Assad A. Oberai

Abstract Liquid biopsy offers a minimally invasive means to interrogate tumor biology; however, the extreme rarity and phenotypic diversity of circulating tumor cells (CTCs) and related cellular events remain major obstacles to sensitive detection and meaningful analysis. Conventional workflows frequently rely on biophysical enrichment or predefined biomarker panels, both of which can bias cell recovery and constrain discovery. There is therefore a critical need for scalable computational approaches capable of analyzing millions of single-cell observations directly and extracting biological structure without dependence on prior labels. We developed deep learning–based pipelines to analyze nucleated cells isolated from peripheral blood. For each patient, approximately five million cells are obtained via buffy coat preparation, stained with a five-marker fluorescence panel, and imaged by whole-slide microscopy without any enrichment steps. The first pipeline is an unsupervised rare-event detector built on a denoising autoencoder. Applied to samples from 11 breast cancer patients, the method recovered 91 additional events—including CTCs, endothelial cells, cancer-associated fibroblasts (CAFs), and extracellular vesicles—representing a greater than 50% increase with minimal manual tuning. This form of label-free outlier detection is broadly generalizable to high-content imaging studies in which unbiased identification of infrequent or unexpected populations is essential. The second pipeline uses representation learning to derive stable single-cell embeddings. These embeddings support phenotype classification with 92.64% accuracy and also enable unsupervised clustering that reflects intrinsic variation in morphology and marker expression. Notably, the learned features are robust to imaging artifacts, ensuring consistent phenotyping across heterogeneous datasets. Collectively, these deep learning frameworks establish an integrated strategy for enrichment-free rare-event detection, clustering, and cell-type characterization in liquid biopsy, providing a scalable foundation for biomarker discovery Citation Format: Dean Tessone, Amin Naghdloo, Javier Murgoitio-Esandi, Jeremy Mason, Assad Oberai, James B. Hicks, Peter Kuhn. Scalable, unsupervised deep learning frameworks for rare event detection and single cell phenotyping in enrichment free liquid biopsies [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1434.

This study presents a conditional flow matching framework for solving physics-constrained Bayesian inverse problems. In this setting, samples from the joint distribution of inferred variables and measurements are assumed available, while explicit evaluation of the prior and likelihood densities is not required. We derive a simple and self-contained formulation of both the unconditional and conditional flow matching algorithms, tailored specifically to inverse problems. In the conditional setting, a neural network is trained to learn the velocity field of a probability flow ordinary differential equation that transports samples from a chosen source distribution directly to the posterior distribution conditioned on observed measurements. This black-box formulation accommodates nonlinear, high-dimensional, and potentially non-differentiable forward models without restrictive assumptions on the noise model. We further analyze the behavior of the learned velocity field in the regime of finite training data. Under mild architectural assumptions, we show that overtraining can induce degenerate behavior in the generated conditional distributions, including variance collapse and a phenomenon termed selective memorization, wherein generated samples concentrate around training data points associated with similar observations. A simplified theoretical analysis explains this behavior, and numerical experiments confirm it in practice. We demonstrate that standard early-stopping criteria based on monitoring test loss effectively mitigate such degeneracy. The proposed method is evaluated on several physics-based inverse problems. We investigate the impact of different choices of source distributions, including Gaussian and data-informed priors. Across these examples, conditional flow matching accurately captures complex, multimodal posterior distributions while maintaining computational efficiency.

This study aimed to develop and evaluate a machine learning pipeline using multiphase contrast-enhanced CT images and clinical data to classify renal tumors as benign, malignant-indolent, or malignant-aggressive, while assessing the contribution of each data source to the classification. In this retrospective study, 448 patients (mean age: 60.7 ± 12.6 years, 306 male, 142 female) who underwent nephrectomy and preoperative CECT between June 2008 and July 2018 were included. Tumors were histologically categorized as benign-indolent, malignant-indolent, or malignant-aggressive. Self-supervised feature extraction converted 4-phase CECT images into 512 real-valued features, combined with clinical data and tumor size for classification. Two machine learning classifiers, random forest (RF) and multi-layer perceptron (MLP), were used to predict tumor type. Nested five-fold cross-validation was employed for hyperparameter tuning and model evaluation, and performance was assessed using area under the curve (AUC) analysis. The best-performing models achieved an AUC of 0.90 (95% CI: 0.88-0.93) for classifying indolent versus aggressive tumors and 0.76 (95% CI: 0.71-0.81) for malignant versus benign tumors. Models incorporating tumor size significantly improved classification accuracy. RF classifiers excelled in distinguishing indolent from aggressive tumors, while MLP classifiers performed better for malignant versus benign classification. The machine learning pipeline demonstrated high accuracy in differentiating aggressive from indolent renal tumors, offering valuable prognostic insights for personalized treatment. Tumor size was a critical factor, complementing CECT images and clinical data. These findings highlight the potential of ML techniques in enhancing renal tumor risk stratification.

