Anatomically-aware uncertainty for semi-supervised image segmentation.

Semi-supervised learning relaxes the need of large pixel-wise labeled datasets for image segmentation by leveraging unlabeled data. A prominent way to exploit unlabeled data is to regularize model predictions. Since the predictions of unlabeled data can be unreliable, uncertainty-aware schemes are typically employed to gradually learn from meaningful and reliable predictions. Uncertainty estimation methods, however, rely on multiple inferences from the model predictions that must be computed for each training step, which is computationally expensive. Moreover, these uncertainty maps capture pixel-wise disparities and do not consider global information. This work proposes a novel method to estimate segmentation uncertainty by leveraging global information from the segmentation masks. More precisely, an anatomically-aware representation is first learnt to model the available segmentation masks. The learnt representation thereupon maps the prediction of a new segmentation into an anatomically-plausible segmentation. The deviation from the plausible segmentation aids in estimating the underlying pixel-level uncertainty in order to further guide the segmentation network. The proposed method consequently estimates the uncertainty using a single inference from our representation, thereby reducing the total computation. We evaluate our method on two publicly available segmentation datasets of left atria in cardiac MRIs and of multiple organs in abdominal CTs. Our anatomically-aware method improves the segmentation accuracy over the state-of-the-art semi-supervised methods in terms of two commonly used evaluation metrics.

Learning joint surface reconstruction and segmentation, from brain images to cortical surface parcellation.

Reconstructing and segmenting cortical surfaces from MRI is essential to a wide range of brain analyses. However, most approaches follow a multi-step slow process, such as a sequential spherical inflation and registration, which requires considerable computation times. To overcome the limitations arising from these multi-steps, we propose SegRecon, an integrated end-to-end deep learning method to jointly reconstruct and segment cortical surfaces directly from an MRI volume in one single step. We train a volume-based neural network to predict, for each voxel, the signed distances to multiple nested surfaces and their corresponding spherical representation in atlas space. This is, for instance, useful for jointly reconstructing and segmenting the white-to-gray-matter interface and the gray-matter-to-CSF (pial) surface. We evaluate the performance of our surface reconstruction and segmentation method with a comprehensive set of experiments on the MindBoggle, ABIDE and OASIS datasets. Our reconstruction error is found to be less than 0.52 mm and 0.97 mm in terms of average Hausdorff distance to the FreeSurfer generated surfaces. Likewise, the parcellation results show over 4% improvements in average Dice with respect to FreeSurfer, in addition to an observed drastic speed-up from hours to seconds of computation on a standard desktop station.

Active learning for medical image segmentation with stochastic batches.

The performance of learning-based algorithms improves with the amount of labelled data used for training. Yet, manually annotating data is particularly difficult for medical image segmentation tasks because of the limited expert availability and intensive manual effort required. To reduce manual labelling, active learning (AL) targets the most informative samples from the unlabelled set to annotate and add to the labelled training set. On the one hand, most active learning works have focused on the classification or limited segmentation of natural images, despite active learning being highly desirable in the difficult task of medical image segmentation. On the other hand, uncertainty-based AL approaches notoriously offer sub-optimal batch-query strategies, while diversity-based methods tend to be computationally expensive. Over and above methodological hurdles, random sampling has proven an extremely difficult baseline to outperform when varying learning and sampling conditions. This work aims to take advantage of the diversity and speed offered by random sampling to improve the selection of uncertainty-based AL methods for segmenting medical images. More specifically, we propose to compute uncertainty at the level of batches instead of samples through an original use of stochastic batches (SB) during sampling in AL. Stochastic batch querying is a simple and effective add-on that can be used on top of any uncertainty-based metric. Extensive experiments on two medical image segmentation datasets show that our strategy consistently improves conventional uncertainty-based sampling methods. Our method can hence act as a strong baseline for medical image segmentation. The code is available on: https://github.com/Minimel/StochasticBatchAL.git.

Real-time simulation of viscoelastic tissue behavior with physics-guided deep learning.

