidneys, renal tumors, arteries, and veins in one inference. Clinicians will benefit from the 3D IRS visual model for accurate preoperative planning and intraoperative guidance of laparoscopic partial nephrectomy (LPN). However, no success has been reported in 3D IRS segmentation due to the inherent challenges in grayscale distribution: low contrast caused by the narrow task-dependent distribution range of regions of interest (ROIs), and the networks representation preferences caused by the distribution variation inter-images. In this paper, we propose the Meta Greyscale Adaptive Network (MGANet), the first deep learning framework to simultaneously segment the kidney, renal tumors, arteries and veins on CTA images in one inference. It makes innovations in two collaborate aspects: 1) The Grayscale Interest Search (GIS) adaptively focuses segmentation networks on task-dependent grayscale distributions via scaling the window width and center with two cross-correlated coefficients for the first time, thus learning the fine-grained representation for fine segmentation. 2) The Meta Grayscale Adaptive (MGA) learning makes an image-level meta-learning strategy. It represents diverse robust features from multiple distributions, perceives the distribution characteristic, and generates the model parameters to fuse features dynamically according to image's distribution, thus adapting the grayscale distribution variation. This study enrolls 123 patients and the average Dice coefficients of the renal structures are up to 87.9%. Fine selection of the task-dependent grayscale distribution ranges and personalized fusion of multiple representations on different distributions will lead to better 3D IRS segmentation quality. Extensive experiments with promising results on renal structures reveal powerful segmentation accuracy and great clinical significance in renal cancer treatment.

International audience; Limited-angle CT is an indispensable tool for some practical applications when the projection data can be only collected within a limited-angle range due to the constraints of scanning conditions. However, the limited-angle scanning mode will lead to severely degraded images with excessive artifacts. Meanwhile, existing methods fail to reconstruct satisfactory images in limited-angle CT because of the unguaranteed measurement consistency caused by serious projection missing. In this paper, we developed a method termed Single-shot Projection Error Correction Integrated Adversarial Learning (SPECIAL) progressive-improvement strategy, which could effectively combine the complementary information contained in the image domain and projection domain, and greatly improve the reconstructions at the expense of small computational cost. Specifically, enhanced adversarial learning is used in different stages to remove artifacts without losing high-frequency component. A projection error correction module is used to boost the performance in high-attenuation tissue restoration. Compared with other competitive algorithms, quantitative and qualitative results show that the proposed method could make a promising improvement on artifact removal, edge preservation and tiny structure restoration.

International audience; In this paper, a new matrix-based method is proposed to real-time determine the guidewire position inside an arterial system. The guidewire path is obtained by the optimal path method, particularly, the fusiform ternary tree method according to the principle of minimum output value of root node. An adaptive sampling strategy, and an optimization strategy based on proximal end and distal end of the guidewire are proposed to change the guidewire position for obtaining an ideal guidewire path. Compared to the existing methods, the proposed method can achieve 74%, 64% and 70% improvements in accuracy for phantoms 1, 2, and 3 investigated in this work.

International audience; Unlabelled - The electricity demand for space cooling in the non-residential building (NRB) sector of China is growing significantly and is becoming increasingly critical with rapid economic development and mounting impacts of climate change. The growing demand for space cooling will increase global warming due to emissions of hydrofluorocarbons used in cooling equipment and carbon dioxide emissions from the mostly fossil fuel-based electricity currently powering space cooling. This study uses the Greenhouse Gas and Air Pollution Interaction and Synergies (GAINS) model framework to estimate current and future emissions of hydrofluorocarbons and their abatement potentials for space cooling in the NRB sector of China and assess the co-benefits in the form of savings in electricity and associated reductions in greenhouse gas (GHG), air pollution, and short-lived climate pollutant emissions. Co-benefits of space cooling are assessed by taking into account (a) regional and urban/rural heterogeneities and climatic zones among different provinces; (b) technical/economic energy efficiency improvements of the cooling technologies; and (c) transition towards lower global warming potential (GWP) refrigerants under the Kigali Amendment. Under the business-as-usual (BAU) scenario, the total energy consumption for space cooling in the NRB sector will increase from 166 TWh in 2015 to 564 TWh in 2050, primarily due to the rapid increase in the floor space area of non-residential buildings. The total GHG mitigation potential due to the transition towards low-GWP refrigerants and technical energy efficiency improvement of cooling technologies will approximately be equal to 10% of the total carbon emissions from the building sector of China in 2050. Supplementary information - The online version contains supplementary material available at 10.1007/s11027-022-10021-w.