Abstract Brain magnetic resonance imaging (MRI) is pivotal in diagnosing and monitoring neurological disorders. However, despite their extensive applications, MR images have certain shortcomings. In particular, factors other than the anatomy of brain tissues influence the intensity distribution of voxels in MR images. These factors include hardware, software, magnetic field strength, and acquisition protocol. This inconsistency poses challenges in multi-site neuroimaging studies, where images are obtained from various devices with minimal control over acquisition parameters. Image harmonization algorithms aim to eliminate non-biological characteristics in MR images through various approaches, including converting images from multiple sites into a format resembling that of a designated target site. Among image harmonization methods, those relying on deep learning algorithms have gained significant attention recently. Nevertheless, certain aspects of deep learning-based image harmonization remain unexplored, notably the integration of probabilistic deep generative models to transform the distribution of MR images to a desired distribution. Inspired by this, we introduced a feature preserving conditional generative adversarial network (FP-cGAN) that converts images from multiple origins into the format of a target site while preserving anatomical features by imposing a novel regularizing constraint. We conduct our experiments on MR images from the SRPBS dataset, which comprises unpaired images in addition to paired (traveling subjects) images from multiple sites. We utilize the unpaired data for training our models and the paired data for evaluation. Furthermore, we compare our results with cycleGAN and histogram matching, two widely used image harmonization methods. Our experiments reveal that our approach surpasses the other techniques. Article Highlights We introduce an FP-cGAN model to harmonize multi-site MRI scans into a target site. A novel anatomical feature invariant constraint maintains tissue integrity. The models are trained in an unpaired setting without requiring traveling subjects. We evaluate harmonization performance using paired traveling subjects.

In metastatic prostate adenocarcinoma, accurate identification and quantification of metastatic lesions from CT images are essential for monitoring disease progression and evaluating therapeutic response. This study conducts a comparative evaluation of supervised learning and foundation model approaches for lesion segmentation, focusing on the zero-shot performance in a minimal computation setting. This comparison reflects real-world scenarios, where supervised learning is feasible when labeled data is available, whereas foundation models provide a flexible alternative in the absence of labeled. Furthermore, we investigate strategies to integrate supervised learning with foundation model methods. We first assess the performance of nnUNetv2, a state-of-the-art supervised segmentation framework, which achieved a Dice Similarity Coefficient (DSC) of 0.73 (95% CI: 0.68–0.78). This serves as the baseline for comparison against three segmentation foundation models SaMMed2D, SAM, and SAM2, when prompted using domain expert annotations (DEP) via bounding box. In the integrated approach, we examine nnUNetv2-generated prompts (nP) and domain expert prompts (DEP), along with their combinatorial operations with the baseline. While nP-based prompting combinations, such as SAM (DSC: 0.74 (95% CI: 0.66–0.82; p=0.06)) and SAM2 (DSC: 0.74 (95% CI: 0.64–0.84; p=0.07)), showed improved mean performance, none achieved statistically significant gains over the nnUNetv2 baseline. These findings highlight the robustness of supervised approaches like nnUNetv2 and suggest that while foundation models offer flexibility, their integration through prompting may require more adaptive strategies to yield tangible improvements in medical image segmentation.

Tumor-associated cells in liquid biopsy are promising biomarkers for cancer detection, diagnosis, prognosis, and monitoring. Yet, their rarity, heterogeneity, and plasticity pose challenges for accurate identification and characterization. Enrichment-free whole slide imaging of all circulating cells offers a comprehensive, unbiased approach to capture this phenotypic diversity. However, current analysis methods often rely on engineered features and manual expert review, making them prone to technical variability and subjective bias. To address this, we present a deep contrastive learning framework for feature extraction from whole slide immunofluorescence microscopy images, enabling robust identification and stratification of single circulating cells. Our learned features achieve 92.64% accuracy in classifying diverse cell phenotypes and improve downstream tasks such as outlier detection and clustering. Additionally, our model enables automated identification and enumeration of rare phenotypes, reaching an average F1-score of 0.93 on contrived samples mimicking circulating tumor and endothelial cells, and 0.858 across circulating tumor cell phenotypes in clinical samples. This workflow provides a scalable, reproducible solution for analyzing tumor-associated cellular biomarkers, with strong potential to enhance clinical prognosis and guide personalized treatment strategies.

Kidney cancer is among the top 10 most common malignancies in adults, and is commonly evaluated with four-phase computed tomography (CT) imaging. However, the presence of missing or corrupted images remains a significant problem in medical imaging that impairs the detection, diagnosis, and treatment planning of kidney cancer. Deep learning approaches through conditional generative adversarial networks (cGANs) have recently shown technical promise in the task of imputing missing imaging data from these four-phase studies. In this study, we explored the clinical utility of these imputed images. We utilized a cGAN trained on 333 patients, with the task of the cGAN being to impute the image of any phase given the other three phases. We tested the clinical utility on the imputed images of the 37 patients in the test set by manually extracting 21 clinically relevant imaging features and comparing them to their ground truth counterpart. All 13 categorical clinical features had greater than 85% agreement rate between true images and their imputed counterparts. This high accuracy is maintained when stratifying across imaging phases. Imputed images also show good agreement with true images in select radiomic features including mean intensity and enhancement. Imputed images possess the features characteristic of benign or malignant diagnosis at an equivalent rate to true images. In conclusion, imputed images from cGANs have large potential for clinical use due to their ability to retain clinically relevant qualitative and quantitative features.

