Prior Information Research Articles

Normalizing flows as an enhanced sampling method for atomistic supercooled liquids

Abstract Normalizing flows can transform a simple prior probability distribution into a more complex target distribution. Here, we evaluate the ability and efficiency of generative machine learning methods to sample the Boltzmann distribution of an atomistic model for glass-forming liquids. This is a notoriously difficult task, as it amounts to ergodically exploring the complex free energy landscape of a disordered and frustrated many-body system. We optimize a normalizing flow model to successfully transform high-temperature configurations of a dense liquid into low-temperature ones, near the glass transition. We perform a detailed comparative analysis with established enhanced sampling techniques developed in the physics literature to assess and rank the performance of normalizing flows against state-of-the-art algorithms. We demonstrate that machine learning methods are very promising, showing a large speedup over conventional molecular dynamics. Normalizing flows show performances comparable to parallel tempering and population annealing, while still falling far behind the swap Monte Carlo algorithm. Our study highlights the potential of generative machine learning models in scientific computing for complex systems, but also points to some of its current limitations and the need for further improvement.

Enhanced IDOL segmentation framework using personalized hyperspace learning IDOL.

Adaptive radiotherapy (ART) workflows have been increasingly adopted to achieve dose escalation and tissue sparing under shifting anatomic conditions, but the necessity of recontouring and the associated time burden hinders a real-time or online ART workflow. In response to this challenge, approaches to auto-segmentation involving deformable image registration, atlas-based segmentation, and deep learning-based segmentation (DLS) have been developed. Despite the particular promise shown by DLS methods, implementing these approaches in a clinical setting remains a challenge, namely due to the difficulty of curating a data set of sufficient size and quality so as to achieve generalizability in a trained model. To address this challenge, we have developed an intentional deep overfit learning (IDOL) framework tailored to the auto-segmentation task. However, certain limitations were identified, particularly the insufficiency of the personalized dataset to effectively overfit the model. In this study, we introduce a personalized hyperspace learning (PHL)-IDOL segmentation framework capable of generating datasets that induce the model to overfit specific patient characteristics for medical image segmentation. The PHL-IDOL model is trained in two stages. In the first, a conventional, general model is trained with a diverse set of patient data (n=100 patients) consisting of CT images and clinical contours. Following this, the general model is tuned with a data set consisting of two components: (a) selection of a subset of the patient data (m<n) using the similarity metrics (mean square error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and the universal quality image index (UQI) values); (b) adjust the CT and the clinical contours using a deformed vector generated from the reference patient and the selected patients using (a). After training, the general model, the continual model, the conventional IDOL model, and the proposed PHL-IDOL model were evaluated using the volumetric dice similarity coefficient (VDSC) and the Hausdorff distance 95% (HD95%) computed for 18 structures in 20 test patients. Implementing the PHL-IDOL framework resulted in improved segmentation performance for each patient. The Dice scores increased from 0.81 0.05 with the general model, 0.83 for the continual model, 0.83 for the conventional IDOL model to an average of 0.87 with the PHL-IDOL model. Similarly, the Hausdorff distance decreased from 3.06 with the general model, 2.84 for the continual model, 2.79 for the conventional IDOL model and 2.36 for the PHL-IDOL model. All the standard deviations were decreased by nearly half of the values comparing the general model and the PHL-IDOL model. The PHL-IDOL framework applied to the auto-segmentation task achieves improved performance compared to the general DLS approach, demonstrating the promise of leveraging patient-specific prior information in a task central to online ART workflows.

Treating severe paediatric asthma with mepolizumab or omalizumab: a protocol for the TREAT randomised non-inferiority trial

IntroductionA minority of school-aged children with asthma have persistent poor control and experience frequent asthma attacks despite maximal prescribed maintenance therapy. These children have higher morbidity and risk of death....