International audience; Dual-energy computed tomography (DECT) is of great significance for clinical practice due to its huge potential to provide material-specific information. However, DECT scanners are usually more expensive than standard single-energy CT (SECT) scanners and thus are less accessible to undeveloped regions. In this paper, we show that the energy-domain correlation and anatomical consistency between standard DECT images can be harnessed by a deep learning model to provide high-performance DECT imaging from fully-sampled low-energy data together with single-view high-energy data. We demonstrate the feasibility of the approach with two independent cohorts (the first cohort including contrast-enhanced DECT scans of 5753 image slices from 22 patients and the second cohort including spectral CT scans without contrast injection of 2463 image slices from other 22 patients) and show its superior performance on DECT applications. The deep-learning-based approach could be useful to further significantly reduce the radiation dose of current premium DECT scanners and has the potential to simplify the hardware of DECT imaging systems and to enable DECT imaging using standard SECT scanners.

International audience; Diabetic retinopathy (DR) is the most common eye complication of diabetes and one of the leading causes of blindness and vision impairment. Automated and accurate DR grading is of great significance for the timely and effective treatment of fundus diseases. Current clinical methods remain subject to potential time-consumption and high-risk. In this paper, a hierarchically Coarse-to-fine network (CF-DRNet) is proposed as an automatic clinical tool to classify five stages of DR severity grades using convolutional neural networks (CNNs). The CF-DRNet conforms to the hierarchical characteristic of DR grading and effectively improves the classification performance of five-class DR grading, which consists of the following: (1) The Coarse Network performs two-class classification including No DR and DR, where the attention gate module highlights the salient lesion features and suppresses irrelevant background information. (2) The Fine Network is proposed to classify four stages of DR severity grades of the grade DR from the Coarse Network including mild, moderate, severe non-proliferative DR (NPDR) and proliferative DR (PDR). Experimental results show that proposed CF-DRNet outperforms some state-of-art methods in the publicly available IDRiD and Kaggle fundus image datasets. These results indicate our method enables an efficient and reliable DR grading diagnosis in clinic.

International audience; Deep learning is a hot research topic in the field of machine learning methods and applications. Real-value neural networks (Real NNs), especially deep real networks (DRNs), have been widely used in many research fields. In recent years, the deep complex networks (DCNs) and the deep quaternion networks (DQNs) have attracted more and more attentions. The octonion algebra, which is an extension of complex algebra and quaternion algebra, can provide more efficient and compact expressions. This paper constructs a general framework of deep octonion networks (DONs) and provides the main building blocks of DONs such as octonion convolution, octonion batch normalization and octonion weight initialization; DONs are then used in image classification tasks for CIFAR-10 and CIFAR-100 data sets. Compared with the DRNs, the DCNs, and the DQNs, the proposed DONs have better convergence and higher classification accuracy. The success of DONs is also explained by multi-task learning.

