Encoding the Latent Posterior of Bayesian Neural Networks for Uncertainty Quantification.

Bayesian Neural Networks (BNNs) have long been considered an ideal, yet unscalable solution for improving the robustness and the predictive uncertainty of deep neural networks. While they could capture more accurately the posterior distribution of the network parameters, most BNN approaches are either limited to small networks or rely on constraining assumptions, e.g., parameter independence. These drawbacks have enabled prominence of simple, but computationally heavy approaches such as Deep Ensembles, whose training and testing costs increase linearly with the number of networks. In this work we aim for efficient deep BNNs amenable to complex computer vision architectures, e.g., ResNet-50 DeepLabv3+, and tasks, e.g., semantic segmentation and image classification, with fewer assumptions on the parameters. We achieve this by leveraging variational autoencoders (VAEs) to learn the interaction and the latent distribution of the parameters at each network layer. Our approach, called Latent-Posterior BNN (LP-BNN), is compatible with the recent BatchEnsemble method, leading to highly efficient (in terms of computation and memory during both training and testing) ensembles. LP-BNNs attain competitive results across multiple metrics in several challenging benchmarks for image classification, semantic segmentation, and out-of-distribution detection.

Tubular structures segmentation of pediatric abdominal-visceral ceCT images with renal tumors: Assessment, comparison and improvement.

Renal tubular structures, such as ureters, arteries and veins, are very important for building a complete digital 3D anatomical model of a patient. However, they can be challenging to segment from ceCT images due to their elongated shape, diameter variation and intra- and inter-patient contrast heterogeneity. This task is even more difficult in pediatric and pathological subjects, due to high inter-subject anatomical variations, potential presence of tumors, small volume of these structures compared to the surrounding, and small available labeled datasets. Given the limited literature on methods dedicated to children, and in order to find inspirational approaches, a complete assessment of state-of-the-art methods for the segmentation of renal tubular structures on ceCT images on adults is presented. Then, these methods are tested and compared on a private pediatric and pathological dataset of 79 abdominal-visceral ceCT images with arteriovenous phase acquisitions. To the best of our knowledge, both assessment and comparison in this specific case are novel. Eventually, we also propose a new loss function which leverages for the first time the use of vesselness functions on the predicted segmentation. We show that the combination of this loss function with state-of-the-art methods improves the topological coherence of the segmented tubular structures.2.

Super-resolution in brain positron emission tomography using a real-time motion capture system.

Super-resolution (SR) is a methodology that seeks to improve image resolution by exploiting the increased spatial sampling information obtained from multiple acquisitions of the same target with accurately known sub-resolution shifts. This work aims to develop and evaluate an SR estimation framework for brain positron emission tomography (PET), taking advantage of a high-resolution infra-red tracking camera to measure shifts precisely and continuously. Moving phantoms and non-human primate (NHP) experiments were performed on a GE Discovery MI PET/CT scanner (GE Healthcare) using an NDI Polaris Vega (Northern Digital Inc), an external optical motion tracking device. To enable SR, a robust temporal and spatial calibration of the two devices was developed as well as a list-mode Ordered Subset Expectation Maximization PET reconstruction algorithm, incorporating the high-resolution tracking data from the Polaris Vega to correct motion for measured line of responses on an event-by-event basis. For both phantoms and NHP studies, the SR reconstruction method yielded PET images with visibly increased spatial resolution compared to standard static acquisitions, allowing improved visualization of small structures. Quantitative analysis in terms of SSIM, CNR and line profiles were conducted and validated our observations. The results demonstrate that SR can be achieved in brain PET by measuring target motion in real-time using a high-resolution infrared tracking camera.

Manifold Learning via Linear Tangent Space Alignment (LTSA) for Accelerated Dynamic MRI With Sparse Sampling.