Convergent least-squares optimization methods for variational data assimilation

Data assimilation combines prior (or background) information with observations to estimate the initial state of a dynamical system over a given time-window. A common application is in numerical weather prediction where a previous forecast and atmospheric observations are used to obtain the initial conditions for a numerical weather forecast. In four-dimensional variational data assimilation (4D-Var), the problem is formulated as a nonlinear least-squares problem, usually solved using a variant of the classical Gauss-Newton (GN) method. However, we show that GN may not converge if poorly initialized. In particular, we show that this may occur when there is greater uncertainty in the background information compared to the observations, or when a long time-window is used in 4D-Var allowing more observations. The difficulties GN encounters may lead to inaccurate initial state conditions for subsequent forecasts. To overcome this, we apply two convergent GN variants (line search and regularization) to the long time-window 4D-Var problem and investigate the cases where they locate a more accurate estimate compared to GN within a given budget of computational time and cost. We show that these methods are able to improve the estimate of the initial state, which may lead to a more accurate forecast. Highlights Poor initialization of Gauss-Newton method may result in failure to converge. Safeguarded Gauss-Newton improves initial state estimate within limited time/cost. Results using twin experiments with long time-window and chaotic Lorenz models. Apply state of the art least-squares convergence theory to data assimilation. Improvements to initial state estimate may lead to a more accurate forecast.

False Discovery Rate Control for Lesion-Symptom Mapping With Heterogeneous Data via Weighted p-Values.

Lesion-symptom mapping studies provide insight into what areas of the brain are involved in different aspects of cognition. This is commonly done via behavioral testing in patients with a naturally occurring brain injury or lesions (e.g., strokes or brain tumors). This results in high-dimensional observational data where lesion status (present/absent) is nonuniformly distributed, with some voxels having lesions in very few (or no) subjects. In this situation, mass univariate hypothesis tests have severe power heterogeneity where many tests are known a priori to have little to no power. Recent advancements in multiple testing methodologies allow researchers to weigh hypotheses according to side information (e.g., information on power heterogeneity). In this paper, we propose the use of p-value weighting for voxel-based lesion-symptom mapping studies. The weights are created using the distribution of lesion status and spatial information to estimate different non-null prior probabilities for each hypothesis test through some common approaches. We provide a monotone minimum weight criterion, which requires minimum a priori power information. Our methods are demonstrated on dependent simulated data and an aphasia study investigating which regions of the brain are associated with the severity of language impairment among stroke survivors. The results demonstrate that the proposed methods have robust error control and can increase power. Further, we showcase how weights can be used to identify regions that are inconclusive due to lack ofpower.

DB-FCN: An end-to-end dual-branch fully convolutional nucleus detection model

Detecting cells or nuclei in histopathological image analysis is a crucial prerequisite for subsequent cancer diagnosis. By precisely locating cell nuclei, pathologists can quantitatively analyze the morphology of each cell nucleus. This enables accurate cancer grading, allowing for the implementation of tailored treatment plans based on distinct cancer stages. Manual nucleus detection methods are labor-intensive, making the development of automatic cell nucleus detection algorithms highly necessary. However, due to the small size of the nucleus and increased adhesions between cells, tissue staining may be uneven, leading to a higher occurrence of false positives in the results of the current automatic nucleus detection algorithm. This paper introduces an end-to-end dual-branch-based fully convolutional neural network (DB-FCN) that effectively addresses the aforementioned challenges, thereby enhancing the accuracy of automatic nucleus detection. This algorithm introduces for the first time the use of two detection branches, namely the coarse detection branch and the fine detection branch, to accomplish the detection task. The role of the coarse detection branch is to identify all cell regions in the pathological image as comprehensively as possible and then transmit the detection results as prior information to the fine detection branch. The fine detection branch is necessary solely to conduct more precise detection based on the coarse detection results. Given that the coarse detection branch has already eliminated interference from many background regions, the fine detection branch can focus on detecting the nucleus region, thereby significantly enhancing the efficiency and accuracy of model detection. The detection model proposed in this paper was evaluated on three classic datasets and compared with many existing detection algorithms. Compared with other algorithms, the detection algorithm proposed in this paper has made significant progress in detecting cell nuclei in histopathological images.

