Image Analysis

We introduce Point2SSM, a novel unsupervised learning approach that can accurately construct correspondence-based statistical shape models (SSMs) of anatomy directly from point clouds. SSMs are crucial in clinical research for analyzing the population-level morphological variation in bones and organs. However, traditional methods for creating SSMs have limitations that hinder their widespread adoption, such as the need for noise-free surface meshes or binary volumes, reliance on assumptions or predefined templates, and simultaneous optimization of the entire cohort leading to lengthy inference times given new data. Point2SSM overcomes these barriers by providing a data-driven solution that infers SSMs directly from raw point clouds, reducing inference burdens and increasing applicability as point clouds are more easily acquired. Deep learning on 3D point clouds has seen recent success in unsupervised representation learning, point-to-point matching, and shape correspondence; however, their application to constructing SSMs of anatomies is largely unexplored. In this work, we benchmark state-of-the-art point cloud deep networks on the task of SSM and demonstrate that they are not robust to the challenges of anatomical SSM, such as noisy, sparse, or incomplete input and significantly limited training data. Point2SSM addresses these challenges via an attention-based module that provides correspondence mappings from learned point features. We demonstrate that the proposed method significantly outperforms existing networks in terms of both accurate surface sampling and correspondence, better capturing population-level statistics.

Statistical shape modeling (SSM) is an essential tool for analyzing variations in anatomical morphology. In a typical SSM pipeline, 3D anatomical images, gone through segmentation and rigid registration, are represented using lower-dimensional shape features, on which statistical analysis can be performed. Various methods for constructing compact shape representations have been proposed, but they involve laborious and costly steps. We propose Image2SSM, a novel deep-learning-based approach for SSM that leverages image-segmentation pairs to learn a radial-basis-function (RBF)-based representation of shapes directly from images. This RBF-based shape representation offers a rich self-supervised signal for the network to estimate a continuous, yet compact representation of the underlying surface that can adapt to complex geometries in a data-driven manner. Image2SSM can characterize populations of biological structures of interest by constructing statistical landmark-based shape models of ensembles of anatomical shapes while requiring minimal parameter tuning and no user assistance. Once trained, Image2SSM can be used to infer low-dimensional shape representations from new unsegmented images, paving the way toward scalable approaches for SSM, especially when dealing with large cohorts. Experiments on synthetic and real datasets show the efficacy of the proposed method compared to the state-of-art correspondence-based method for SSM.

Statistical shape modeling is the computational process of discovering significant shape parameters from segmented anatomies captured by medical images (such as MRI and CT scans), which can fully describe subject-specific anatomy in the context of a population. The presence of substantial non-linear variability in human anatomy often makes the traditional shape modeling process challenging. Deep learning techniques can learn complex non-linear representations of shapes and generate statistical shape models that are more faithful to the underlying population-level variability. However, existing deep learning models still have limitations and require established/optimized shape models for training. We propose Mesh2SSM, a new approach that leverages unsupervised, permutation-invariant representation learning to estimate how to deform a template point cloud to subject-specific meshes, forming a correspondence-based shape model. Mesh2SSM can also learn a population-specific template, reducing any bias due to template selection. The proposed method operates directly on meshes and is computationally efficient, making it an attractive alternative to traditional and deep learning-based SSM approaches.

Statistical Shape Modeling (SSM) is a valuable tool for investigating and quantifying anatomical variations within populations of anatomies. However, traditional correspondence-based SSM generation methods require a time-consuming re-optimization process each time a new subject is added to the cohort, making the inference process prohibitive for clinical research. Additionally, they require complete geometric proxies (e.g., high-resolution binary volumes or surface meshes) as input shapes to construct the SSM. Unordered 3D point cloud representations of shapes are more easily acquired from various medical imaging practices (e.g., thresholded images and surface scanning). Point cloud deep networks have recently achieved remarkable success in learning permutation-invariant features for different point cloud tasks (e.g., completion, semantic segmentation, classification). However, their application to learning SSM from point clouds is to-date unexplored. In this work, we demonstrate that existing point cloud encoder-decoder-based completion networks can provide an untapped potential for SSM, capturing population-level statistical representations of shapes while reducing the inference burden and relaxing the input requirement. We discuss the limitations of these techniques to the SSM application and suggest future improvements. Our work paves the way for further exploration of point cloud deep learning for SSM, a promising avenue for advancing shape analysis literature and broadening SSM to diverse use cases.