The spatial resolution and temporal frame-rate of dynamic magnetic resonance imaging (MRI) can be improved by reconstructing images from sparsely sampled k -space data with mathematical modeling of the underlying spatiotemporal signals. These models include sparsity models, linear subspace models, and non-linear manifold models. This work presents a novel linear tangent space alignment (LTSA) model-based framework that exploits the intrinsic low-dimensional manifold structure of dynamic images for accelerated dynamic MRI. The performance of the proposed method was evaluated and compared to state-of-the-art methods using numerical simulation studies as well as 2D and 3D in vivo cardiac imaging experiments. The proposed method achieved the best performance in image reconstruction among all the compared methods. The proposed method could prove useful for accelerating many MRI applications, including dynamic MRI, multi-parametric MRI, and MR spectroscopic imaging.

Instruments Segmentation in X-ray Fluoroscopic Images for Endoscopic Retrograde Cholangio Pancreatography.

In this work, we propose a method to segment endoscope and guidewire from 2D X-ray fluoroscopic images of an endoscopic retrograde cholangiopancreatography (ERCP). We used an improved U-Net model. We obtained a Dice score of 0.94±0.05 for endoscope segmentation and a Hausdorff distance of 24.26 pixels for the guidewire segmentation. These preliminary results pave the way for further applications aiming at aiding the medical procedure.

Computer-aided diagnosis tool for cervical cancer screening with weakly supervised localization and detection of abnormalities using adaptable and explainable classifier.

While pap test is the most common diagnosis methods for cervical cancer, their results are highly dependent on the ability of the cytotechnicians to detect abnormal cells on the smears using brightfield microscopy. In this paper, we propose an explainable region classifier in whole slide images that could be used by cyto-pathologists to handle efficiently these big images (100,000x100,000 pixels). We create a dataset that simulates pap smears regions and uses a loss, we call classification under regression constraint, to train an efficient region classifier (about 66.8% accuracy on severity classification, 95.2% accuracy on normal/abnormal classification and 0.870 KAPPA score). We explain how we benefit from this loss to obtain a model focused on sensitivity and, then, we show that it can be used to perform weakly supervised localization (accuracy of 80.4%) of the cell that is mostly responsible for the malignancy of regions of whole slide images. We extend our method to perform a more general detection of abnormal cells (66.1% accuracy) and ensure that at least one abnormal cell will be detected if malignancy is present. Finally, we experiment our solution on a small real clinical slide dataset, highlighting the relevance of our proposed solution, adapting it to be as easily integrated in a pathology laboratory workflow as possible, and extending it to make a slide-level prediction.

Automated neurosurgical stereotactic planning for intraoperative use: a comprehensive review of the literature and perspectives.

The creation of intracranial stereotactic trajectories, from entry point to target point, is still mostly done manually by the neurosurgeon. The development of automated stereotactic planning tools has been described in the literature. This systematic review aims to assess the effectiveness of stereotactic planning procedure automation and develop tools for patients undergoing neurosurgical stereotactic procedures. PubMed/MEDLINE, EMBASE, Google Scholar, CINAHL, PsycINFO, and Cochrane Register of Controlled Trials databases were searched from inception to September 1, 2019, at the exception of Google Scholar (from 1 January 2010 to September 1, 2019) in French and English. Eligible studies included all studies proposing automated stereotactic planning. A total of 1543 studies were screened. Forty-two studies were included in the systematic review, including 18 (42.9%) conference papers. The surgical procedures planned automatically were mainly deep brain stimulation (n = 14, 33.3%), stereoelectroencephalography (n = 12, 28.6%), and not specified (n = 10, 23.8%). The most frequently used surgical constraints to plan the trajectory were blood vessels (n = 32, 76.2%), cerebral sulci (n = 27, 64.3%), and cerebral ventricles (n = 23, 54.8%). The distance from blood vessels ranged from 1.96 to 4.78 mm for manual trajectories and from 2.47 to 7.0 mm for automated trajectories. At least one neurosurgeon was involved in 36 studies (85.7%). The automated stereotactic trajectory was preferred in 75.4% of the studied cases (range 30-92.9). Only 3 (7.1%) studies were multicentric. No study reported prospective use of the planning software. Stereotactic planning automation is a promising tool to provide valuable stereotactic trajectories for clinical applications.