Enhancing Study Power in Pediatric Research through Bayesian Extrapolation

Pediatric populations, comprising individuals aged 18 or younger, and rare disease populations pose significant challenges in clinical trial design due to limited participant numbers, patient sensitivity, and insufficient natural history data. Addressing these challenges often involves data extrapolation, leveraging existing data from adults to inform pediatric trials. Bayesian hierarchical modeling is increasingly recognized as a valuable tool for combining information across disparate sources, such as adult and pediatric datasets. This manuscript introduces an extension of an existing statistical model to enhance the efficiency of borrowing strength from multiple historical trials under consistent assumptions. A quantitative method is developed to improve the borrowing of historical information. The methods are illustrated with a simulation study motivated by real case study in pediatric clinical trial, and practical considerations are provided regarding the selection of prior distributions. This work aims to provide comprehensive insights and practical guidance for leveraging historical data effectively in clinical trial design.

Bayesian method for fitting the low-energy constants in chiral perturbation theory

The values of the low-energy constants (LECs) are very important in the chiral perturbation theory. This paper adopts a Bayesian method with the truncation errors to globally fit eight next-to-leading order (NLO) LECs Lir and next-to-next-leading order (NNLO) LECs Cir. With the estimation of the truncation errors, the fitting results of Lir in the NLO and NNLO are very close. The posterior distributions of Cir indicate the boundary-dependent relations of these Cir. Ten Cir are weakly dependent on the boundaries and their values are reliable. The other Cir are required more experimental data to constrain their boundaries. Some linear combinations of Cir are also fitted with more reliable posterior distributions. If one knows some more precise values of Cir, some other Cir can be obtained by these values. With these fitting LECs, most observables provide a good convergence, except for the πK scattering lengths a03/2 and a01/2. An example is also introduced to test the improvement of the method. All the computations indicate that considering the truncation errors can improve the global fit greatly, and more prior information can obtain better fitting results. This fitting method can be extended to the other effective field theories and the perturbation theory.

Intrinsic Rewards for Exploration Without Harm From Observational Noise: A Simulation Study Based on the Free Energy Principle.

In reinforcement learning (RL), artificial agents are trained to maximize numerical rewards by performing tasks. Exploration is essential in RL because agents must discover information before exploiting it. Two rewards encouraging efficient exploration are the entropy of action policy and curiosity for information gain. Entropy is well established in the literature, promoting randomized action selection. Curiosity is defined in a broad variety of ways in literature, promoting discovery of novel experiences. One example, prediction error curiosity, rewards agents for discovering observations they cannot accurately predict. However, such agents may be distracted by unpredictable observational noises known as curiosity traps. Based on the free energy principle (FEP), this letter proposes hidden state curiosity, which rewards agents by the KL divergence between the predictive prior and posterior probabilities of latent variables. We trained six types of agents to navigate mazes: baseline agents without rewards for entropy or curiosity and agents rewarded for entropy and/or either prediction error curiosity or hidden state curiosity. We find that entropy and curiosity result in efficient exploration, especially both employed together. Notably, agents with hidden state curiosity demonstrate resilience against curiosity traps, which hinder agents with prediction error curiosity. This suggests implementing the FEP that may enhance the robustness and generalization of RL models, potentially aligning the learning processes of artificial and biological agents.

