The prediction of Intensive Care Unit (ICU) readmission has become a crucial area of research due to the increasing demand for ICU resources and the need to provide timely interventions to critically ill patients. In recent years, several studies have explored the use of statistical, machine learning (ML), and deep learning (DL) models to predict ICU readmission. This review paper presents an extensive overview of these studies and discusses the challenges associated with ICU readmission prediction. We categorize the studies based on the type of model used and evaluate their strengths and limitations. We also discuss the performance metrics used to evaluate the models and their potential clinical applications. In addition, this review explores current methodologies, data usage, and recent advances in interpretability and explainable AI for medical applications, offering insights to guide future research and development in this field. Finally, we identify gaps in the current literature and provide recommendations for future research. Recent advances like ML and DL have moderately improved the prediction of the risk of ICU readmission. However, more progress is needed to reach the precision required to build computerized decision support tools.

Designing dental crowns with computer-aided design software in dental laboratories is complex and time-consuming. Using real clinical datasets, we developed an end-to-end deep learning model that automatically generates personalized dental crown meshes. The input context includes the prepared tooth, its adjacent teeth, and the two closest teeth in the opposing jaw. The training set contains this context, the ground truth crown, and the extracted margin line. Our model consists of two components: First, a feature extractor converts the input point cloud into a set of local feature vectors, which are then fed into a transformer-based model to predict the geometric features of the crown. Second, a point-to-mesh module generates a dense array of points with normal vectors, and a differentiable Poisson surface reconstruction method produces an accurate crown mesh. Training is conducted with three losses: (1) a customized margin line loss; (2) a contrastive-based Chamfer distance loss; and (3) a mean square error (MSE) loss to control mesh quality. We compare our method with our previously published method, Dental Mesh Completion (DMC). Extensive testing confirms our method's superiority, achieving a 12.32% reduction in Chamfer distance and a 46.43% reduction in MSE compared to DMC. Margin line loss improves Chamfer distance by 5.59%.

Generating computational anatomical models of cerebrovascular networks is vital for improving clinical practice and understanding brain oxygen transport. This is achieved by extracting graph-based representations based on pre-mapping of vascular structures. Recent graphing methods can provide smooth vessels trajectories and well-connected vascular topology. However, they require water-tight surface meshes as inputs. Furthermore, adding vessels radii information on their graph compartments restricts their alignment along vascular centerlines. Here, we propose a novel graphing scheme that works with relaxed input requirements and intrinsically captures vessel radii information. The proposed approach is based on deforming geometric graphs constructed within vascular boundaries. Under a laplacian optimization framework, we assign affinity weights on the initial geometry that drives its iterative contraction toward vessels centerlines. We present a mechanism to decimate graph structure at each run and a convergence criterion to stop the process. A refinement technique is then introduced to obtain final vascular models. Our implementation is available on https://github.com/Damseh/VascularGraph. We benchmarked our results with that obtained using other efficient and state-of-the-art graphing schemes, validating on both synthetic and real angiograms acquired with different imaging modalities. The experiments indicate that the proposed scheme produces the lowest geometric and topological error rates on various angiograms. Furthermore, it surpasses other techniques in providing representative models that capture all anatomical aspects of vascular structures.

Intra-slice motion correction is an important step for analyzing volume variations and pathological formations from intravascular imaging. Optical coherence tomography (OCT) has been recently introduced for intravascular imaging and assessment of coronary artery disease. Two-dimensional (2-D) cross-sectional OCT images of coronary arteries play a crucial role to characterize the internal structure of the tissues. Adjacent images could be compounded; however, they might not fully match due to motion, which is a major hurdle for analyzing longitudinally each tissue in 3-D. The aim of this study is to develop a robust tissue-matching-based motion correction approach from a sequence of 2-D intracoronary OCT images. Our motion correction technique is based on the correlation between deep features obtained from a convolutional neural network (CNN) for each frame of a sequence. The optimal transformation of each frame is obtained by maximizing the similarity between the tissues of reference and moving frames. The results show a good alignment of the tissues after applying CNN features and determining the transformation parameters.