Automated brain MRI metrics in the EPIRMEX cohort of preterm newborns: Correlation with the neurodevelopmental outcome at 2 years.

PURPOSE: The purpose of this study was to identify in the EPIRMEX cohort the correlations between MRI brain metrics, including diffuse excessive high signal intensities (DEHSI) obtained with an automated quantitative method and neurodevelopmental outcomes at 2 years. MATERIALS AND METHODS: A total of 390 very preterm infants (gestational age at birth≤32 weeks) who underwent brain MRI at term equivalent age at 1.5T (n=338) or 3T (n=52) were prospectively included. Using a validated algorithm, automated metrics of the main brain surfaces (cortical and deep gray matter, white matter, cerebrospinal fluid) and DEHSI with three thresholds were obtained. Linear adjust regressions were performed to assess the correlation between brain metrics with the ages and stages questionnaire (ASQ) score at 2 years. RESULTS: Basal ganglia and thalami, cortex and white matter surfaces positively and significantly correlated with the global ASQ score. For all ASQ sub-domains, basal ganglia and thalami surfaces significantly correlated with the scores. DEHSI was present in 289 premature newborns (74%) without any correlation with the ASQ score. Metrics of DEHSI were greater at 3T than at 1.5T. CONCLUSION: Brain MRI metrics obtained in our multicentric cohort correlate with the neurodevelopmental outcome at 2 years of age. The quantitative detection of DEHSI is not predictive of adverse outcomes. Our automated algorithm might easily provide useful predictive information in daily practice.

Motion correction for PET data using subspace-based real-time MR imaging in simultaneous PET/MR.

Image quality of positron emission tomography (PET) reconstructions is degraded by subject motion occurring during the acquisition. Magnetic resonance (MR)-based motion correction approaches have been studied for PET/MR scanners and have been successful at capturing regular motion patterns, when used in conjunction with surrogate signals (e.g. navigators) to detect motion. However, handling irregular respiratory motion and bulk motion remains challenging. In this work, we propose an MR-based motion correction method relying on subspace-based real-time MR imaging to estimate motion fields used to correct PET reconstructions. We take advantage of the low-rank characteristics of dynamic MR images to reconstruct high-resolution MR images at high frame rates from highly undersampled k-space data. Reconstructed dynamic MR images are used to determine motion phases for PET reconstruction and estimate phase-to-phase nonrigid motion fields able to capture complex motion patterns such as irregular respiratory and bulk motion. MR-derived binning and motion fields are used for PET reconstruction to generate motion-corrected PET images. The proposed method was evaluated on in vivo data with irregular motion patterns. MR reconstructions accurately captured motion, outperforming state-of-the-art dynamic MR reconstruction techniques. Evaluation of PET reconstructions demonstrated the benefits of the proposed method in terms of motion artifacts reduction, improving the contrast-to-noise ratio by up to a factor 3 and achieveing a target-to-background ratio up to 90% superior compared to standard/uncorrected methods. The proposed method can improve the image quality of motion-corrected PET reconstructions in clinical applications.

Comprehensive Review of 3D Segmentation Software Tools for MRI Usable for Pelvic Surgery Planning.

Patient-specific 3D modeling is the first step towards image-guided surgery, the actual revolution in surgical care. Pediatric and adolescent patients with rare tumors and malformations should highly benefit from these latest technological innovations, allowing personalized tailored surgery. This study focused on the pelvic region, located at the crossroads of the urinary, digestive, and genital channels with important vascular and nervous structures. The aim of this study was to evaluate the performances of different software tools to obtain patient-specific 3D models, through segmentation of magnetic resonance images (MRI), the reference for pediatric pelvis examination. Twelve software tools freely available on the Internet and two commercial software tools were evaluated using T2-w MRI and diffusion-weighted MRI images. The software tools were rated according to eight criteria, evaluated by three different users: automatization degree, segmentation time, usability, 3D visualization, presence of image registration tools, tractography tools, supported OS, and potential extension (i.e., plugins). A ranking of software tools for 3D modeling of MRI medical images, according to the set of predefined criteria, was given. This ranking allowed us to elaborate guidelines for the choice of software tools for pelvic surgical planning in pediatric patients. The best-ranked software tools were Myrian Studio, ITK-SNAP, and 3D Slicer, the latter being especially appropriate if nerve fibers should be included in the 3D patient model. To conclude, this study proposed a comprehensive review of software tools for 3D modeling of the pelvis according to a set of eight criteria and delivered specific conclusions for pediatric and adolescent patients that can be directly applied to clinical practice.