Closed-Form Method for Unified Far-Field and Near-Field Localization Based on TDOA and FDOA Measurements

When the near-field and far-field information of a target is uncertain, it is necessary to choose a suitable localization method. The modified polar representation (MPR) method integrates the two scenarios and achieves a unified localization with direction of arrival (DOA) estimation in the far field and position estimation in the near field. Previous studies have only proposed solutions for stationary environments and have not considered the motion factor. Therefore, this paper proposes a new unified positioning algorithm using multi-sensor time difference of arrival (TDOA) and frequency difference of arrival (FDOA) measurements without prior target source information. The method represents the position of the target source using MPR and describes the localization problem as a weighted least squares (WLS) problem with two constraints. We first obtain the initial estimates by WLS without considering the constraints and then investigate a two-step error correction method based on the constraints. The first step corrects the initial estimate using the Taylor series expansion technique, and the second step corrects the DOA estimate in the previous step using the direct error compensation technique based on the properties of the second constraint. Simulation experiments show that the method is effective for the unified positioning of moving targets and can achieve the Cramer–Rao lower bound (CRLB).

New clinical trial design borrowing information across patient subgroups based on fusion-penalized regression models.

In cancer research, basket trials aim to assess the efficacy of a drug using baskets, wherein patients are organized into subgroups according to their tumor type. In this context, using information borrowing strategy may increase the probability of detecting drug efficacy in active baskets, by shrinking together the estimates of the parameters characterizing the drug efficacy in baskets with similar drug activity. Here, we propose to use fusion-penalized logistic regression models to borrow information in the setting of a phase 2 single-arm basket trial with binary outcome. We describe our proposed strategy and assess its performance via a simulation study. We assessed the impact of heterogeneity in drug efficacy, prevalence of each tumor types and implementation of interim analyses on the operating characteristics of our proposed design. We compared our approach with two existing designs, relying on the specification of prior information in a Bayesian framework to borrow information across similar baskets. Notably, our approach performed well when the effect of the drug varied greatly across the baskets. Our approach offers several advantages, including limited implementation efforts and fast computation, which is essential when planning a new trial as such planning requires intensive simulation studies.

Senecio inaequidens DC. will thrive in future climate: A case study in a Mediterranean biodiversity hotspot

Monitoring the expansion of invasive non-native plants under current and future climatic conditions is crucial for understanding biodiversity threats, addressing the ecological impact, and developing effective management strategies. This study focuses on modelling the expansion and distribution of Senecio inaequidens DC. on the island of Sardinia (Italy) to address these environmental challenges. The objectives were to identify bio-climatically suitable areas under current conditions, project potential future distribution, and evaluate invasion dynamics on the island to localize suitable areas for effective management strategies.Species data were collected from 1991 to the present, supplemented by global databases and analyzed using an ensemble species distribution model approach. This approach utilized presence data, high-resolution current bioclimatic variables (40 m2), developed explicitly for our study area, and two future scenarios derived from the newly Coupled Model Intercomparison Project Phase 6 (CMIP6) under Shared Socio-economic Pathways (SSP2–4.5 and SSP5–8.5 for 2040 and 2060).The ensemble model's findings suggest a close alignment between the currently documented occurrences of S. inaequidens and its bio-climatically suitable habitats in Sardinia. Moreover, predictions indicate high bioclimatic suitability for S. inaequidens in the western and southwestern coastal regions, contrasting with its known occurrences at higher altitudes. Notably, the model also forecasts high bio-climatic suitability across most small islands surrounding the region and in central-east Sardinia, potentially indicating habitats at lower altitudes compared to current records.Under the SSP2–4.5 scenario, suitable areas are expected to nearly double by 2040 and more than double by 2060, compared to current conditions. Under the SSP5–8.5 scenario, the increase in suitable habitats is projected to be about 83.31% by 2040 and more than double by 2060. These results highlight the species' ability to thrive under climate change, with a more pronounced range expansion under the pessimistic management regime (SSP5–8.5) than under the intermediate one (SSP2–4.5), particularly in the central region of the island. This expansion under the more severe management scenario is particularly alarming as it reflects limited implementation of environmental management policies.The study underscores the potential ecological risk posed by S. inaequidens due to its potential range expansion and ability to invade different habitat types, from coastal regions to mountainous areas, under current and future scenarios. Based on these findings, we propose targeted management actions for monitoring and eradicating the species, leveraging prior information and local experiences to mitigate its impact.