International audience; The integration of segmentation and direct quantification on the left ventricle (LV) from the paired apical views(i.e., apical 4-chamber and 2-chamber together) echo sequence clinically achieves the comprehensive cardiac assessment multiview segmentation for anatomical morphology, and multidimensional quantification for contractile function. Direct quantification of LV, i.e., to automatically quantify multiple LV indices directly from the image via task-aware feature representation and regression, avoids accumulative error from the inter-step target. This integration sequentially makes a stereoscopical reflection of cardiac activity jointly from the paired orthogonal cross views sequences, overcoming limited observation with a single plane. We propose a K-shaped Unified Network (K-Net), the first end-to-end framework to simultaneously segment LV from apical 4-chamber and 2-chamber views, and directly quantify LV from major- and minor-axis dimensions (1D), area (2D), and volume (3D), in sequence. It works via four components 1) the K-Net architecture with the Attention Junction enables heterogeneous tasks learning of segmentation task of pixel-wise classification, and direct quantification task of image-wise regression, by interactively introducing the information from segmentation to jointly promote spatial attention map to guide quantification focusing on LV-related region, and transferring quantification feedback to make global constraint on segmentation; 2) the Bi-ResLSTMs distributed in K-Net layer-by-layer hierarchically extract spatial-temporal information in echo sequence, with bidirectional recurrent and short-cut connection to model spatial-temporal information among all frames; 3) the Information Valve tailing the Bi-ResLSTMs selectively exchanges information among multiple views, by stimulating complementary information and suppressing redundant information to make the efficient cross-flow for each view; 4) the Evolution Loss comprehensively guides sequential data learning, with static constraint for frame values, and dynamic constraint for inter-frame value changes. The experiments show that our K-Net gains high performance with a Dice coefficient up to 91.44% and a mean absolute error of the major-axis dimension down to 2.74mm, which reveal its clinical potential.

International audience; A novel network-based method to realistically simulate the guidewire position in the vascular system is presented in this paper. We can always obtain an ideal guidewire path by the shortest path algorithm, particularly, critical path algorithm based on the principle of maximum path length. Besides, we propose an adaptive sampling algorithm and optimizations based on two ends of the guidewire to adjust the guidewire position toward achieving an ideal simulation result. Compared with previous results, we can achieve 29%, 20% and 47% improvements in accuracy for three phantoms under the same conditions.

Background Renal cancer is one of the 10 most common cancers in human beings. The laparoscopic partial nephrectomy (LPN) is an effective way to treat renal cancer. Localization and delineation of the renal tumor from pre-operative CT Angiography (CTA) is an important step for LPN surgery planning. Recently, with the development of the technique of deep learning, deep neural networks can be trained to provide accurate pixel-wise renal tumor segmentation in CTA images. However, constructing the training dataset with a large amount of pixel-wise annotations is a time-consuming task for the radiologists. Therefore, weakly-supervised approaches attract more interest in research. Methods In this paper, we proposed a novel weakly-supervised convolutional neural network (CNN) for renal tumor segmentation. A three-stage framework was introduced to train the CNN with the weak annotations of renal tumors, i.e. the bounding boxes of renal tumors. The framework includes pseudo masks generation, group and weighted training phases. Clinical abdominal CT angiographic images of 200 patients were applied to perform the evaluation. Results Extensive experimental results show that the proposed method achieves a higher dice coefficient (DSC) of 0.826 than the other two existing weakly-supervised deep neural networks. Furthermore, the segmentation performance is close to the fully supervised deep CNN. Conclusions The proposed strategy improves not only the efficiency of network training but also the precision of the segmentation.

International audience; IntraVoxel Incoherent Motion (IVIM) Diffusion-Weighted Magnetic Resonance Imaging (DW-MRI) is of great interest for evaluating tissue diffusion and perfusion and producing parametric maps in clinical applications for liver pathologies. However, the presence of macroscopic blood vessels (not capillaries) in a given Region of Interest (ROI) results in a confounding effect that bias the quantification of tissue perfusion. Therefore, it is necessary to identify those voxels affected by blood vessels. In this paper, an efficient algorithm for an automatic identification of blood vessels in a given ROI is proposed. It relies on the sparsity of the spatial distribution of blood vessels. This sparsity prior can be easily incorporated using the all-voxel IVIM-MRI model introduced in this paper. In addition to the identification of blood vessels, the proposed algorithm provides a quantification of blood vessels, tissue diffusion and tissue perfusion of all voxels in a given ROI, in one single step. Besides, two strategies are proposed in this paper to deal with the nonnegativity of the model parameters. The efficiency of the proposed algorithm compared to the Non-Negative Least Square (NNLS)-based method, recently introduced to deal with the confounding blood vessel effect in the IVIM-MRI model, is confirmed using both realistic and real DW-MR images.