An iterative multi-atlas patch-based approach for cortex segmentation from neonatal MRI.

Brain structure analysis in the newborn is a major health issue. This is especially the case for preterm neonates, in order to obtain predictive information related to the child development. In particular, the cortex is a structure of interest, that can be observed in magnetic resonance imaging (MRI). However, neonatal MRI data present specific properties that make them challenging to process. In this context, multi-atlas approaches constitute an efficient strategy, taking advantage of images processed beforehand. The method proposed in this article relies on such a multi-atlas strategy. More precisely, it uses two paradigms: first, a non-local model based on patches; second, an iterative optimization scheme. Coupling both concepts allows us to consider patches related not only to the image information, but also to the current segmentation. This strategy is compared to other multi-atlas methods proposed in the literature. Experiments on dHCP datasets show that the proposed approach provides robust cortex segmentation results.

Objective Evaluation of Multiple Sclerosis Lesion Segmentation using a Data Management and Processing Infrastructure.

We present a study of multiple sclerosis segmentation algorithms conducted at the international MICCAI 2016 challenge. This challenge was operated using a new open-science computing infrastructure. This allowed for the automatic and independent evaluation of a large range of algorithms in a fair and completely automatic manner. This computing infrastructure was used to evaluate thirteen methods of MS lesions segmentation, exploring a broad range of state-of-theart algorithms, against a high-quality database of 53 MS cases coming from four centers following a common definition of the acquisition protocol. Each case was annotated manually by an unprecedented number of seven different experts. Results of the challenge highlighted that automatic algorithms, including the recent machine learning methods (random forests, deep learning, …), are still trailing human expertise on both detection and delineation criteria. In addition, we demonstrate that computing a statistically robust consensus of the algorithms performs closer to human expertise on one score (segmentation) although still trailing on detection scores.

The challenge of cerebral magnetic resonance imaging in neonates: A new method using mathematical morphology for the segmentation of structures including diffuse excessive high signal intensities.

Preterm birth is a multifactorial condition associated with increased morbidity and mortality. Diffuse excessive high signal intensity (DEHSI) has been recently described on T2-weighted MR sequences in this population and thought to be associated with neuropathologies. To date, no robust and reproducible method to assess the presence of white matter hyperintensities has been developed, perhaps explaining the current controversy over their prognostic value. The aim of this paper is to propose a new semi-automated framework to detect DEHSI on neonatal brain MR images having a particular pattern due to the physiological lack of complete myelination of the white matter. A novel method for semi- automatic segmentation of neonatal brain structures and DEHSI, based on mathematical morphology and on max-tree representations of the images is thus described. It is a mandatory first step to identify and clinically assess homogeneous cohorts of neonates for DEHSI and/or volume of any other segmented structures. Implemented in a user-friendly interface, the method makes it straightforward to select relevant markers of structures to be segmented, and if needed, apply eventually manual corrections. This method responds to the increasing need for providing medical experts with semi-automatic tools for image analysis, and overcomes the limitations of visual analysis alone, prone to subjectivity and variability. Experimental results demonstrate that the method is accurate, with excellent reproducibility and with very few manual corrections needed. Although the method was intended initially for images acquired at 1.5T, which corresponds to the usual clinical practice, preliminary results on images acquired at 3T suggest that the proposed approach can be generalized.

Dual-rotation C-arm cone-beam computed tomography to increase low-contrast detection.