A mechanistic framework for the leakage risk reduction of mobile hydrogen refueling stations based on inherent safety concepts

This study presents a mechanistic framework for assessing and mitigating hydrogen leakage risks in Mobile Hydrogen Refueling Stations (MHRS) based on inherent safety concepts. A bow-tie model is proposed to identify potential hazards, event pathways, and consequences. Subsequently, this model is mapped into Bayesian Networks (BN) to enable quantitative risk assessment. Fuzzy set theory is adopted to address the uncertainty of prior probabilities, using a Noisy-MAX model to optimize the BN conditional probabilities. To mitigate leakage risks, a dedicated inherently safer design checklist is developed to identify opportunities for incorporating inherent safety. A case study demonstrates that the leakage potentials decrease from 0.2600 to 0.1997, and the leakage consequences decrease by approximately 22%. This work presents a novel mechanism using inherent safety to reduce the hydrogen leakage risk of MHRS, which is useful for creating an inherently safer MHRS with fundamental and natural robustness against hydrogen leakage.

A global argument-based algorithm for finding complex zeros and poles to investigate plasma kinetic instabilities

Root-finding is a common problem in various fields of physics and mathematics. The basic version of this problem can be formulated as an equation f(z)=0 where z∈ℂ and f is a meromorphic function on a complex plane ℂ. In some cases, a conjugated problem has to be solved as well to investigate the properties of the inspected system: f(z)−1=0. Iterative Householder-like methods show high convergence rate, however, the convergence itself is strongly dependent on the input parameters such as an initial guess. If the function f is complicated enough, there can be no prior information about its zeros and poles, and thus, no way to correctly pick the initial guess. The aim of the algorithm proposed in this article is to find all zeros and poles inside a selected region without any information about the local behavior of the function f. Additionally, the method is generalized to a parametric problem with continuous solutions: f(z)→f(z;p) where p∈Rd, d∈N∪{0}. The basis of the algorithm consists of a triangulation process and an implementation of the Cauchy’s argument principle to inspect the constructed subregions. The refinement process is accompanied by an absolute value gradient ∇ln|f| analysis that significantly decreases the number of points needed to correctly identify the locations of all zeros and poles. The algorithm is presented as a hybrid class for d∈{0,1} that can be directly used to solve corresponding problems.

Reliable multi-modal medical image-to-image translation independent of pixel-wise aligned data.

The current mainstream multi-modal medical image-to-image translation methods face a contradiction. Supervised methods with outstanding performance rely on pixel-wise aligned training data to constrain the model optimization. However, obtaining pixel-wise aligned multi-modal medical image datasets is challenging. Unsupervised methods can be trained without paired data, but their reliability cannot be guaranteed. At present, there is no ideal multi-modal medical image-to-image translation method that can generate reliable translation results without the need for pixel-wise aligned data. This work aims to develop a novel medical image-to-image translation model that is independent of pixel-wise aligned data (MITIA), enabling reliable multi-modal medical image-to-image translation under the condition of misaligned trainingdata. The proposed MITIA model utilizes a prior extraction network composed of a multi-modal medical image registration module and a multi-modal misalignment error detection module to extract pixel-level prior information from training data with misalignment errors to the largest extent. The extracted prior information is then used to construct a regularization term to constrain the optimization of the unsupervised cycle-consistent Generative Adversarial Network model, restricting its solution space and thereby improving the performance and reliability of the generator. We trained the MITIA model using six datasets containing different misalignment errors and two well-aligned datasets. Subsequently, we conducted quantitative analysis using peak signal-to-noise ratio and structural similarity as metrics. Moreover, we compared the proposed method with six other state-of-the-art image-to-image translationmethods. The results of both quantitative analysis and qualitative visual inspection indicate that MITIA achieves superior performance compared to the competing state-of-the-art methods, both on misaligned data and aligned data. Furthermore, MITIA shows more stability in the presence of misalignment errors in the training data, regardless of their severity or type. The proposed method achieves outstanding performance in multi-modal medical image-to-image translation tasks without aligned training data. Due to the difficulty in obtaining pixel-wise aligned data for medical image translation tasks, MITIA is expected to generate significant application value in this scenario compared to existingmethods.