Finite element methods (FEM) are popular approaches for simulation of soft tissues with elastic or viscoelastic behavior. However, their usage in real-time applications, such as in virtual reality surgical training, is limited by computational cost. In this application scenario, which typically involves transportable simulators, the computing hardware severely constrains the size or the level of details of the simulated scene. To address this limitation, data-driven approaches have been suggested to simulate mechanical deformations by learning the mapping rules from FEM generated datasets. Prior data-driven approaches have ignored the physical laws of the underlying engineering problem and have consequently been restricted to simulation cases of simple hyperelastic materials where the temporal variations were effectively ignored. However, most surgical training scenarios require more complex hyperelastic models to deal with the viscoelastic properties of tissues. This type of material exhibits both viscous and elastic behaviors when subjected to external force, requiring the implementation of time-dependant state variables. Herein, we propose a deep learning method for predicting displacement fields of soft tissues with viscoelastic properties. The main contribution of this work is the use of a physics-guided loss function for the optimization of the deep learning model parameters. The proposed deep learning model is based on convolutional (CNN) and recurrent layers (LSTM) to predict spatiotemporal variations. It is augmented with a mass conservation law in the lost function to prevent the generation of physically inconsistent results. The deep learning model is trained on a set of FEM datasets that are generated from a commercially available state-of-the-art numerical neurosurgery simulator. The use of the physics-guided loss function in a deep learning model has led to a better generalization in the prediction of deformations in unseen simulation cases. Moreover, the proposed method achieves a better accuracy over the conventional CNN models, where improvements were observed in unseen tissue from 8% to 30% depending on the magnitude of external forces. It is hoped that the present investigation will help in filling the gap in applying deep learning in virtual reality simulators, hence improving their computational performance (compared to FEM simulations) and ultimately their usefulness.

Source-free domain adaptation for image segmentation.

Domain adaptation (DA) has drawn high interest for its capacity to adapt a model trained on labeled source data to perform well on unlabeled or weakly labeled target data from a different domain. Most common DA techniques require concurrent access to the input images of both the source and target domains. However, in practice, privacy concerns often impede the availability of source images in the adaptation phase. This is a very frequent DA scenario in medical imaging, where, for instance, the source and target images could come from different clinical sites. We introduce a source-free domain adaptation for image segmentation. Our formulation is based on minimizing a label-free entropy loss defined over target-domain data, which we further guide with weak labels of the target samples and a domain-invariant prior on the segmentation regions. Many priors can be derived from anatomical information. Here, a class-ratio prior is estimated from anatomical knowledge and integrated in the form of a Kullback-Leibler (KL) divergence in our overall loss function. Furthermore, we motivate our overall loss with an interesting link to maximizing the mutual information between the target images and their label predictions. We show the effectiveness of our prior-aware entropy minimization in a variety of domain-adaptation scenarios, with different modalities and applications, including spine, prostate and cardiac segmentation. Our method yields comparable results to several state-of-the-art adaptation techniques, despite having access to much less information, as the source images are entirely absent in our adaptation phase. Our straightforward adaptation strategy uses only one network, contrary to popular adversarial techniques, which are not applicable to a source-free DA setting. Our framework can be readily used in a breadth of segmentation problems, and our code is publicly available: https://github.com/mathilde-b/SFDA.

Attention-Based Dynamic Subspace Learners for Medical Image Analysis.

Learning similarity is a key aspect in medical image analysis, particularly in recommendation systems or in uncovering the interpretation of anatomical data in images. Most existing methods learn such similarities in the embedding space over image sets using a single metric learner. Images, however, have a variety of object attributes such as color, shape, or artifacts. Encoding such attributes using a single metric learner is inadequate and may fail to generalize. Instead, multiple learners could focus on separate aspects of these attributes in subspaces of an overarching embedding. This, however, implies the number of learners to be found empirically for each new dataset. This work, Dynamic Subspace Learners, proposes to dynamically exploit multiple learners by removing the need of knowing apriori the number of learners and aggregating new subspace learners during training. Furthermore, the visual interpretability of such subspace learning is enforced by integrating an attention module into our method. This integrated attention mechanism provides a visual insight of discriminative image features that contribute to the clustering of image sets and a visual explanation of the embedding features. The benefits of our attention-based dynamic subspace learners are evaluated in the application of image clustering, image retrieval, and weakly supervised segmentation. Our method achieves competitive results with the performances of multiple learners baselines and significantly outperforms the classification network in terms of clustering and retrieval scores on three different public benchmark datasets. Moreover, our method also provides an attention map generated directly during inference to illustrate the visual interpretability of the embedding features. These attention maps offer a proxy-labels, which improves the segmentation accuracy up to 15% in Dice scores when compared to state-of-the-art interpretation techniques.