The use of liquid biopsies in the detection, diagnosis and treatment monitoring of different types of cancers and other diseases often requires identifying and enumerating instances of analytes that are rare. Most current techniques that aim to computationally isolate these rare instances or events first learn the signature of the event, and then scan the appropriate biological assay for this signature. While such techniques have proven to be very useful, they are limited because they must first establish what signature to look for, and only then identify events that are consistent with this signature. In contrast to this, in this study, we present an automated approach that does not require the knowledge of the signature of the rare event. It works by breaking the assay into a sequence of components, learning the probability distribution of these components, and then isolating those that are rare. This is done with the help of deep generative algorithms in an unsupervised manner, meaning without a-priori knowledge of the rare event associated with an analyte. In this study, this approach is applied to immunofluorescence microscopy images of peripheral blood, where it is shown that it successfully isolates biologically relevant events in blood from normal donors spiked with cancer-related cells and in blood from patients with late-stage breast cancer.

We propose a data-driven method to learn the time-dependent probability density of a multivariate stochastic process from sample paths, assuming that the initial probability density is known and can be evaluated. Our method uses a novel time-dependent binary classifier trained using a contrastive estimation-based objective that trains the classifier to discriminate between realizations of the stochastic process at two nearby time instants. Significantly, the proposed method explicitly models the time-dependent probability distribution, which means that it is possible to obtain the value of the probability density within the time horizon of interest. Additionally, the input before the final activation in the time-dependent classifier is a second-order approximation to the partial derivative, with respect to time, of the logarithm of the density. We apply the proposed approach to approximate the time-dependent probability density functions for systems driven by stochastic excitations. We also use the proposed approach to synthesize new samples of a random vector from a given set of its realizations. In such applications, we generate sample paths necessary for training using stochastic interpolants. Subsequently, new samples are generated using gradient-based Markov chain Monte Carlo methods because automatic differentiation can efficiently provide the necessary gradient. Further, we demonstrate the utility of an explicit approximation to the time-dependent probability density function through applications in unsupervised outlier detection. Through several numerical experiments, we show that the proposed method accurately reconstructs complex time-dependent, multi-modal, and near-degenerate densities, scales effectively to moderately high-dimensional problems, and reliably detects rare events among real-world data.

Increasing wildfire occurrence has spurred growing interest in wildfire spread prediction. However, even the most complex wildfire models diverge from observed progression during multi-day simulations, motivating need for data assimilation. A useful approach to assimilating measurement data into complex coupled atmosphere-wildfire models is to estimate wildfire progression from measurements and use this progression to develop a matching atmospheric state. In this study, an approach is developed for estimating fire progression from VIIRS active fire measurements, GOES-derived ignition times, and terrain height data. A conditional Generative Adversarial Network is trained with simulations of historic wildfires from the atmosphere-wildfire model WRF-SFIRE, thus allowing incorporation of WRF-SFIRE physics into estimates. Fire progression is succinctly represented by fire arrival time, and measurements for training are obtained by applying an approximate observation operator to WRF-SFIRE solutions, eliminating need for satellite data during training. The model is trained on tuples of fire arrival times, measurements, and terrain, and once trained leverages measurements of real fires and corresponding terrain data to generate samples of fire arrival times. The approach is validated on five Pacific US wildfires, with results compared against high-resolution perimeters measured via aircraft, finding an average Sorensen-Dice coefficient of 0.81. The influence of terrain height on the arrival time inference is also evaluated and it is observed that terrain has minimal influence when the inference is conditioned on satellite measurements.

Tumor-associated cells derived from a liquid biopsy are promising biomarkers for cancer detection, diagnosis, prognosis, and monitoring. However, their rarity, heterogeneity and plasticity make precise identification and biological characterization challenging for clinical utility. Enrichment-free approaches using whole slide imaging of all circulating cells offer a comprehensive and unbiased strategy for capturing the full spectrum of tumor-associated cell phenotypes. However, current analysis methods often depend on engineered features and manual expert review, making them sensitive to technical variations and subjective biases. These limitations highlight the need for a better feature representation to improve performance and reproducibility of applications in large-scale patient cohort analyses. In this study, we present a deep contrastive learning framework for learning features of all circulating cells, enabling robust identification and stratification of single cells in whole slide immunofluorescence microscopy images. We demonstrate performance of learned features in classification of diverse cell phenotypes in the liquid biopsy, achieving an accuracy of 92.64%. We further demonstrate that learned features improve performance in downstream applications such as outlier detection and clustering. Lastly, our feature representation enables automated identification and enumeration of distinct rare cell phenotypes, achieving average F1-score of 0.93 across cell lines mimicking circulating tumor cells and endothelial cells in contrived samples and average F1-score of 0.858 across CTC phenotypes in clinical samples. This workflow has significant implications for scalable analysis of tumor-associated cellular biomarkers in clinical prognosis and personalized treatment strategies.

: Nuclear energy is among the cleanest and most efficient energy sources currently available.