Feature extractor optimization for discriminative representations in Generalized Category Discovery

Generalized Category Discovery (GCD) task involves transferring knowledge from labeled known categories to recognize both known and novel categories within an unlabeled dataset. A significant challenge arises from the lack of prior information for novel categories. To address this, we develop a feature extractor that can learn discriminative features for both known and novel categories. Our approach leverages the observation that similar samples often belong to the same class. We construct a similarity matrix and employ similarity contrastive loss to increase the similarity between similar samples in the feature space. Additionally, we incorporate cluster labels to further refine the feature extractor, utilizing K-means clustering to assign these labels to unlabeled data, providing valuable supervision. Our feature extractor is optimized through the utilization of instance-level contrastive learning and class-level contrastive learning constraints. These constraints promote similarity maximization in both the instance space and the label space for instances sharing the same pseudo-labels. These three components complement each other, facilitating the learning of discriminative representations for both known and novel categories. Through comprehensive evaluations of generic image recognition datasets and challenging fine-grained datasets, we demonstrate that our proposed method achieves state-of-the-art performance.

Three layered sparse dictionary learning algorithm for enhancing the subject wise segregation of brain networks

Independent component analysis (ICA) and dictionary learning (DL) are the most successful blind source separation (BSS) methods for functional magnetic resonance imaging (fMRI) data analysis. However, ICA to higher and DL to lower extent may suffer from performance degradation by the presence of anomalous observations in the recovered time courses (TCs) and high overlaps among spatial maps (SMs). This paper addressed both problems using a novel three-layered sparse DL (TLSDL) algorithm that incorporated prior information in the dictionary update process and recovered full-rank outlier-free TCs from highly corrupted measurements. The associated sequential DL model involved factorizing each subject’s data into a multi-subject (MS) dictionary and MS sparse code while imposing a low-rank and a sparse matrix decomposition restriction on the dictionary matrix. It is derived by solving three layers of feature extraction and component estimation. The first and second layers captured brain regions with low and moderate spatial overlaps, respectively. The third layer that segregated regions with significant spatial overlaps solved a sequence of vector decomposition problems using the proximal alternating linearized minimization (PALM) method and solved a decomposition restriction using the alternating directions method (ALM). It learned outlier-free dynamics that integrate spatiotemporal diversities across brains and external information. It differs from existing DL methods owing to its unique optimization model, which incorporates prior knowledge, subject-wise/multi-subject representation matrices, and outlier handling. The TLSDL algorithm was compared with existing dictionary learning algorithms using experimental and synthetic fMRI datasets to verify its performance. Overall, the mean correlation value was found to be 26%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$26\\%$$\\end{document} higher for the TLSDL than for the state-of-the-art subject-wise sequential DL (swsDL) technique.

A unified framework of semi-supervised community detection integrating network topology and node content