Statistical shape modeling (SSM) enables population-based quantitative analysis of anatomical shapes, informing clinical diagnosis. Deep learning approaches predict correspondence-based SSM directly from unsegmented 3D images but require calibrated uncertainty quantification, motivating Bayesian formulations. Variational information bottleneck DeepSSM (VIB-DeepSSM) is an effective, principled framework for predicting probabilistic shapes of anatomy from images with aleatoric uncertainty quantification. However, VIB is only half-Bayesian and lacks epistemic uncertainty inference. We derive a fully Bayesian VIB formulation from both the probably approximately correct (PAC)-Bayes and variational inference perspectives. We demonstrate the efficacy of two scalable approaches for Bayesian VIB with epistemic uncertainty: concrete dropout and batch ensemble. Additionally, we introduce a novel combination of the two that further enhances uncertainty calibration via multimodal marginalization. Experiments on synthetic shapes and left atrium data demonstrate that the fully Bayesian VIB network predicts SSM from images with improved uncertainty reasoning without sacrificing accuracy.

Semantic segmentation is a critical step in automated image interpretation and analysis where pixels are classified into one or more predefined semantically meaningful classes. Deep learning approaches for semantic segmentation rely on harnessing the power of annotated images to learn features indicative of these semantic classes. Nonetheless, they often fail to generalize when there is a significant domain (i.e., distributional) shift between the training (i.e., source) data and the dataset(s) encountered when deployed (i.e., target), necessitating manual annotations for the target data to achieve acceptable performance. This is especially important in medical imaging because different image modalities have significant intra- and inter-site variations due to protocol and vendor variability. Current techniques are sensitive to hyperparameter tuning and target dataset size. This paper presents an unsupervised domain adaptation approach for semantic segmentation that alleviates the need for annotating target data. Using kernel density estimation, we match the target data distribution to the source data in the feature space. We demonstrate that our results are comparable or superior on multiple-site prostate MRI and histopathology images, which mitigates the need for annotating target data.

Homography estimation is a basic image-alignment method in many applications. Recently, with the development of convolutional neural networks (CNNs), some learning based approaches have shown great success in this task. However, the performance across different domains has never been researched. Unlike other common tasks (e.g., classification, detection, segmentation), CNN based homography estimation models show a domain shift immunity, which means a model can be trained on one dataset and tested on another without any transfer learning. To explain this unusual performance, we need to determine how CNNs estimate homography. In this study, we first show the domain shift immunity of different deep homography estimation models. We then use a shallow network with a specially designed dataset to analyze the features used for estimation. The results show that networks use low-level texture information to estimate homography. We also design some experiments to compare the performance between different texture densities and image features distorted on some common datasets to demonstrate our findings. Based on these findings, we provide an explanation of the domain shift immunity of deep homography estimation.

Neural operator learning as a means of mapping between complex function spaces has garnered significant attention in the field of computational science and engineering (CS&E). In this paper, we apply Neural operator learning to the time-of-flight ultrasound computed tomography (USCT) problem. We learn the mapping between time-of-flight (TOF) data and the heterogeneous sound speed field using a full-wave solver to generate the training data. This novel application of operator learning circumnavigates the need to solve the computationally intensive iterative inverse problem. The operator learns the non-linear mapping offline and predicts the heterogeneous sound field with a single forward pass through the model. This is the first time operator learning has been used for ultrasound tomography and is the first step in potential real-time predictions of soft tissue distribution for tumor identification in beast imaging.

Introduction:
The diagnosis and management of metopic craniosynostosis involves subjective decision-making at the point of care. The purpose of this work is to describe a quantitative severity metric and point-of-care user interface to aid clinicians in the management of metopic craniosynostosis and to provide a platform for future research through deep phenotyping.

Methods:
Two machine-learning algorithms were developed that quantify the severity of craniosynostosis – a supervised model specific to metopic craniosynostosis (Metopic Severity Score) and an unsupervised model used for cranial morphology in general (Cranial Morphology Deviation). CT imaging from multiple institutions were compiled to establish the spectrum of severity and a point-of-care tool was developed and validated.