State-of-the-art retinal vessel segmentation with minimalistic models.

The segmentation of retinal vasculature from eye fundus images is a fundamental task in retinal image analysis. Over recent years, increasingly complex approaches based on sophisticated Convolutional Neural Network architectures have been pushing performance on well-established benchmark datasets. In this paper, we take a step back and analyze the real need of such complexity. We first compile and review the performance of 20 different techniques on some popular databases, and we demonstrate that a minimalistic version of a standard U-Net with several orders of magnitude less parameters, carefully trained and rigorously evaluated, closely approximates the performance of current best techniques. We then show that a cascaded extension (W-Net) reaches outstanding performance on several popular datasets, still using orders of magnitude less learnable weights than any previously published work. Furthermore, we provide the most comprehensive cross-dataset performance analysis to date, involving up to 10 different databases. Our analysis demonstrates that the retinal vessel segmentation is far from solved when considering test images that differ substantially from the training data, and that this task represents an ideal scenario for the exploration of domain adaptation techniques. In this context, we experiment with a simple self-labeling strategy that enables moderate enhancement of cross-dataset performance, indicating that there is still much room for improvement in this area. Finally, we test our approach on Artery/Vein and vessel segmentation from OCTA imaging problems, where we again achieve results well-aligned with the state-of-the-art, at a fraction of the model complexity available in recent literature. Code to reproduce the results in this paper is released.

Manifold-aware synthesis of high-resolution diffusion from structural imaging.

The physical and clinical constraints surrounding diffusion-weighted imaging (DWI) often limit the spatial resolution of the produced images to voxels up to eight times larger than those of T1w images. The detailed information contained in accessible high-resolution T1w images could help in the synthesis of diffusion images with a greater level of detail. However, the non-Euclidean nature of diffusion imaging hinders current deep generative models from synthesizing physically plausible images. In this work, we propose the first Riemannian network architecture for the direct generation of diffusion tensors (DT) and diffusion orientation distribution functions (dODFs) from high-resolution T1w images. Our integration of the log-Euclidean Metric into a learning objective guarantees, unlike standard Euclidean networks, the mathematically-valid synthesis of diffusion. Furthermore, our approach improves the fractional anisotropy mean squared error (FA MSE) between the synthesized diffusion and the ground-truth by more than 23% and the cosine similarity between principal directions by almost 5% when compared to our baselines. We validate our generated diffusion by comparing the resulting tractograms to our expected real data. We observe similar fiber bundles with streamlines having <3% difference in length, <1% difference in volume, and a visually close shape. While our method is able to generate diffusion images from structural inputs in a high-resolution space within 15 s, we acknowledge and discuss the limits of diffusion inference solely relying on T1w images. Our results nonetheless suggest a relationship between the high-level geometry of the brain and its overall white matter architecture that remains to be explored.

Learnable Pooling in Graph Convolutional Networks for Brain Surface Analysis.