Detecting the community structure within networks is important because it aids in the analysis of complex networks. However, the existence of diverse complex structures in the network topology can limit the topological information’s ability to represent communities. For instance, the accuracy of many traditional methods based on topology alone decreases sharply when a link is missing or noisy. To solve the above-mentioned issues simultaneously, we propose a unified framework of semi-supervised community detection integrating content information. Three models are proposed, instead of a special model-handling problem. The framework used performs the same interpretation for a target problem. The inputs contain the adjacent and node content matrix. The proposed models describe the network topology using a non-negative matrix factorization (NMF) algorithm–based generative model or modularity maximization algorithm. Additionally, we use the equivalence theory of the pLSI, and NMF models the content information based on NMF. Community membership is also used as the relationship between communities and node contents. The models uniformly integrate content information and network topology. Further, a constraint matrix that depicts must-links is built by calculating the similarity of neighbors between any two nodes. Must-link pairs can represent the network topology’s prior information. The models’ community memberships are adjusted by using a strong constraint based on the connected component of this matrix. The models also introduce prior information into the topological and content information in the mentioned models. We compared the proposed models with the state-of-the-art community detection models on artificial networks and five real networks. The experimental results show that the proposed models outperformed the models combining topology and content and semi-supervised models. Fusing the topology with the content and the prior information in the models was also effective for community detection. Thus, the models under the framework are more advantageous than other semi-supervised models when incorporating the same amount of prior information into the models. The semi-supervised framework used also has a certain level of competitiveness in detecting communities.

Exploring the Trade-Off between generalist and specialized Models: A center-based comparative analysis for glioblastoma segmentation

IntroductionInherent variations between inter-center data can undermine the robustness of segmentation models when applied at a specific center (dataset shift). We investigated whether specialized center-specific models are more effective compared to generalist models based on multi-center data, and how center-specific data could enhance the performance of generalist models within a particular center using a fine-tuning transfer learning approach. For this purpose, we studied the dataset shift at center level and conducted a comparative analysis to assess the impact of data source on glioblastoma segmentation models. Methods & MaterialsThe three key components of dataset shift were studied: prior probability shift—variations in tumor size or tissue distribution among centers; covariate shift—inter-center MRI alterations; and concept shift—different criteria for tumor segmentation. BraTS 2021 dataset was used, which includes 1251 cases from 23 centers. Thereafter, 155 deep-learning models were developed and compared, including 1) generalist models trained with multi-center data, 2) specialized models using only center-specific data, and 3) fine-tuned generalist models using center-specific data. ResultsThe three key components of dataset shift were characterized. The amount of covariate shift was substantial, indicating large variations in MR imaging between different centers. Glioblastoma segmentation models tend to perform best when using data from the application center. Generalist models, trained with over 700 samples, achieved a median Dice score of 88.98%. Specialized models surpassed this with 200 cases, while fine-tuned models outperformed with 50 cases. ConclusionsThe influence of dataset shift on model performance is evident. Fine-tuned and specialized models, utilizing data from the evaluated center, outperform generalist models, which rely on data from other centers. These approaches could encourage medical centers to develop customized models for their local use, enhancing the accuracy and reliability of glioblastoma segmentation in a context where dataset shift is inevitable.

Top-Down Priors Disambiguate Target and Distractor Features in Simulated Covert Visual Search.

Several models of visual search consider visual attention as part of a perceptual inference process, in which top-down priors disambiguate bottom-up sensory information. Many of these models have focused on gaze behavior, but there are relatively fewer models of covert spatial attention, in which attention is directed to a peripheral location in visual space without a shift in gaze direction. Here, we propose a biologically plausible model of covert attention during visual search that helps to bridge the gap between Bayesian modeling and neurophysiological modeling by using top-down priors over target features that are acquired through Hebbian learning and spatial resampling of modeled cortical receptive fields to enhance local spatial resolution of image representations for downstream target classification. By training a simple generative model using a Hebbian update rule, top-down priors for target features naturally emerge without the need for hand-tuned or predetermined priors. Furthermore, the implementation of covert spatial attention in our model is based on a known neurobiological mechanism, providing a plausible process through which Bayesian priors could locally enhance the spatial resolution of image representations. We validate this model during simulated visual search for handwritten digits among nondigit distractors, demonstrating that top-down priors improve accuracy for estimation of target location and classification, relative to bottom-up signals alone. Our results support previous reports in the literature that demonstrated beneficial effects of top-down priors on visual search performance while extending this literature to incorporate known neural mechanisms of covert spatial attention.