International audience; Compressed sensing MRI (CS-MRI) is considered as a powerful technique for decreasing the scan time of MRI while ensuring the image quality. However, state of the art reconstruction algorithms are still subjected to two challenges including terrible parameters tuning and image details loss resulted from over-smoothing. In this paper, we propose a deep frequency-division network (DFDN) to face these two image reconstruction issues. The proposed DFDN approach applies a deep iterative reconstruction network (DIRN) to replace the regularization terms and the corresponding parameters by a stacked convolution neural network (CNN). And then multiple DIRN blocks are cascaded continuously as one deeper neural network. Data consistency (DC) layer is incorporated after each DIRN block to correct the k-space data of intermediate results. Image content loss is computed after each DC layer and frequency-division loss is gained by weighting the high frequency loss and low frequency loss after each DIRN block. The combination of image content loss and frequency-division loss is considered as the total loss for constraining the network training procedure. Validations of the proposed method have been performed on two brain datasets. Visual results and quantitative evaluations show that the proposed DFDN algorithm has better performance in sparse MRI reconstruction than other comparative methods.

Summary: Various bioinformatics protocols have been developed for trimming the number of operational taxonomic units (OTUs) in phylogenetic datasets, but they typically require significant manual intervention. Here we present TreeTuner, a semiautomated pipeline that allows both coarse and fine-scale tuning of large protein sequence phylogenetic datasets via the minimization of OTU redundancy. TreeTuner facilitates preliminary investigation of such datasets as well as more rigorous downstream analysis of specific subsets of OTUs.For complete details on the use and execution of this protocol, please refer to Maruyama et al. (2013) and Sibbald et al. (2019).

International audience; To apply the Virtual Phantom (ViP) method for generating reference signals, to the chemical shift imaging (CSI) technique. The ViP method, previously used for generating reference images in magnetic resonance imaging (MRI), was here extended to generate reference peaks in the MR spectra acquired with a 2D-CSI sequence. Theoretical analysis was carried out to design the ViP signal. ViP-2D-CSI experiments were performed on a 4.7 Tesla MR scanner. Data analysis was carried out with the jMRUI software to fit the ViP peaks to determine their amplitude and frequency. Based on the theoretical analysis, it was found that the ViP peak could be generated by transmitting the ViP signal only in one repetition time of the 2D-CSI sequence. The amplitude and frequency of the ViP peak could be precisely adjusted and fine-tuned. The ViP peak was uniform over all CSI voxels, both in amplitude and in frequency. Furthermore, a good stability of the ViP reference amplitude and frequency was observed. The ViP method provides a means to generate a reference MR peak in 2D-CSI experiments. This could be of interest for signal quantification in CSI experiments.