1 Abstract Purpose This study aimed to develop and evaluate a machine learning pipeline using multiphase contrast-enhanced CT images and clinical data to classify renal tumors as benign, malignant-indolent, or malignant-aggressive, while assessing the contribution of each data source to the classification. Methods In this retrospective study, 448 patients (mean age: 60.7±12.6 years, 306 male, 142 female) who underwent nephrectomy and preoperative CECT between June 2008 and July 2018 were included. Tumors were histologically categorized as benign-indolent, malignant-indolent, or malignant-aggressive. Self-supervised feature extraction converted 4-phase CECT images into 512 real-valued features, combined with clinical data and tumor size for classification. Two machine learning classifiers, random forest (RF) and multi-layer perceptron (MLP), were used to predict tumor type. Nested five-fold cross-validation was employed for hyperparameter tuning and model evaluation, and performance was assessed using area under the curve (AUC) analysis. Results The best-performing models achieved an AUC of 0.90 (95% CI: 0.88–0.93) for classifying indolent versus aggressive tumors and 0.76 (95% CI: 0.71–0.81) for benign versus malignant tumors. Models incorporating tumor size significantly improved classification accuracy. RF classifiers excelled in distinguishing indolent from aggressive tumors, while MLP classifiers performed better for benign versus malignant classification. Conclusion The machine learning pipeline demonstrated high accuracy in differentiating aggressive from indolent renal tumors, offering valuable prognostic insights for personalized treatment. Tumor size was a critical factor, complementing CECT images and clinical data. These findings highlight the potential of ML techniques in enhancing renal tumor risk stratification.

The use of liquid biopsies in the detection, diagnosis and treatment monitoring of different types of cancers and other diseases often requires identifying and enumerating instances of analytes that are rare. Most current techniques that aim to computationally isolate these rare instances or events first learn the signature of the event, and then scan the appropriate biological assay for this signature. While such techniques have proven to be very useful, they are limited because they must first establish what signature to look for, and only then identify events that are consistent with this signature. In contrast to this, in this study, we present an automated approach that does not require the knowledge of the signature of the rare event. It works by breaking the assay into a sequence of components, learning the probability distribution of these components, and then isolating those that are rare. This is done with the help of deep generative algorithms in an unsupervised manner, meaning without a-priori knowledge of the rare event associated with an analyte. In this study, this approach is applied to immunofluorescence microscopy images of peripheral blood, where it is shown that it successfully isolates biologically relevant events in blood from normal donors spiked with cancer-related cells and in blood from patients with late-stage breast cancer.

Diffusion models have emerged as a powerful framework for generative modeling. At the heart of the methodology is score matching: learning gradients of families of log-densities for noisy versions of the data distribution at different scales. When the loss function adopted in score matching is evaluated using empirical data, rather than the population loss, the minimizer corresponds to the score of a time-dependent Gaussian mixture. However, use of this analytically tractable minimizer leads to data memorization: in both unconditioned and conditioned settings, the generative model returns the training samples. This paper contains an analysis of the dynamical mechanism underlying memorization. The analysis highlights the need for regularization to avoid reproducing the analytically tractable minimizer; and, in so doing, lays the foundations for a principled understanding of how to regularize. Numerical experiments investigate the properties of: (i) Tikhonov regularization; (ii) regularization designed to promote asymptotic consistency; and (iii) regularizations induced by under-parameterization of a neural network or by early stopping when training a neural network. These experiments are evaluated in the context of memorization, and directions for future development of regularization are highlighted.

Introduction: Prostate adenocarcinoma frequently metastasizes to bone and is detected via computed tomography (CT) scans. Accurate detection and segmentation of these lesions are critical for diagnosis, prognosis, and monitoring. This study aims to automate lesion detection and segmentation using deep learning models. Methods and Materials: We evaluated deep learning models for lesion detection (EfficientNet, ResNet34, DenseNet) and segmentation (nnUNetv2, UNet, ResUNet, ResAttUNet). Performance metrics included F1 score, precision, recall, Area Under the Curve (AUC), and Dice Similarity Coefficient (DSC). Pairwise t-tests compared segmentation accuracy. Radiomic analyses compared lesions segmented by deep learning to manual segmentations. Results: EfficientNet achieved the highest detection performance, with an F1 score of 0.82, precision of 0.88, recall of 0.79, and AUC of 0.71. Among segmentation models, nnUNetv2 performed best, achieving a DSC of 0.74, precision of 0.73, and recall of 0.83. Pairwise t-tests showed that nnUNetv2 outperformed other models in segmentation accuracy (p < 0.01). Clinically, nnUNetv2 also demonstrated superior specificity for lesion detection (0.90) compared to the other models. All models performed similarly in distinguishing diffuse and focal lesions, predicting weight-bearing lesions, and identifying lesion locations, although nnUNetv2 had higher specificity. Sensitivity was highest for rib lesions and lowest for spine lesions across all models. Conclusions: EfficientNet and nnUNetv2 were the top-performing models for detection and segmentation, respectively. Radiomic features derived from deep learning-based segmentations were comparable to manual segmentations, supporting clinical applicability. Further analysis of lesion detection and spatial distribution underscores the models' potential for improving diagnostic workflows and patient outcomes.