PURPOSE: This paper investigates the capabilities of a dual-rotation C-arm cone-beam computed tomography (CBCT) framework to improve non-contrast-enhanced low-contrast detection for full volume or volume-of-interest (VOI) brain imaging. METHOD: The idea is to associate two C-arm short-scan rotational acquisitions (spins): one over the full detector field of view (FOV) at low dose, and one collimated to deliver a higher dose to the central densest parts of the head. The angular sampling performed by each spin is allowed to vary in terms of number of views and angular positions. Collimated data is truncated and does not contain measurement of the incoming X-ray intensities in air (air calibration). When targeting full volume reconstruction, the method is intended to act as a virtual bow-tie. When targeting VOI imaging, the method is intended to provide the minimum full detector FOV data that sufficiently corrects for truncation artifacts. A single dedicated iterative algorithm is described that handles all proposed sampling configurations despite truncation and absence of air calibration. RESULTS: Full volume reconstruction of dual-rotation simulations and phantom acquisitions are shown to have increased low-contrast detection for less dose, with respect to a single-rotation acquisition. High CNR values were obtained on 1% inserts of the Catphan® 515 module in 0.94 mm thick slices. Image quality for VOI imaging was preserved from truncation artifacts even with less than 10 non-truncated views, without using the sparsity a priori common to such context. CONCLUSION: A flexible dual-rotation acquisition and reconstruction framework is proposed that has the potential to improve low-contrast detection in clinical C-arm brain soft-tissue imaging.

Functional and high-resolution retinal imaging monitoring photoreceptor damage in acute macular neuroretinopathy.

BACKGROUND: To report functional and high-resolution retinal imaging abnormalities, including adaptive optics (AO) throughout the course of acute macular neuroretinopathy (AMNR). METHODS: Two female patients (four eyes) with a diagnosis of AMNR were observed at the Clinical Investigation Center, CHNO des Quinze-Vingts, Paris, France. The patients underwent detailed ophthalmic examination including best-corrected visual acuity, slit-lamp examination, kinetic and static perimetry, full-field and multifocal electroretinogram, infrared reflectance, autofluorescence imaging and spectral-domain optical coherence tomography (SD-OCT) and AO fundus imaging at presentation and during follow-up. RESULTS: Both cases showed concomitant loss of integrity of the outer retinal structures on SD-OCT, and marked abnormalities on AO imaging with disruption of the visibility of the cone mosaic. In the first case, photoreceptor damage was seen to progress during several weeks before healing. In both cases, there were persistent morphological abnormalities of photoreceptors 1 year after onset. CONCLUSION: This study further highlights the value of AO fundus imaging to facilitate detection, mapping, and monitoring of damage to the cone outer segments during AMNR. In particular, residual damage to the cone mosaic can be precisely documented.

A flexible patch based approach for combined denoising and contrast enhancement of digital X-ray images.

Denoising and contrast enhancement play key roles in optimizing the trade-off between image quality and X-ray dose. However, these tasks present multiple challenges raised by noise level, low visibility of fine anatomical structures, heterogeneous conditions due to different exposure parameters, and patient characteristics. This work proposes a new method to address these challenges. We first introduce a patch-based filter adapted to the properties of the noise corrupting X-ray images. The filtered images are then used as oracles to define non parametric noise containment maps that, when applied in a multiscale contrast enhancement framework, allow optimizing the trade-off between improvement of the visibility of anatomical structures and noise reduction. A significant amount of tests on both phantoms and clinical images has shown that the proposed method is better suited than others for visual inspection for diagnosis, even when compared to an algorithm used to process low dose images in clinical routine.

A challenging issue: Detection of white matter hyperintensities in neonatal brain MRI.

The progress of magnetic resonance imaging (MRI) allows for a precise exploration of the brain of premature infants at term equivalent age. The so-called DEHSI (diffuse excessive high signal intensity) of the white matter of premature brains remains a challenging issue in terms of definition, and thus of interpretation. We propose a semi-automatic detection and quantification method of white matter hyperintensities in MRI relying on morphological operators and max-tree representations, which constitutes a powerful tool to help radiologists to improve their interpretation. Results show better reproducibility and robustness than interactive segmentation.