Graph models of cerebral vasculature derived from two-photon microscopy have shown to be relevant to study brain microphysiology. Automatic graphing of these microvessels remain problematic due to the vascular network complexity and two-photon sensitivity limitations with depth. In this paper, we propose a fully automatic processing pipeline to address this issue. The modeling scheme consists of a fully-convolution neural network to segment microvessels, a three-dimensional surface model generator, and a geometry contraction algorithm to produce graphical models with a single connected component. Based on a quantitative assessment using NetMets metrics, at a tolerance of 60 μm, false negative and false positive geometric error 19 rates are 3.8% and 4.2%, respectively, whereas false nega- 20 tive and false positive topological error rates are 6.1% and 4.5%, respectively. Our qualitative evaluation confirms the efficiency of our scheme in generating useful and accurate graphical models.

Obtaining the complete segmentation map of retinal lesions is the first step toward an automated diagnosis tool for retinopathy that is interpretable in its decision-making. However, the limited availability of ground truth lesion detection maps at a pixel level restricts the ability of deep segmentation neural networks to generalize over large databases. In this paper, we propose a novel approach for training a convolutional multi-task architecture with supervised learning and reinforcing it with weakly supervised learning. The architecture is simultaneously trained for three tasks: segmentation of red lesions and of bright lesions, those two tasks done concurrently with lesion detection. In addition, we propose and discuss the advantages of a new preprocessing method that guarantees the color consistency between the raw image and its enhanced version. Our complete system produces segmentations of both red and bright lesions. The method is validated at the pixel level and per-image using four databases and a cross-validation strategy. When evaluated on the task of screening for the presence or absence of lesions on the Messidor image set, the proposed method achieves an area under the ROC curve of 0.839, comparable with the state-of-the-art.

Retinal vessel segmentation constitutes an essential part of computer‐assisted tools for the diagnosis of ocular diseases. In this study, the authors propose an unsupervised retinal blood vessels segmentation approach based on the elite‐guided multi‐objective artificial bee colony (EMOABC) algorithm. The proposed method exploits several criteria simultaneously to improve the accuracy of the segmentation results. An energy curve function is used to calculate the values of the thresholding criteria, in order to reduce the noise response from lesions and select the optimal thresholds that separate the blood vessels from the background. In order to achieve computational speed up, a stopping criterion method is used to adjust the parameters of the EMOABC algorithm. The proposed method is computationally simple and faster than most of the available unsupervised algorithms, demonstrating fast convergence to the final segmentation. Additionally, the proposed vessel segmentation method outperforms the metaheuristics vessels segmentation algorithms reported in the literature. The achieved mean discrepancy metrics for the proposed approach are 94.5% accuracy, 97.4% specificity and 73.9% sensitivity for DRIVE database, and 94% accuracy, 96.2% specificity and 73.7% sensitivity for STARE database.

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.

Efficient implementation of non-linear activation functions is essential to the implementation of deep learning models on FPGAs. We introduce such an implementation based on the Discrete Cosine Transform Interpolation Filter (DCTIF). The proposed interpolation architecture combines simple arithmetic operations on the stored samples of the hyperbolic tangent function and on input data. It achieves almost 3× better precision than previous works while using a similar amount computational resources and a small amount of memory. Various combinations of DCTIF parameters can be chosen to trade off the accuracy and the overall circuit complexity of the tanh function. In one case, the proposed architecture approximates the hyperbolic tangent activation function with 0.004 maximum error while requiring only 1.45 kbits BRAM memory and 21 LUTs of a Virtex-7 FPGA.

A local feature descriptor for image analysis is a tool of interest for many applications. In this paper we propose an in context feature descriptor. An instance of this descriptor corresponding to a specific feature includes information from all other features in the image; it is a feature in context descriptor. This descriptor is thus unique for a feature in an ensemble of features. Many medical and industrial imaging applications are possible, one of them could be anomaly detection. Automatic anomaly detection could be implemented using this technique, thanks to the fact that this descriptor forms a metric space. Metric spaces are useful for indexing purposes and for statistical analysis.