'Correction to: Deep Learning and Computational Physics' published in 'Deep Learning and Computational Physics'

We present a novel probabilistic approach for generating multi-fidelity data while accounting for errors inherent in both low- and high-fidelity data. In this approach a graph Laplacian constructed from the low-fidelity data is used to define a multivariate Gaussian prior density for the coordinates of the true data points. In addition, few high-fidelity data points are used to construct a conjugate likelihood term. Thereafter, Bayes rule is applied to derive an explicit expression for the posterior density which is also multivariate Gaussian. The maximum \textit{a posteriori} (MAP) estimate of this density is selected to be the optimal multi-fidelity estimate. It is shown that the MAP estimate and the covariance of the posterior density can be determined through the solution of linear systems of equations. Thereafter, two methods, one based on spectral truncation and another based on a low-rank approximation, are developed to solve these equations efficiently. The multi-fidelity approach is tested on a variety of problems in solid and fluid mechanics with data that represents vectors of quantities of interest and discretized spatial fields in one and two dimensions. The results demonstrate that by utilizing a small fraction of high-fidelity data, the multi-fidelity approach can significantly improve the accuracy of a large collection of low-fidelity data points.

Brain MR image transformation, which is the process of transforming one type of MR image into another, is a critical neuroimaging task that is needed when the target image type is missing or corrupted. Accordingly, several methods have been developed to tackle this problem, with a recent focus on deep learning-based models. In this paper, we investigate the performance of the conditional version of three such probabilistic generative models, including conditional Generative Adversarial Networks (cGAN), Noise Conditioned Score Networks (NCSN), and De-noising Diffusion Probabilistic Models (DDPM). We also compare their performance against a more traditional deterministic U-Net based model. We train and test these models using MR images from publicly available datasets IXI and OASIS. For images from the IXI dataset, we conduct experiments on combinations of transformations between T1-weighted (T1), T2-weighted (T2), and proton density (PD) images, whereas for the OASIS dataset, we consider combinations of T1, T2, and Fluid Attenuated Inversion Recovery (FLAIR) images. In evaluating these models, we measure the similarity between the transformed image and the target image using metrics like PSNR and SSIM. In addition, for the three probabilistic generative models, we evaluate the utility of generating an ensemble of predictions by computing a metric that measures the variance in their predictions and demonstrate that it can be used to identify out-of-distribution (OOD) input images. We conclude that the NCSN model yields the most accurate transformations, while the DDPM model yields variance results that most clearly detect OOD inputs. We also note that while the results for the two diffusion models (NCSN and DDPM) are more accurate than those for the cGAN, the latter was significantly more efficient in generating multiple samples. Overall, our work demonstrates the utility of probabilistic conditional generative models for MR image transformation and highlights the role of generating an ensemble of outputs in identifying OOD input images.

Currently, spent nuclear fuel (SNF) from commercial nuclear power plants is stored in stainless-steel canisters for interim dry storage. To provide an inert environment, these canisters are backfilled with helium after vacuum drying. However, the helium environment may be contaminated during extended storage because of the material degradation. For example, the heavier fission gas xenon may be released from the fuel rods into the canister cavity should the fuel cladding be breached. Other gases such as air and water vapor may also be present as a result of leakage caused by chloride-induced stress corrosion cracking on the canister walls or by insufficient vacuum drying. Therefore, monitoring the gas composition can provide critical information about the health of SNF canisters. In this study, noninvasive testing was conducted on a 2/3-scaled SNF canister mock-up using acoustic sensing. Ultrasonic transducers were placed on the exterior surface of the canister to probe the gas composition. A dataset was collected by sealing the canister mock-up and introducing up to 1.53% argon or 1.29% air into the helium background gas. Three methods were used to detect changes in the gas composition: the time-of-flight (TOF) method, the differential method, and the autoencoder method. Results showed that the TOF method had sufficient resolution to detect abnormal gas concentrations of less than 1.0%. The differential method demonstrated a periodic in-phase and out-of-phase behavior between the benchmark (i.e., pure helium) and abnormal (i.e., with argon or air) state signals. The variational autoencoder (VAE) and the Wasserstein autoencoder (WAE) were trained on the benchmark data and were applied directly to the abnormal state data. It was found that both the unsupervised VAE and the WAE were able to distinguish the benchmark and abnormal states of the canister mock-up based on the reconstruction error.

Abstract Purpose Prostate adenocarcinoma frequently metastasizes to bone and is detected via computed tomography (CT) scans. Accurate detection and segmentation of these lesions are critical for diagnosis, prognosis, and monitoring. This study aims to automate lesion detection and segmentation using deep learning models. Methods and Materials We evaluated several deep learning models for lesion detection (EfficientNet, ResNet34, DenseNet) and segmentation (nnUNetv2, UNet, ResUNet, ResAttUNet). Performance metrics included F1 score, precision, recall, Area Under the Curve (AUC), and Dice Similarity Coefficient (DSC). Pairwise t-tests compared segmentation accuracy. Additionally, we conducted radiomic analyses to compare lesions segmented by deep learning to manual segmentations Results EfficientNet achieved the highest detection performance, with an F1 score of 0.82, precision of 0.88, recall of 0.79, and AUC of 0.71. Among segmentation models, nnUNetv2 performed best, achieving a DSC of 0.74, with precision and recall values of 0.73 and 0.83, respectively. Pairwise t-tests showed that nnUNetv2 outperformed ResAttUNet, ResUNet, and UNet in segmentation accuracy (p &lt; 0.01). Clinically, nnUNetv2 also demonstrated superior specificity for lesion detection (0.9) compared to the other models. All models performed similarly in distinguishing diffuse and focal lesions, predicting weight-bearing lesions, and identifying lesion locations, although nnUNetv2 had higher specificity for these tasks. Sensitivity was highest for rib lesions and lowest for spine lesions across all models. Conclusions EfficientNet and nnUNetv2 were the top-performing models for detection and segmentation, respectively. The radiomic features derived from deep learning-based segmentations were comparable to those from manual segmentations, supporting the clinical applicability of these methods. Further analysis of lesion detection and spatial distribution, as well as lesion quality differentiation, underscores the models’ potential for improving diagnostic workflows and patient outcomes in clinical settings.