[Aims]: Dust is observed in the polar regions of nearby active galactic nuclei (AGN) and it is known to contribute substantially to their mid-IR emission and to the obscuration of their UV to optical emission. We aim to carry out a statistical test to check whether this component is a common feature based on an analysis of the integrated spectral energy distributions of these composite sources., [Methods]: We selected a sample of 1275 broad-line AGN in the XMM-XXL field, with optical to infrared photometric data. These AGN are seen along their polar direction and we expect a maximal impact of dust located around the poles when it is present. We used X-CIGALE, which introduces a dust component to account for obscuration along the polar directions, modeled as a foreground screen, and an extinction curve that is chosen as it steepens significantly at short wavelengths or is much grayer. By comparing the results of different fits, we are able to define subsamples of sources with positive statistical evidence in favor of or against polar obscuration (if present) and described using the gray or steep extinction curve., [Results]: We find a similar fraction of sources with positive evidence for and against polar dust. Applying statistical corrections, we estimate that half of our sample could contain polar dust and among them, 60% exhibit a steep extinction curve and 40% a flat extinction curve; although these latter percentages are found to depend on the adopted extinction curves. The obscuration in the V-band is not found to correlate with the X-ray column density, while AV/NH ratios span a large range of values and higher dust temperatures are found with the flat, rather than with the steep extinction curve. Ignoring this polar dust component in the fit of the spectral energy distribution of these composite systems leads to an overestimation of the stellar contribution. A single fit with a polar dust component described with an SMC extinction curve efficiently overcomes this issue but it fails at identifying all the AGN with polar dust obscuration., The project has received funding from Excellence Initiative of Aix-Marseille University – AMIDEX, a French ‘Investissements d’Avenir’ programme. GM acknowledges support by the Agencia Estatal de Investigacíon, Unidad de Excelencia María de Maeztu, ref. MDM-2017-0765. M.S. acknowledges support by the Ministry of Education, Science and Technological Development of the Republic of Serbia through the contract no. 451-03- 9/2021-14/200002 and by the Science Fund of the Republic of Serbia, PROMIS 6060916, BOWIE.

International audience; This paper addresses the question of effective connectivity in the human cerebral cortex in the context of epilepsy. Among model based approaches to infer brain connectivity, spectral Dynamic Causal Modelling is a conventional technique for which we propose an alternative to estimate cross spectral density. The proposed strategy we investigated tackles the sub-estimation of the free energy using the well-known variational Expectation-Maximization algorithm highly sensitive to the initialization of the parameters vector by a permanent local adjustment of the initialization process. The performance of the proposed strategy in terms of effective connectivity identification is assessed using simulated data generated by a neuronal mass model (simulating unidirectional and bidirectional flows) and real epileptic intracerebral Electroencephalographic signals. Results show the efficiency of proposed approach compared to the conventional Dynamic Causal Modelling and the one wherein a deterministic annealing scheme is employed.

International audience; By providing fast scanning with low radiation doses, sparse-view (or sparse-projection) reconstruction has attracted much research attention in X-ray computerized tomography (CT) imaging. Recent contributions have demonstrated that the total variation (TV) constraint can lead to improved solution by regularizing the underdetermined ill-posed problem of sparse-view reconstruction. However, when the projection views are reduced below certain numbers, the performance of TV regularization tends to deteriorate with severe artifacts. In this paper, we explore the applicability of Gamma regularization for the sparse-view CT reconstruction. Experiments on simulated data and clinical data demonstrate that the Gamma regularization can lead to good performance in sparse-view reconstruction.

Electroencephalography (EEG) is a common tool used for the detection of epileptic seizures. However, the visual analysis of long-term EEG recordings is characterized by its subjectivity, time-consuming procedure and its erroneous detection. Various epileptic seizure detection algorithms have been proposed to deal with such issues. In this study, a novel automatic seizure-detection approach is proposed. Three different strategies are suggested to the user whereby he/she could choose the appropriate one for a given classification problem. Indeed, the feature extraction step, including both linear and nonlinear measures, is performed either directly from the EEG signals, or from the derived sub-bands of tunable-Q wavelet transform (TQWT), or even from the intrinsic mode functions (IMFs) of multivariate empirical mode decomposition (MEMD). The classification procedure is executed using a support vector machine (SVM). The performance of the proposed method is evaluated through a publicly available database from which six binary classification cases are formulated to discriminate between healthy, seizure and non-seizure EEG signals. Our results show high performance in terms of accuracy (ACC), sensitivity (SEN) and specificity (SPE) compared to the state-of-the-art approaches. Thus, the proposed approach for automatic seizure detection can be considered as a valuable alternative to existing methods, able to alleviate the overload of visual analysis and accelerate the seizure detection.