The development of an automatic telemedicine system for computer-aided screening and grading of diabetic retinopathy depends on reliable detection of retinal lesions in fundus images. In this paper, a novel method for automatic detection of both microaneurysms and hemorrhages in color fundus images is described and validated. The main contribution is a new set of shape features, called Dynamic Shape Features, that do not require precise segmentation of the regions to be classified. These features represent the evolution of the shape during image flooding and allow to discriminate between lesions and vessel segments. The method is validated per-lesion and per-image using six databases, four of which are publicly available. It proves to be robust with respect to variability in image resolution, quality and acquisition system. On the Retinopathy Online Challenge's database, the method achieves a FROC score of 0.420 which ranks it fourth. On the Messidor database, when detecting images with diabetic retinopathy, the proposed method achieves an area under the ROC curve of 0.899, comparable to the score of human experts, and it outperforms state-of-the-art approaches.

The optical flow estimation is addressed in the context of video sequences, where temporal information can be exploited to increase the accuracy and the convergence speed of the algorithm. This paper presents an unsupervised optical flow algorithm based on robust Student's t data and regularization terms, which automatically tunes the relative weight of the data adequacy and regularization terms. The contribution of this paper is twofold. Firstly, it gives a more tractable and fully parallel formulation of the aforementioned algorithm, which significantly enhances the speed performances, and, secondly, it exploits the temporal smoothness information by introducing spatio-temporal regularization.

The main objective of this letter is to formulate a new approach of learning a Mahalanobis distance metric for nearest neighbor regression from a training sample set. We propose a modified version of the large margin nearest neighbor metric learning method to deal with regression problems. As an application, the prediction of post-operative trunk 3-D shapes in scoliosis surgery using nearest neighbor regression is described. Accuracy of the proposed method is quantitatively evaluated through experiments on real medical data.

Existing methods for surface matching are limited by the tradeoff between precision and computational efficiency. Here, we present an improved algorithm for dense vertex-to-vertex correspondence that uses direct matching of features defined on a surface and improves it by using spectral correspondence as a regularization. This algorithm has the speed of both feature matching and spectral matching while exhibiting greatly improved precision (distance errors of 1.4 percent). The method, FOCUSR, incorporates implicitly such additional features to calculate the correspondence and relies on the smoothness of the lowest-frequency harmonics of a graph Laplacian to spatially regularize the features. In its simplest form, FOCUSR is an improved spectral correspondence method that nonrigidly deforms spectral embeddings. We provide here a full realization of spectral correspondence where virtually any feature can be used as an additional information using weights on graph edges, but also on graph nodes and as extra embedded coordinates. As an example, the full power of FOCUSR is demonstrated in a real-case scenario with the challenging task of brain surface matching across several individuals. Our results show that combining features and regularizing them in a spectral embedding greatly improves the matching precision (to a submillimeter level) while performing at much greater speed than existing methods.

Besides the spinal deformity, scoliosis modifies notably the general appearance of the trunk resulting in trunk rotation, imbalance and asymmetries which constitutes patients' major concern. Existing classifications of scoliosis, based on the type of spinal curve as depicted on radiographs, are currently used to guide treatment strategies. Unfortunately, even though a perfect correction of the spinal curve is achieved, some trunk deformities remain, making patients dissatisfied with the treatment received. The purpose of this study is to identify possible shape patterns of trunk surface deformity associated with scoliosis. First, trunk surface is represented by a multivariate functional trunk shape descriptor based on 3D clinical measurements computed on cross sections of the trunk. Then, the classical formulation of hierarchical clustering is adapted to the case of multivariate functional data and applied to a set of 236 trunk surface 3D reconstructions. The highest internal validity is obtained when considering 11 clusters that explain up to 65% of the variance in our dataset. Our clustering result shows a concordance with the radiographic classification of spinal curves in 68% of the cases. As opposed to radiographic evaluation, the trunk descriptor is three-dimensional and its functional nature offers a compact and elegant description of not only the type, but also the severity and extent of the trunk surface deformity along the trunk length. In future work, new management strategies based on the resulting trunk shape patterns could be thought of in order to improve the esthetic outcome after treatment, and thus patients satisfaction.