Brain surface analysis is essential to neuroscience, however, the complex geometry of the brain cortex hinders computational methods for this task. The difficulty arises from a discrepancy between 3D imaging data, which is represented in Euclidean space, and the non-Euclidean geometry of the highly-convoluted brain surface. Recent advances in machine learning have enabled the use of neural networks for non-Euclidean spaces. These facilitate the learning of surface data, yet pooling strategies often remain constrained to a single fixed-graph. This paper proposes a new learnable graph pooling method for processing multiple surface-valued data to output subject-based information. The proposed method innovates by learning an intrinsic aggregation of graph nodes based on graph spectral embedding. We illustrate the advantages of our approach with in-depth experiments on two large-scale benchmark datasets. The ablation study in the paper illustrates the impact of various factors affecting our learnable pooling method. The flexibility of the pooling strategy is evaluated on four different prediction tasks, namely, subject-sex classification, regression of cortical region sizes, classification of Alzheimer's disease stages, and brain age regression. Our experiments demonstrate the superiority of our learnable pooling approach compared to other pooling techniques for graph convolutional networks, with results improving the state-of-the-art in brain surface analysis.

Realistic image normalization for multi-Domain segmentation.

Image normalization is a building block in medical image analysis. Conventional approaches are customarily employed on a per-dataset basis. This strategy, however, prevents the current normalization algorithms from fully exploiting the complex joint information available across multiple datasets. Consequently, ignoring such joint information has a direct impact on the processing of segmentation algorithms. This paper proposes to revisit the conventional image normalization approach by, instead, learning a common normalizing function across multiple datasets. Jointly normalizing multiple datasets is shown to yield consistent normalized images as well as an improved image segmentation when intensity shifts are large. To do so, a fully automated adversarial and task-driven normalization approach is employed as it facilitates the training of realistic and interpretable images while keeping performance on par with the state-of-the-art. The adversarial training of our network aims at finding the optimal transfer function to improve both, jointly, the segmentation accuracy and the generation of realistic images. We have evaluated the performance of our normalizer on both infant and adult brain images from the iSEG, MRBrainS and ABIDE datasets. The results indicate that our contribution does provide an improved realism to the normalized images, while retaining a segmentation accuracy at par with the state-of-the-art learnable normalization approaches.

Non-uniform Label Smoothing for Diabetic Retinopathy Grading from Retinal Fundus Images with Deep Neural Networks.

Purpose: Introducing a new technique to improve deep learning (DL) models designed for automatic grading of diabetic retinopathy (DR) from retinal fundus images by enhancing predictions' consistency. Methods: A convolutional neural network (CNN) was optimized in three different manners to predict DR grade from eye fundus images. The optimization criteria were (1) the standard cross-entropy (CE) loss; (2) CE supplemented with label smoothing (LS), a regularization approach widely employed in computer vision tasks; and (3) our proposed non-uniform label smoothing (N-ULS), a modification of LS that models the underlying structure of expert annotations. Results: Performance was measured in terms of quadratic-weighted κ score (quad-κ) and average area under the receiver operating curve (AUROC), as well as with suitable metrics for analyzing diagnostic consistency, like weighted precision, recall, and F1 score, or Matthews correlation coefficient. While LS generally harmed the performance of the CNN, N-ULS statistically significantly improved performance with respect to CE in terms quad-κ score (73.17 vs. 77.69, P < 0.025), without any performance decrease in average AUROC. N-ULS achieved this while simultaneously increasing performance for all other analyzed metrics. Conclusions: For extending standard modeling approaches from DR detection to the more complex task of DR grading, it is essential to consider the underlying structure of expert annotations. The approach introduced in this article can be easily implemented in conjunction with deep neural networks to increase their consistency without sacrificing per-class performance. Translational Relevance: A straightforward modification of current standard training practices of CNNs can substantially improve consistency in DR grading, better modeling expert annotations and human variability.

Graph Convolutions on Spectral Embeddings for Cortical Surface Parcellation.

Neuronal cell bodies mostly reside in the cerebral cortex. The study of this thin and highly convoluted surface is essential for understanding how the brain works. The analysis of surface data is, however, challenging due to the high variability of the cortical geometry. This paper presents a novel approach for learning and exploiting surface data directly across multiple surface domains. Current approaches rely on geometrical simplifications, such as spherical inflations, a popular but costly process. For instance, the widely used FreeSurfer takes about 3 hours to parcellate brain surfaces on a standard machine. Direct learning of surface data via graph convolutions would provide a new family of fast algorithms for processing brain surfaces. However, the current limitation of existing state-of-the-art approaches is their inability to compare surface data across different surface domains. Surface bases are indeed incompatible between brain geometries. This paper leverages recent advances in spectral graph matching to transfer surface data across aligned spectral domains. This novel approach enables direct learning of surface data across compatible surface bases. It exploits spectral filters over intrinsic representations of surface neighborhoods. We illustrate the benefits of this approach with an application to brain parcellation. We validate the algorithm over 101 manually labeled brain surfaces. The results show a significant improvement in labeling accuracy over recent Euclidean approaches while gaining a drastic speed improvement over conventional methods.