We propose a framework to perform Bayesian inference using conditional score-based diffusion models to solve a class of inverse problems in mechanics involving the inference of a specimen's spatially varying material properties from noisy measurements of its mechanical response to loading. Conditional score-based diffusion models are generative models that learn to approximate the score function of a conditional distribution using samples from the joint distribution. More specifically, the score functions corresponding to multiple realizations of the measurement are approximated using a single neural network, the so-called score network, which is subsequently used to sample the posterior distribution using an appropriate Markov chain Monte Carlo scheme based on Langevin dynamics. Training the score network only requires simulating the forward model. Hence, the proposed approach can accommodate black-box forward models and complex measurement noise. Moreover, once the score network has been trained, it can be re-used to solve the inverse problem for different realizations of the measurements. We demonstrate the efficacy of the proposed approach on a suite of high-dimensional inverse problems in mechanics that involve inferring heterogeneous material properties from noisy measurements. Some examples we consider involve synthetic data, while others include data collected from actual elastography experiments. Further, our applications demonstrate that the proposed approach can handle different measurement modalities, complex patterns in the inferred quantities, non-Gaussian and non-additive noise models, and nonlinear black-box forward models. The results show that the proposed framework can solve large-scale physics-based inverse problems efficiently.

The safe storage of spent nuclear fuel (SNF) in dry cask storage systems (DCSSs) is critical to the nuclear fuel cycle and the future of nuclear energy. A critical component of DCSSs is the SNF canister. The canister is a sealed stainless-steel structure, which is first vacuum dried and then backfilled with helium. The structural deterioration within a canister can be monitored through its internal gas properties. This monitoring serves as the driving force behind the non-invasive ultrasonic sensing approach in this paper. A major challenge in collecting gas-borne signals using ultrasonic sensing is the impedance mismatch between the stainless-steel canister and the helium gas inside. Only a small fraction of the ultrasonic signal makes its way from the transmitter to the receiver through the gas medium. In this paper, experimental studies on a partial full-scale canister mock-up were carried out to capture the gas-borne signals. Damping materials were applied on the outside, and blocking and unblocking tests were conducted to identify the gas-borne signal. The research results showed that the excitation frequency played an important role in maximizing the gas-borne signals. The gas-borne signal was successfully detected at around the theoretical time-of-flight (TOF) at 225 kHz. A high signal-to-noise ratio (SNR) was achieved in the measurements. Next, acoustic impedance matching (AIM) layers were added, and it was found that the gas signal energy was improved by 160.4% compared with that of no AIM layers. Subsequently, the relative humidity (RH) level and temperature of the gas were varied to simulate abnormal internal conditions of the canister. The non-invasive testing system demonstrated reliability and sensitivity in detecting gas temperature and RH variations. Theoretical calculations demonstrated the potential for detecting low-level xenon and air within an actual SNF canister filled with helium. Last, an active noise cancellation (ANC) method, previously developed by the authors, was verified on the canister mock-up for the first time. The results showed that the SNR of the gas signal was improved by 213.6% compared with that of no ANC.

We present a novel probabilistic approach for generating multi-fidelity data while accounting for errors inherent in both low- and high-fidelity data. In this approach a graph Laplacian constructed from the low-fidelity data is used to define a multivariate Gaussian prior density for the coordinates of the true data points. In addition, few high-fidelity data points are used to construct a conjugate likelihood term. Thereafter, Bayes rule is applied to derive an explicit expression for the posterior density which is also multivariate Gaussian. The maximum \textit{a posteriori} (MAP) estimate of this density is selected to be the optimal multi-fidelity estimate. It is shown that the MAP estimate and the covariance of the posterior density can be determined through the solution of linear systems of equations. Thereafter, two methods, one based on spectral truncation and another based on a low-rank approximation, are developed to solve these equations efficiently. The multi-fidelity approach is tested on a variety of problems in solid and fluid mechanics with data that represents vectors of quantities of interest and discretized spatial fields in one and two dimensions. The results demonstrate that by utilizing a small fraction of high-fidelity data, the multi-fidelity approach can significantly improve the accuracy of a large collection of low-fidelity data points.