Cardiac fibers, as well as their local arrangement in laminar sheets, have a complex spatial variation of their orientation that has an important role in mechanical and electrical cardiac functions. In this paper, a statistical atlas of this cardiac fiber architecture is built for the first time using human datasets. This atlas provides an average description of the human cardiac fiber architecture along with its variability within the population. In this study, the population is composed of ten healthy human hearts whose cardiac fiber architecture is imaged ex vivo with DT-MRI acquisitions. The atlas construction is based on a computational framework that minimizes user interactions and combines most recent advances in image analysis: graph cuts for segmentation, symmetric log-domain diffeomorphic demons for registration, and log-Euclidean metric for diffusion tensor processing and statistical analysis. Results show that the helix angle of the average fiber orientation is highly correlated to the transmural depth and ranges from -41° on the epicardium to +66° on the endocardium. Moreover, we find that the fiber orientation dispersion across the population (±13°) is lower than for the laminar sheets (±31°) . This study, based on human hearts, extends previous studies on other mammals with concurring conclusions and provides a description of the cardiac fiber architecture more specific to human and better suited for clinical applications. Indeed, this statistical atlas can help to improve the computational models used for radio-frequency ablation, cardiac resynchronization therapy, surgical ventricular restoration, or diagnosis and followups of heart diseases due to fiber architecture anomalies.

Thin Plate Splines are often used in image registration to model deformations. Its physical analogy involves a thin lying sheet of metal that is deformed and forced to pass through a set of control points. The Thin Plate Spline equation minimizes that thin plate bending energy. Rather than using Euclidean distances between control points for image deformation, we are using geodesic distances for image segmentation. Control points become seed points and force the thin plate to pass through given heights. Intuitively, the thin plate surface in the vicinity of a seed point within a region should have similar heights. The minimally bended thin plate actually gives a "confidence" map telling what the closest seed point is for every surface point. The Thin Plate Spline has a closed-form solution which is fast to compute and global optimal. This method shows comparable results to the Graph Cuts method.

This paper presents a unified framework for automatic segmentation of intervertebral disks of scoliotic spines from different types of magnetic resonance (MR) image sequences. The method exploits a combination of statistical and spectral texture features to discriminate closed regions representing intervertebral disks from background in MR images of the spine. Specific texture features are evaluated for three types of MR sequences acquired in the sagittal plane: 2-D spin echo, 3-D multiecho data image combination, and 3-D fast imaging with steady state precession. A total of 22 texture features (18 statistical and 4 spectral) are extracted from every closed region obtained from an automatic segmentation procedure based on the watershed approach. The feature selection step based on principal component analysis and clustering process permit to decide among all the extracted features which ones resulted in the highest rate of good classification. The proposed method is validated using a supervised k-nearest-neighbor classifier on 505 MR images coming from three different scoliotic patients and three different MR acquisition protocols. Results suggest that the selected texture features and classification can contribute to solve the problem of oversegmentation inherent to existing automatic segmentation methods by successfully discriminating intervertebral disks from the background on MRI of scoliotic spines.