HyperDense-Net: A Hyper-Densely Connected CNN for Multi-Modal Image Segmentation.

Recently, dense connections have attracted substantial attention in computer vision because they facilitate gradient flow and implicit deep supervision during training. Particularly, DenseNet that connects each layer to every other layer in a feed-forward fashion and has shown impressive performances in natural image classification tasks. We propose HyperDenseNet, a 3-D fully convolutional neural network that extends the definition of dense connectivity to multi-modal segmentation problems. Each imaging modality has a path, and dense connections occur not only between the pairs of layers within the same path but also between those across different paths. This contrasts with the existing multi-modal CNN approaches, in which modeling several modalities relies entirely on a single joint layer (or level of abstraction) for fusion, typically either at the input or at the output of the network. Therefore, the proposed network has total freedom to learn more complex combinations between the modalities, within and in-between all the levels of abstraction, which increases significantly the learning representation. We report extensive evaluations over two different and highly competitive multi-modal brain tissue segmentation challenges, iSEG 2017 and MRBrainS 2013, with the former focusing on six month infant data and the latter on adult images. HyperDenseNet yielded significant improvements over many state-of-the-art segmentation networks, ranking at the top on both benchmarks. We further provide a comprehensive experimental analysis of features re-use, which confirms the importance of hyper-dense connections in multi-modal representation learning. Our code is publicly available.

Spectral Shape Analysis of Human Torsos: Application to the Evaluation of Scoliosis Surgery Outcome.

This paper aims at evaluating the effect of spinal surgery on the torso shape appearance of adolescent patients. Current methods that assess the surgical outcome on the trunk shape are limited to its global asymmetry or rely on unreliable manual measurements. We introduce a novel framework to evaluate pre- to postoperative local asymmetry changes using a spectral representation of the torso shape, more specifically, the Laplacian spectrum (eigenvalues and eigenvectors) of a graph. We conduct a statistical analysis on the eigenvalues to efficiently select the spectral space and determine the significant components between preop and postop groups. On the selected eigenvectors, we propose a local analysis based on the concept of Euler characteristic to detect their local maxima and minima, which are then used to compute local left-right (L-R) asymmetries of torso shape. On 49 patients with a thoracic spinal deformity, the method captures significant pre- to postoperative changes of asymmetry at the waist, shoulder blades, shoulders, and breasts. We have evaluated average correction rates for L-R asymmetry of the waist height (67%), shoulder-blade height (64%) and depth (67%), lateral offset between shoulder and neck (61%), and breast height (52%). Spectral torso shape analysis provides a novel approach to quantify the surgical correction of the scoliotic trunk from local shape asymmetry. The proposed method could help the surgeon to understand the impact of different spinal surgery strategies on the postoperative appearance and choose the one that should provide better patient's satisfaction.

Erratum: Cortical folding of the preterm brain: a longitudinal analysis of extremely preterm born neonates using spectral matching.

[This corrects the article DOI: 10.1002/brb3.488.].

Statistical shape analysis of subcortical structures using spectral matching.