Abstract Monitoring impurity gases in spent nuclear fuel (SNF) canisters is a novel structural health monitoring approach for SNF in dry storage. The SNF canisters are sealed containers that do not facilitate visual access to the inside. Acoustic sensing can be deployed by taking advantage of the pathways unobstructed by internal hardware. Although the ultrasonic time-of-flight measurement can provide valuable information, it is limited in its ability to discern the concentration of only one impurity gas. As such, deep learning algorithms, particularly convolutional neural networks (CNNs), offer a promising solution. In this study, CNN-based probabilistic deep learning models were implemented to detect and quantify multiple impurity gases in helium. An experimental platform was established to simulate canister conditions, and ultrasonic test data were collected. The presence of argon and air in helium at concentrations ranging from 0% to 1.2% at increments of 0.05% was considered. The multi-layer perceptron, decision tree, and logistic regression classifiers achieved high accuracies when distinguishing pure helium from helium with impurities. CNN with dropout layers and CNN using maximum likelihood estimation showed a similar performance, indicating their ability to capture uncertainties. The ensemble CNN model exhibited improved predictions and the ability to balance individual gas concentration by integrating 1D- and 2D-CNN models. These findings contribute probabilistic deep learning solutions for impurity gas detection and analysis within SNF canisters, thus ensuring safe storage and management of SNFs.

BACKGROUND AND OBJECTIVES: Techniques for imaging the mechanical properties of cells are needed to study how cell mechanics influence cell function and disease progression. Mechano-microscopy (a high-resolution variant of compression optical coherence elastography) generates elasticity images of a sample undergoing compression from the phase difference between optical coherence microscopy (OCM) B-scans. However, the existing mechano-microscopy signal processing chain (referred to as the algebraic method) assumes the sample stress is uniaxial and axially uniform, such that violation of these assumptions reduces the accuracy and precision of elasticity images. Furthermore, it does not account for prior information regarding the sample geometry or mechanical property distribution. In this study, we investigate the feasibility of training a conditional generative adversarial network (cGAN) to generate elasticity images from phase difference images of samples containing a cell spheroid embedded in a hydrogel. METHODS: To construct the cGAN training and simulated test sets, we generated 30,000 artificial elasticity images using a parametric model and computed the corresponding phase difference images using finite element analysis to simulate compression applied to the artificial samples. We also imaged real MCF7 breast tumor spheroids embedded in hydrogel using mechano-microscopy to construct the experimental test set and evaluated the cGAN using the algebraic elasticity images and co-registered OCM and confocal fluorescence microscopy (CFM) images. RESULTS: Comparison with the simulated test set ground truth elasticity images shows the cGAN produces a lower root mean square error (median: 3.47 kPa, 95 % confidence interval (CI) [3.41, 3.52]) than the algebraic method (median: 4.91 kPa, 95 % CI [4.85, 4.97]). For the experimental test set, the cGAN elasticity images contain features resembling stiff nuclei at locations corresponding to nuclei seen in the algebraic elasticity, OCM, and CFM images. Furthermore, the cGAN elasticity images are higher resolution and more robust to noise than the algebraic elasticity images. CONCLUSIONS: The cGAN elasticity images exhibit better accuracy, spatial resolution, sensitivity, and robustness to noise than the algebraic elasticity images for both simulated and real experimental data.

We propose a framework to perform Bayesian inference using conditional score-based diffusion models to solve a class of inverse problems in mechanics involving the inference of a specimen's spatially varying material properties from noisy measurements of its mechanical response to loading. Conditional score-based diffusion models are generative models that learn to approximate the score function of a conditional distribution using samples from the joint distribution. More specifically, the score functions corresponding to multiple realizations of the measurement are approximated using a single neural network, the so-called score network, which is subsequently used to sample the posterior distribution using an appropriate Markov chain Monte Carlo scheme based on Langevin dynamics. Training the score network only requires simulating the forward model. Hence, the proposed approach can accommodate black-box forward models and complex measurement noise. Moreover, once the score network has been trained, it can be re-used to solve the inverse problem for different realizations of the measurements. We demonstrate the efficacy of the proposed approach on a suite of high-dimensional inverse problems in mechanics that involve inferring heterogeneous material properties from noisy measurements. Some examples we consider involve synthetic data, while others include data collected from actual elastography experiments. Further, our applications demonstrate that the proposed approach can handle different measurement modalities, complex patterns in the inferred quantities, non-Gaussian and non-additive noise models, and nonlinear black-box forward models. The results show that the proposed framework can solve large-scale physics-based inverse problems efficiently.

Generative modeling has drawn much attention in creative and scientific data generation tasks. Score-based Diffusion Models, a type of generative model that iteratively learns to denoise data, have shown state-of-the-art results on tasks such as image generation, multivariate time series forecasting, and robotic trajectory planning. Using score-based diffusion models, this work implements a novel generative framework to generate ballistic transfers from Earth to Mars. We further analyze the model's ability to learn the characteristics of the original dataset and its ability to produce transfers that follow the underlying dynamics. Ablation studies were conducted to determine how model performance varies with model size and trajectory temporal resolution. In addition, a performance benchmark is designed to assess the generative model's usefulness for trajectory design, conduct model performance comparisons, and lay the groundwork for evaluating different generative models for trajectory design beyond diffusion. The results of this analysis showcase several useful properties of diffusion models that, when taken together, can enable a future system for generative trajectory design powered by diffusion models.