Results:
Over the study period (2019-2021), 254 patients with metopic craniosynostosis and 92 control patients who underwent CT scan between the ages of 6 and 18 months were included. Scans were processed using an unsupervised machine-learning based dysmorphology quantification tool, CranioRate TM. The average Metopic severity score (MSS) for normal controls was 0.0±1.0 and for metopic synostosis was 4.9±2.3 (p<0.001). The average Cranial Morphology Deviation (CMD) for normal controls was 85.2±19.2 and for metopic synostosis was 189.9±43.4 (p<0.001). A point-of-care user interface (craniorate.org) has processed 46 CT images from 10 institutions.

The microscopic evaluation of glands in the colon is of utmost importance in the diagnosis of inflammatory bowel disease (IBD) and cancer. When properly trained, deep learning pipelines can provide a systematic, reproducible, and quantitative assessment of disease-related changes in glandular tissue architecture. The training and testing of deep learning models require large amounts of manual annotations, which are difficult, time-consuming, and expensive to obtain. Here, we propose a method for the automated generation of ground truth in digital H&E slides using immunohistochemistry (IHC) labels. The image processing pipeline generates annotations of glands in H&E histopathology images from colon biopsies by transfer of gland masks from CK8/18, CDX2, or EpCAM IHC. The IHC gland outlines are transferred to co-registered H&E images for the training of deep learning models. We compare the performance of the deep learning models to manual annotations using an internal held-out set of biopsies as well as two public data sets. Our results show that EpCAM IHC provides gland outlines that closely match manual gland annotations (DICE = 0.89) and are robust to damage by inflammation. In addition, we propose a simple data sampling technique that allows models trained on data from several sources to be adapted to a new data source using just a few newly annotated samples. The best-performing models achieved average DICE scores of 0.902 and 0.89, respectively, on GLAS and CRAG colon cancer public datasets when trained with only 10% of annotated cases from either public cohort. Altogether, the performances of our models indicate that automated annotations using cell type-specific IHC markers can safely replace manual annotations. The automated IHC labels from single institution cohorts can be combined with small numbers of hand-annotated cases from multi-institutional cohorts to train models that generalize well to diverse data sources.

Traditionally, two-dimensional conventional radiographs have been the primary tool to measure the complex morphology of the foot and ankle. However, the subtalar, talonavicular, and calcaneocuboid joints are challenging to assess due to their bone morphology and locations within the ankle. Weightbearing computed tomography is a novel high-resolution volumetric imaging mechanism that allows detailed generation of 3D bone reconstructions. This study aimed to develop a multi-domain statistical shape model to assess morphologic and alignment variation of the subtalar, talonavicular, and calcaneocuboid joints across an asymptomatic population and calculate 3D joint measurements in a consistent weightbearing position. Specific joint measurements included joint space distance, congruence, and coverage. Noteworthy anatomical variation predominantly included the talus and calcaneus, specifically an inverse relationship regarding talar dome heightening and calcaneal shortening. While there was minimal navicular and cuboid shape variation, there were alignment variations within these joints; the most notable is the rotational aspect about the anterior-posterior axis. This study also found that multi-domain modeling may be able to predict joint space distance measurements within a population. Additionally, variation across a population of these four bones may be driven far more by morphology than by alignment variation based on all three joint measurements. These data are beneficial in furthering our understanding of joint-level morphology and alignment variants to guide advancements in ankle joint pathological care and operative treatments.

Large-scale three-dimensional (3D) microscopy acquisitions fre-quently create terabytes of image data at high resolution and magni-fication. Imaging large specimens at high magnifications requires acquiring 3D overlapping image stacks as tiles arranged on a two-dimensional (2D) grid that must subsequently be aligned and fused into a single 3D volume. Due to their sheer size, aligning many overlapping gigabyte-sized 3D tiles in parallel and at full resolution is memory intensive and often I/O bound. Current techniques trade accuracy for scalability, perform alignment on subsampled images, and require additional postprocess algorithms to refine the alignment quality, usually with high computational requirements. One common solution to the memory problem is to subdivide the overlap region into smaller chunks (sub-blocks) and align the sub-block pairs in parallel, choosing the pair with the most reliable alignment to determine the global transformation. Yet aligning all sub-block pairs at full resolution remains computationally expensive. The key to quickly developing a fast, high-quality, low-memory solution is to identify a single or a small set of sub-blocks that give good alignment at full resolution without touching all the overlapping data. In this paper, we present a new iterative approach that leverages coarse resolution alignments to progressively refine and align only the promising candidates at finer resolutions, thereby aligning only a small user-defined number of sub-blocks at full resolution to determine the lowest error transformation between pairwise overlapping tiles. Our progressive approach is 2.6x faster than the state of the art, requires less than 450MB of peak RAM (per parallel thread), and offers a higher quality alignment without the need for additional postprocessing refinement steps to correct for alignment errors.