Studying morphological changes of subcortical structures often predicate neurodevelopmental and neurodegenerative diseases, such as Alzheimer's disease and schizophrenia. Hence, methods for quantifying morphological variations in the brain anatomy, including groupwise shape analyses, are becoming increasingly important for studying neurological disorders. In this paper, a novel groupwise shape analysis approach is proposed to detect regional morphological alterations in subcortical structures between two study groups, e.g., healthy and pathological subjects. The proposed scheme extracts smoothed triangulated surface meshes from segmented binary maps, and establishes reliable point-to-point correspondences among the population of surfaces using a spectral matching method. Mean curvature features are incorporated in the matching process, in order to increase the accuracy of the established surface correspondence. The mean shapes are created as the geometric mean of all surfaces in each group, and a distance map between these shapes is used to characterize the morphological changes between the two study groups. The resulting distance map is further analyzed to check for statistically significant differences between two populations. The performance of the proposed framework is evaluated on two separate subcortical structures (hippocampus and putamen). Furthermore, the proposed methodology is validated in a clinical application for detecting abnormal subcortical shape variations in Alzheimer's disease. Experimental results show that the proposed method is comparable to state-of-the-art algorithms, has less computational cost, and is more sensitive to small morphological variations in patients with neuropathologies.

Brain Transfer: Spectral Analysis of Cortical Surfaces and Functional Maps.

The study of brain functions using fMRI often requires an accurate alignment of cortical data across a population. Particular challenges are surface inflation for cortical visualizations and measurements, and surface matching or alignment of functional data on surfaces for group-level analyses. Present methods typically treat each step separately and can be computationally expensive. For instance, smoothing and matching of cortices often require several hours. Conventional methods also rely on anatomical features to drive the alignment of functional data between cortices, whereas anatomy and function can vary across individuals. To address these issues, we propose BrainTransfer, a spectral framework that unifies cortical smoothing, point matching with confidence regions, and transfer of functional maps, all within minutes of computation. Spectral methods decompose shapes into intrinsic geometrical harmonics, but suffer from the inherent instability of eigenbasis. This limits their accuracy when matching eigenbasis, and prevents the spectral transfer of functions. Our contributions consist of, first, the optimization of a spectral transformation matrix, which combines both, point correspondence and change of eigenbasis, and second, focused harmonics, which localize the spectral decomposition of functional data. BrainTransfer enables the transfer of surface functions across interchangeable cortical spaces, accounts for localized confidence, and gives a new way to perform statistics directly on surfaces. Benefits of spectral transfers are illustrated with a variability study on shape and functional data. Matching accuracy on retinotopy is increased over conventional methods.

Laplacian forests: semantic image segmentation by guided bagging.

This paper presents a new, efficient and accurate technique for the semantic segmentation of medical images. The paper builds upon the successful random decision forests model and improves on it by modifying the way in which randomness is injected into the tree training process. The contribution of this paper is two-fold. First, we replace the conventional bagging procedure (the uniform sampling of training images) with a guided bagging approach, which exploits the inherent structure and organization of the training image set. This allows the creation of decision trees that are specialized to a specific sub-type of images in the training set. Second, the segmentation of a previously unseen image happens via selection and application of only the trees that are relevant to the given test image. Tree selection is done automatically, via the learned image embedding, with more precisely a Laplacian eigenmap. We, therefore, call the proposed approach Laplacian Forests. We validate Laplacian Forests on a dataset of 256, manually segmented 3D CT scans of patients showing high variability in scanning protocols, resolution, body shape and anomalies. Compared with conventional decision forests, Laplacian Forests yield both higher training efficiency, due to the local analysis of the training image space, as well as higher segmentation accuracy, due to the specialization of the forest to image sub-types.

Multi-atlas spectral PatchMatch: application to cardiac image segmentation.

The automatic segmentation of cardiac magnetic resonance images poses many challenges arising from the large variation between different anatomies, scanners and acquisition protocols. In this paper, we address these challenges with a global graph search method and a novel spectral embedding of the images. Firstly, we propose the use of an approximate graph search approach to initialize patch correspondences between the image to be segmented and a database of labelled atlases, Then, we propose an innovative spectral embedding using a multi-layered graph of the images in order to capture global shape properties. Finally, we estimate the patch correspondences based on a joint spectral representation of the image and atlases. We evaluated the proposed approach using 155 images from the recent MICCAI SATA segmentation challenge and demonstrated that the proposed algorithm significantly outperforms current state-of-the-art methods on both training and test sets.