Safe storage of spent nuclear fuel (SNF) is critical to the nuclear fuel cycle and the future of nuclear energy. In the United States, SNF is primarily stored via two methods regulated by the U.S. Nuclear Regulatory Commission (U.S. NRC): wet storage in SNF pools, and dry storage in dry cask storage systems (DCSSs). After about five years of cooling in spent fuel pools, the fuel assemblies are transferred into DCSSs, and the systems are filled with helium and sealed by welding. Deterioration of conditions inside of a DCSS will be reflected by changes in the internal gas properties which motivates the development of acoustic techniques to monitor internal gas properties, over extended storage periods, using sensors mounted on the exterior of the storage packages. However, a major challenge in collecting acoustic signals is the impedance mismatch between the steel canister shell and the gas. Only a small fraction of the ultrasonic signal can be transmitted through the gas medium. In this paper, experimental studies on a full-scale canister mock-up were conducted to capture the gas-borne signals. Damping materials were pasted on the outside and blocking and unblocking tests were conducted to identify the gas-borne signal. The results showed that the excitation frequency plays an important role in maximizing the gas-borne signal. The gas-borne signal was successfully detected at around the theoretical time-of-flight (TOF). A high signal-to-noise ratio (SNR) was achieved in the measurements. Next, the acoustic impedance matching (AIM) layers were introduced, and the gas signal was drastically improved compared with no AIM layers.

Abstract Increases in wildfire activity and the resulting impacts have prompted the development of high-resolution wildfire behavior models for forecasting fire spread. Recent progress in using satellites to detect fire locations further provides the opportunity to use measurements toward improving fire spread forecasts from numerical models through data assimilation. This work develops a physics-informed approach for inferring the history of a wildfire from satellite measurements, providing the necessary information to initialize coupled atmosphere–wildfire models from a measured wildfire state. The fire arrival time, which is the time the fire reaches a given spatial location, acts as a succinct representation of the history of a wildfire. In this work, a conditional Wasserstein generative adversarial network (cWGAN), trained with WRF–SFIRE simulations, is used to infer the fire arrival time from satellite active fire data. The cWGAN is used to produce samples of likely fire arrival times from the conditional distribution of arrival times given satellite active fire detections. Samples produced by the cWGAN are further used to assess the uncertainty of predictions. The cWGAN is tested on four California wildfires occurring between 2020 and 2022, and predictions for fire extent are compared against high-resolution airborne infrared measurements. Further, the predicted ignition times are compared with reported ignition times. An average Sørensen’s coefficient of 0.81 for the fire perimeters and an average ignition time difference of 32 min suggest that the method is highly accurate. Significance Statement To initialize coupled atmosphere–wildfire simulations in a physically consistent way based on satellite measurements of active fire locations, it is critical to ensure the state of the fire and atmosphere aligns at the start of the forecast. If known, the history of a wildfire may be used to develop an atmospheric state matching the wildfire state determined from satellite data in a process known as spinup. In this paper, we present a novel method for inferring the early stage history of a wildfire based on satellite active fire measurements. Here, inference of the fire history is performed in a probabilistic sense and physics is further incorporated through the use of training data derived from a coupled atmosphere–wildfire model.

Abstract Techniques for imaging the mechanical properties of cells are needed to study how cell mechanics influence cell function and disease progression. Mechano-microscopy (a high-resolution variant of compression optical coherence elastography) generates elasticity images of a sample undergoing compression from the phase difference between optical coherence microscopy (OCM) B-scans. However, the existing mechano-microscopy signal processing chain (referred to as the algebraic method) assumes the sample stress is uniaxial and axially uniform, such that violation of these assumptions reduces the accuracy and precision of elasticity images. Furthermore, it does not account for prior information regarding the sample geometry or mechanical property distribution. In this study, we investigate the feasibility of training a conditional generative adversarial network (cGAN) to generate elasticity images from phase difference images of samples containing a cell spheroid embedded in a hydrogel. To train and test the cGAN, we constructed 30,000 elasticity and phase difference image pairs, where elasticity images were generated using a parametric model to simulate artificial samples, and phase difference images were computed using finite element analysis to simulate compression applied to the artificial samples. By applying both the cGAN and algebraic methods to simulated phase difference images, our results indicate the cGAN elasticity images exhibit better spatial resolution and sensitivity. We also evaluated the cGAN on experimental phase difference images of real spheroids embedded in hydrogels and compared the cGAN elasticity with the algebraic elasticity, OCM, and confocal fluorescence microscopy, and found the cGAN elasticity is often more robust to noise, especially within stiff nuclei.

Image imputation refers to the task of generating a type of medical image given images of another type. This task becomes challenging when the difference between the available images, and the image to be imputed is large. In this manuscript, one such application is considered. It is derived from the dynamic contrast enhanced computed tomography (CECT) imaging of the kidneys: given an incomplete sequence of three CECT images, we are required to impute the missing image. This task is posed as one of probabilistic inference and a generative algorithm to generate samples of the imputed image, conditioned on the available images, is developed, trained, and tested. The output of this algorithm is the "best guess" of the imputed image, and a pixel-wise image of variance in the imputation. It is demonstrated that this best guess is more accurate than those generated by other, deterministic deep-learning based algorithms, including ones which utilize additional information and more complex loss terms. It is also shown that the pixel-wise variance image, which quantifies the confidence in the reconstruction, can be used to determine whether the result of the imputation meets a specified accuracy threshold and is therefore appropriate for a downstream task.