The Negative Cone Mosaic: A New Manifestation of the Optical Stiles-Crawford Effect in Normal Eyes.

PURPOSE: The purpose of this study was to describe a previously unreported manifestation of the optical Stiles-Crawford effect (oSCE) in normal eyes. METHODS: In a cohort of 50 normal subjects, the directional reflectance of cones in the retinal periphery was explored by flood-illuminated adaptive optics (FIAO) and optical coherence tomography (OCT). RESULTS: In 32 eyes (64%), off-axis FIAO images of the retinal periphery (∼15-20° from the fovea) showed variably sized patches of hyporeflective dots (called here negative mosaic) coexisting with hyperreflective (positive) cones. In nine cases, shifting the entry pupil toward the optical axis restored the positive cone mosaic, with a point-by-point correspondence between positive and negative mosaics. Rods remained hyperreflective around negative and positive cones. These changes were paralleled by changes of the OCT reflectance of the cone outer segment tips and, to a lesser extent, of the inner/outer segment limit. CONCLUSIONS: By en face FIAO imaging of the retina, the contrast of cones over rods may be strongly dependent on the entry pupil to such an extent that their reflectance is lower than that of rods. We hypothesized that the negative cone mosaic aspect results from the differential Stiles-Crawford effect of cones and rods. Cone reflectance by en face FIAO parallels the reflectance from the cone outer segment tip line and to a lesser extent of the inner/outer segment limit by OCT. Taking this into account, the oSCE is of importance for the interpretation of high-resolution images of photoreceptors. (ClinicalTrials.gov number, NCT01546181.)

Venous Nicking Without Arteriovenous Contact: The Role of the Arteriolar Microenvironment in Arteriovenous Nickings.

IMPORTANCE: Arteriovenous nickings (AVNs) in the retina are the cause of retinal vein occlusions and are also surrogates of cerebrovascular aging. The prevalent mechanistic model of AVNs stating that arteries crush veins remains somewhat unchallenged despite the lack of evidence other than fundus photographs. Here, we observed that venous nicking may be observed in the absence of physical contact with an arteriole. OBSERVATIONS: This observational study, conducted from January 2013 to September 2014, included 7 patients showing remodeling of a venous segment close to a retinal arteriole without arteriovenous overlap were imaged by adaptive optics imaging. Affected venous segments showed a variable association of nicking, narrowing, deviation, and opacification. Venous segments were deviated toward the arterioles in 6 of the 7 cases. The degree of venous narrowing ranged from 40% to 77%, while at these sites, the width of the intervascular space ranged from 16 µm to 42 µm. Similar features were identified in typical AVNs. CONCLUSIONS AND RELEVANCE: Arteriovenous nickings do not necessarily involve an arteriovenous compression. Instead, the topology of venous changes suggests a retractile process originating in the intervascular space. These findings have important implications for the understanding of retinal vein occlusions and of cerebrovascular aging.

Multi-organ localization with cascaded global-to-local regression and shape prior.

We propose a method for fast, accurate and robust localization of several organs in medical images. We generalize the global-to-local cascade of regression random forest to multiple organs. A first regressor encodes the global relationships between organs, learning simultaneously all organs parameters. Then subsequent regressors refine the localization of each organ locally and independently for improved accuracy. By combining the regression vote distribution and the organ shape prior (through probabilistic atlas representation) we compute confidence maps that are organ-dedicated probability maps. They are used within the cascade itself, to better select the test voxels for the second set of regressors, and to provide richer information than the classical bounding boxes result thanks to the shape prior. We propose an extensive study of the different learning and testing parameters, showing both their robustness to reasonable perturbations and their influence on the final algorithm accuracy. Finally we demonstrate the robustness and accuracy of our approach by evaluating the localization of six abdominal organs (liver, two kidneys, spleen, gallbladder and stomach) on a large and diverse database of 130 CT volumes. Moreover, the comparison of our results with two existing methods shows significant improvements brought by our approach and our deep understanding and optimization of the parameters.