International audience; Background Using deep learning techniques in image analysis is a dynamically emerging field. This study aims to use a convolutional neural network (CNN), a deep learning approach, to automatically classify esophageal cancer (EC) and distinguish it from premalignant lesions. Methods A total of 1,272 white-light images were adopted from 748 subjects, including normal cases, premalignant lesions, and cancerous lesions; 1,017 images were used to train the CNN, and another 255 images were examined to evaluate the CNN architecture. Our proposed CNN structure consists of two subnetworks (O-stream and P-stream). The original images were used as the inputs of the O-stream to extract the color and global features, and the pre-processed esophageal images were used as the inputs of the P-stream to extract the texture and detail features. Results The CNN system we developed achieved an accuracy of 85.83%, a sensitivity of 94.23%, and a specificity of 94.67% after the fusion of the 2 streams was accomplished. The classification accuracy of normal esophagus, premalignant lesion, and EC were 94.23%, 82.5%, and 77.14%, respectively, which shows a better performance than the Local Binary Patterns (LBP) + Support Vector Machine (SVM) and Histogram of Gradient (HOG) + SVM methods. A total of 8 of the 35 (22.85%) EC lesions were categorized as premalignant lesions because of their slightly reddish and flat lesions. Conclusions The CNN system, with 2 streams, demonstrated high sensitivity and specificity with the endoscopic images. It obtained better detection performance than the currently used methods based on the same datasets and has great application prospects in assisting endoscopists to distinguish esophageal lesion subclasses.

International audience; This paper tackles the slowness issue of the well-known expectation-maximization (EM) algorithm in the context of Gaussian mixture models. To cope with this slowness problem, an Exact Line Search scheme is proposed. It is based on exact computation of the step size required to jump, for a given search direction, towards the final solution. Computing this exact step size is easily done by only rooting a second-order polynomial computed from the initial log-likelihood maximization problem. Numerical results using both simulated and real dataset showed the efficiency of the proposed exact line search scheme when applied to the conventional EM algorithm as well as the anti-annealing based acceleration techniques based on either the EM or the expectation conjugate gradient algorithm. © 2020 Elsevier B.V.

International audience; Purpose - Acute ischemic stroke is one of the most causes of death all over the world. Onset to treatment time is critical in stroke diagnosis and treatment. Considering the time consumption and high price of MR imaging, CT perfusion (CTP) imaging is strongly recommended for acute stroke. However, too much CT radiation during CTP imaging may increase the risk of health problems. How to reduce CT radiation dose in CT perfusion imaging has drawn our great attention. Methods - In this study, the original 30-pass CTP images are downsampled to 15 passes in time sequence, which equals to 50% radiation dose reduction. Then, a residual deep convolutional neural network (DCNN) model is proposed to restore the downsampled 15-pass CTP images to 30 passes to calculate the parameters such as cerebral blood flow, cerebral blood volume, mean transit time, time to peak for stroke diagnosis and treatment. The deep restoration CNN is implemented simply and effectively with 16 successive convolutional layers which form a wide enough receptive field for input image data. 18 patients' CTP images are employed as training set and the other six patients' CTP images are treated as test dataset in this study. Results - Experiments demonstrate that our CNN can restore high-quality CTP images in terms of structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR). The average SSIM and PSNR for test images are 0.981 and 56.25, and the SSIM and PSNR of regions of interest are 0.915 and 42.44, respectively, showing promising quantitative level. In addition, we compare the perfusion maps calculated from the restored images and from the original images, and the average perfusion results of them are extremely close. Areas of hypoperfusion of six test cases could be detected with comparable accuracy by radiologists. Conclusion - The trained model can restore the temporally downsampled 15-pass CTP to 30 passes very well. According to the contrast test, sufficient information cannot be restored with, e.g., simple interpolation method and deep convolutional generative adversarial network, but can be restored with the proposed CNN model. This method can be an optional way to reduce radiation dose during CTP imaging.