Background
We used panoramic images and neural networks to measure street-level built environment features with relevance to pedestrian safety.

Methods
Street-level features were identified from systematic literature search and local experience in Bogota, Colombia (study location). Google Street View© panoramic images were sampled from 10,810 intersection and street segment locations, including 2,642 where pedestrian collisions occurred 2015–2019; the most recent, nearest (<25 meters) available image was selected for each sampled intersection or segment. Human raters annotated image features which were used to train neural networks. Neural networks and human raters were compared across all features using mean Average Recall (mAR) and mean Average Precision (mAP) estimated performance. Feature prevalence was compared by pedestrian vs non-pedestrian collision locations.

Results
Thirty features were identified related to roadway (e.g., medians), crossing areas (e.g., crosswalk), traffic control (e.g., pedestrian signal), and roadside (e.g., trees) with streetlights the most frequently detected object (N=10,687 images). Neural networks achieved mAR=15.4 versus 25.4 for humans, and a mAP=16.0. Bus lanes, pedestrian signals, and pedestrian bridges were significantly more prevalent at pedestrian collision locations, whereas speed bumps, school zones, sidewalks, trees, potholes and streetlights were significantly more prevalent at non-pedestrian collision locations.

Conclusion
Neural networks have substantial potential to obtain timely, accurate built environment data crucial to improve road safety. Training images need to be well-annotated to ensure accurate object detection and completeness.

Learning Outcomes
1) Describe how neural networks can be used for road safety research; 2) Describe challenges of using neural networks.

Physics-informed neural networks (PINNs) are a recent trend in scientific machine learning research and modeling of differential equations. Despite progress in PINN research, large gradients and highly nonlinear patterns remain challenging to model. Thin boundary layer problems are prominent examples of large gradients that commonly arise in transport problems. In this study, boundary-layer PINN (BL-PINN) is proposed to enable a solution to thin boundary layers by considering them as a singular perturbation problem. Inspired by the classical perturbation theory and asymptotic expansions, BL-PINN is designed to replicate the procedure in singular perturbation theory. Namely, different parallel PINN networks are defined to represent different orders of approximation to the boundary layer problem in the inner and outer regions. In different benchmark problems (forward and inverse), BL-PINN shows superior performance compared to the traditional PINN approach and is able to produce accurate results, whereas the classical PINN approach could not provide meaningful solutions. BL-PINN also demonstrates significantly better results compared to other extensions of PINN such as the extended PINN (XPINN) approach. The natural incorporation of the perturbation parameter in BL-PINN provides the opportunity to evaluate parametric solutions without the need for retraining. BL-PINN demonstrates an example of how classical mathematical theory could be used to guide the design of deep neural networks for solving challenging problems.

Pathologists diagnose and grade prostate cancer by examining tissue from needle biopsies on glass slides. The cancer's severity and risk of metastasis are determined by the Gleason grade, a score based on the organization and morphology of prostate cancer glands. For diagnostic work-up, pathologists first locate glands in the whole biopsy core, and---if they detect cancer---they assign a Gleason grade. This time-consuming process is subject to errors and significant inter-observer variability, despite strict diagnostic criteria. This paper proposes an automated workflow that follows pathologists' modus operandi, isolating and classifying multi-scale patches of individual glands in whole slide images (WSI) of biopsy tissues using distinct steps: (1) two fully convolutional networks segment epithelium versus stroma and gland boundaries, respectively; (2) a classifier network separates benign from cancer glands at high magnification; and (3) an additional classifier predicts the grade of each cancer gland at low magnification. Altogether, this process provides a gland-specific approach for prostate cancer grading that we compare against other machine-learning-based grading methods.