This paper presents a novel 3-D reconstruction method of the scoliotic spine using prior vertebra models with image-based information taken from biplanar X-ray images. We first propose a global modeling approach by exploiting the 3-D scoliotic curve reconstructed from a coronal and sagittal X-ray image in order to generate an approximate statistical model from a 3-D database of scoliotic patients based on a transformation algorithm which incorporates intuitive geometrical properties. The personalized 3-D reconstruction of the spine is then achieved with a novel segmentation method which takes into account the variable appearance of scoliotic vertebrae (rotation, wedging) from standard quality images in order to segment and isolate individual vertebrae on the radiographic planes. More specifically, it uses prior 3-D models regulated from 2-D image level set functionals to identify and match corresponding bone structures on the biplanar X-rays. An iterative optimization procedure integrating similarity measures such as deformable vertebral contours regulated from high-level anatomical primitives, morphological knowledge and epipolar constraints is then applied to globally refine the 3-D anatomical landmarks on each vertebra level of the spine. This method was validated on twenty scoliotic patients by comparing results to a standard manual approach. The qualitative evaluation of the retro-projection of the vertebral contours confirms that the proposed method can achieve better consistency to the X-ray image's natural content. A comparison to synthetic models and real patient data also yields good accuracy on the localization of low-level primitives such as anatomical landmarks identified by an expert on each vertebra. The experiments reported in this paper demonstrate that the proposed method offers a better matching accuracy on a set of landmarks from biplanar views when compared to a manual technique for each evaluated cases, and its precision is comparable to 3-D models generated from magnetic resonance images, thus suitable for routine 3-D clinical assessment of spinal deformities.

This paper introduces a method to analyze the variability of the spine shape and of the spine shape deformations using articulated shape models. The spine shape was expressed as a vector of relative poses between local coordinate systems of neighboring vertebrae. Spine shape deformations were then modeled by a vector of rigid transformations that transforms one spine shape into another. Because rigid transformations do not naturally belong to a vector space, conventional mean and covariance could not be applied. The Fréchet mean and a generalized covariance were used instead. The spine shapes of a group of 295 scoliotic patients were quantitatively analyzed as well as the spine shape deformations associated with the Cotrel-Dubousset corrective surgery (33 patients), the Boston brace (39 patients), and the scoliosis progression without treatment (26 patients). The variability of intervertebral poses was found to be inhomogeneous (lumbar vertebrae were more variable than the thoracic ones) and anisotropic (with maximal rotational variability around the coronal axis and maximal translational variability along the axial direction). Finally, brace and surgery were found to have a significant effect on the Fréchet mean and on the generalized covariance in specific spine regions where treatments modified the spine shape.

We propose a new region segmentation method based on non-rigid template matching. We align a binary template to an image by maximizing the likelihood of intensity distributions within a region of interest and its background. The intensity model and the corresponding a posteriori distributions are estimated and updated throughout the alignment. The geometric deformation of the template is based on a fluid registration model. Unlike contour-based segmentation techniques, this registration framework allows for a global regularization of the template variations. This enables the segmentation of irregular shapes while avoiding leaks. We apply our method to the segmentation of the liver in computed tomography images, a challenging task due to the high inter-patient variability in the shape of this organ. We show that our segmentation results are equivalent or superior in accuracy to results obtained using existing techniques based on 3D shape models.

The main objective of this study was to develop a 3D reconstruction technique of the spine and rib cage of idiopathic scoliotic patients using the self-calibration of the imaging system. The proposed approach computes the intrinsic and extrinsic parameters of the radiographic setup with respect to the global coordinate system used at Ste-Justine Hospital. Our approach determines an optimal estimate of the geometrical parameters of the imaging system from a nonlinear minimization of the mean square distance between the observed and analytical projections of a set of matched points identified on a pair of radiographic views. The accuracy of the optimal estimate for the intrinsic parameters was significantly improved when geometric knowledge such as the known length of detectable straight bars is incorporated as a set of equality constraints in the optimization process. Furthermore, in order to retrieve the 3D structure of interest in the global coordinate system, a reference plane including the origin of the global coordinate system is specified. Computer simulations were performed to evaluate the self-calibration procedure and to determine the minimum knowledge required to obtain an accurate 3D reconstruction for clinical applications. An in vitro validation on real images of a dry cadaveric human spine showed that the method is feasible and reaches the expected accuracy.