International audience; In computed tomography, high attenuation occurs when x-rays pass through a dense region or a long path in the scanning object. In this case, only limited photons reach the detector, which causes photon starvation artifacts. The artifacts usually appear as streaks along the directions with high attenuation. It might lower the discrimination of minor structures and lead to misdiagnosis. Applying a local filter to the projection data adaptively is a common solution, however, if the parameters of projection-based filter are not well selected, new artifacts and noise might appear in the final image. In this paper, a post image processing technique was developed to suppress the photon starvation streak artifacts. Based on the directional characteristics of streaks, a semi-adaptive anisotropic diffusion filter was applied to the high frequency sub-bands after wavelet transformation (WASA). Qualitative and quantitative experiments were performed on phantom data and clinical data to prove the effectiveness of this method for photon starvation artifact suppression.

International audience; Accurate segmentation of uterus, uterine fibroids, and spine from MR images is crucial for high intensity fo-cused ultrasound (HIFU) therapy but remains still difficult to achieve because of 1) the large shape and size variations among individuals, 2) the low contrast between adjacent organs and tissues, and 3) the unknown number of uterine fibroids. To tackle this problem, in this paper, we propose a large kernel Encoder-Decoder Network based on a 2D segmentation model. The use of this large kernel can capture multi-scale contexts by enlarging the valid receptive field. In addition, a deep multiple atrous convolution block is also employed to enlarge the receptive field and extract denser feature maps. Our approach is compared to both conventional and other deep learning methods and the experimental results conducted on a large dataset show its effectiveness.

International audience; Minimal path method has been widely recognized as an efficient tool for extracting vascular structures in medical imaging. In a previous paper, a method termed minimal path propagation with backtracking (MPP-BT) was derived to deal with curve-like structures such as vessel centerlines. A robust approach termed CMPP (constrained minimal path propagation) is here proposed to extend this work. The proposed method utilizes another minimal path propagation procedure to extract the complete vessel lumen after the centerlines have been found. Moreover, a process named local MPP-BT is applied to handle structure missing caused by the so-called close loop problems. This approach is fast and unsupervised with only one roughly set start point required in the whole process to get the entire vascular structure. A variety of datasets, including 2D cardiac angiography, 2D retinal images and 3D kidney CT angiography, are used for validation. A quantitative evaluation, together with a comparison to recently reported methods, is performed on retinal images for which a ground truth is available. The proposed method leads to specificity (Sp) and sensitivity (Se) values equal to 0.9750 and 0.6591. This evaluation is also extended to 3D synthetic vascular datasets and shows that the specificity (Sp) and sensitivity (Se) values are higher than 0.99. Parameter setting and computation cost are analyzed in this paper.

International audience; Multi-energy computed tomography (MCT) has a great potential in material decomposition, tissue characterization, lesion detection, and other applications. However, the severe noise that exists within projections makes it difficult to obtain high-quality MCT images. To overcome this limitation, we propose a method termed Spectral-Image Similarity-based Tensor with Enhanced-sparsity Reconstruction (SISTER) method. SISTER utilizes the non-local feature similarity in the spatial-spectral domain by clustering similar spatial-spectral patches within non-local window to a 4th-order tensor group. Compared with the image gradient L-0-norm with tensor dictionary learning (L0TDL) method, by adopting tensor decomposition rather than tensor dictionary learning, SISTER overcomes the instability of tensor dictionary. Besides, in our SISTER method the weight coefficients update strategy is also optimized. Both numerical simulation and preclinical dataset were performed to evaluate and validate the performance of SISTER. Qualitative and quantitative results show that the proposed method can lead to a promising improvement of edge preservation, finer feature recovery, and noise suppression.