Few-shot segmentation has received recent attention because of its promise to segment images containing novel classes based on a handful of annotated examples. Few-shot-based machine learning methods build generic and adaptable models that can quickly learn new tasks. This approach finds potential application in many scenarios that do not benefit from large repositories of labeled data, which strongly impacts the performance of the existing data-driven deep-learning algorithms. This paper presents a few-shot segmentation method for microscopy images that combines a neural-network architecture with a Gaussian-process (GP) regression. The GP regression is used in the latent space of an autoencoder-based segmentation model to learn the distribution of functions from the encoded image representations to the corresponding representation of the segmentation masks in the support set. This regression analysis serves as the prior for predicting the segmentation mask for the query image. The rich latent representation built by the GP using examples in the support set significantly impacts the performance of the segmentation model, demonstrated by extensive experimental evaluation.

Clinical investigations of anatomy’s structural changes over time could greatly benefit from population-level quantification of shape, or spatiotemporal statistic shape modeling (SSM). Such a tool enables characterizing patient organ cycles or disease progression in relation to a cohort of interest. Constructing shape models requires establishing a quantitative shape representation (e.g., corresponding landmarks). Particle-based shape modeling (PSM) is a data-driven SSM approach that captures population-level shape variations by optimizing landmark placement. However, it assumes cross-sectional study designs and hence has limited statistical power in representing shape changes over time. Existing methods for modeling spatiotemporal or longitudinal shape changes require predefined shape atlases and pre-built shape models that are typically constructed cross-sectionally. This paper proposes a data-driven approach inspired by the PSM method to learn population-level spatiotemporal shape changes directly from shape data. We introduce a novel SSM optimization scheme that produces landmarks that are in correspondence both across the population (inter-subject) and across time-series (intra-subject). We apply the proposed method to 4D cardiac data from atrial-fibrillation patients and demonstrate its efficacy in representing the dynamic change of the left atrium. Furthermore, we show that our method outperforms an image-based approach for spatiotemporal SSM with respect to a generative time-series model, the Linear Dynamical System (LDS). LDS fit using a spatiotemporal shape model optimized via our approach provides better generalization and specificity, indicating it accurately captures the underlying time-dependency.

Statistical shape modeling (SSM) is a valuable and powerful tool to generate a detailed representation of complex anatomy that enables quantitative analysis and the comparison of shapes and their variations. SSM applies mathematics, statistics, and computing to parse the shape into a quantitative representation (such as correspondence points or landmarks) that will help answer various questions about the anatomical variations across the population. Complex anatomical structures have many diverse parts with varying interactions or intricate architecture. For example, the heart is a four-chambered anatomy with several shared boundaries between chambers. Coordinated and efficient contraction of the chambers of the heart is necessary to adequately perfuse end organs throughout the body. Subtle shape changes within these shared boundaries of the heart can indicate potential pathological changes that lead to uncoordinated contraction and poor end-organ perfusion. Early detection and robust quantification could provide insight into ideal treatment techniques and intervention timing. However, existing SSM approaches fall short of explicitly modeling the statistics of shared boundaries. In this paper, we present a general and flexible data-driven approach for building statistical shape models of multi-organ anatomies with shared boundaries that captures morphological and alignment changes of individual anatomies and their shared boundary surfaces throughout the population. We demonstrate the effectiveness of the proposed methods using a biventricular heart dataset by developing shape models that consistently parameterize the cardiac biventricular structure and the interventricular septum (shared boundary surface) across the population data.

We present schemes for simulating Brownian bridges on complete and connected Lie groups and homogeneous spaces. We use this to construct an estimation scheme for recovering an unknown left- or right-invariant Riemannian metric on the Lie group from samples. We subsequently show how pushing forward the distributions generated by Brownian motions on the group results in distributions on homogeneous spaces that exhibit a non-trivial covariance structure. The pushforward measure gives rise to new non-parametric families of distributions on commonly occurring spaces such as spheres and symmetric positive tensors. We extend the estimation scheme to fit these distributions to homogeneous space-valued data. We demonstrate both the simulation schemes and estimation procedures on Lie groups and homogenous spaces, including SPD(3)=GL+(3)/SO(3) and S2=SO(3)/SO(2).