Functional connectivity alterations in traumatic brain injury patients with late seizures.

PTE is a neurological disorder characterized by recurrent and spontaneous epileptic seizures. PTE is a major public health problem occurring in 2-50% of TBI patients. Identifying PTE biomarkers is crucial for the development of effective treatments. Functional neuroimaging studies in patients with epilepsy and in epileptic rodents have observed that abnormal functional brain activity plays a role in the development of epilepsy. Network representations of complex systems ease quantitative analysis of heterogeneous interactions within a unified mathematical framework. In this work, graph theory was used to study resting state functional magnetic resonance imaging (rs-fMRI) and reveal functional connectivity abnormalities that are associated with seizure development in traumatic brain injury (TBI) patients. We examined rs-fMRI of 75 TBI patients from Epilepsy Bioinformatics Study for Antiepileptogenic Therapy (EpiBioS4Rx) which aims to identify validated Post-traumatic epilepsy (PTE) biomarkers and antiepileptogenic therapies using multimodal and longitudinal data acquired from 14 international sites. The dataset includes 28 subjects who had at least one late seizure after TBI and 47 subjects who had no seizures within 2 years post-injury. Each subject's neural functional network was investigated by computing the correlation between the low frequency time series of 116 regions of interest (ROIs). Each subject's functional organization was represented as a network consisting of nodes, brain regions, and edges that show the relationship between the nodes. Then, several graph measures concerning the integration and the segregation of the functional brain networks were extracted in order to highlight changes in functional connectivity between the two TBI groups. Results showed that the late seizure-affected group had a compromised balance between integration and segregation and presents functional networks that are hyperconnected, hyperintegrated but at the same time hyposegregated compared with seizure-free patients. Moreover, TBI subjects who developed late seizures had more low betweenness hubs.

Similarity of hand muscle synergies elicited by transcranial magnetic stimulation and those found during voluntary movement.

Converging evidence in human and animal models suggests that exogenous stimulation of the motor cortex (M1) elicits responses in the hand with similar modular structure to that found during voluntary grasping movements. The aim of this study was to establish the extent to which modularity in muscle responses to transcranial magnetic stimulation (TMS) to M1 resembles modularity in muscle activation during voluntary hand movements involving finger fractionation. Electromyography (EMG) was recorded from eight hand-forearm muscles in eight healthy individuals. Modularity was defined using non-negative matrix factorization to identify low-rank approximations (spatial muscle synergies) of the complex activation patterns of EMG data recorded during high-density TMS mapping of M1 and voluntary formation of gestures in the American Sign Language alphabet. Analysis of synergies revealed greater than chance similarity between those derived from TMS and those derived from voluntary movement. Both data sets included synergies dominated by single intrinsic hand muscles presumably to meet the demand for highly fractionated finger movement. These results suggest that corticospinal connectivity to individual intrinsic hand muscles may be combined with modular multimuscle activation via synergies in the formation of hand postures.NEW & NOTEWORTHY This is the first work to examine the similarity of modularity in hand muscle responses to transcranial magnetic stimulation (TMS) of the motor cortex and that derived from voluntary hand movement. We show that TMS-elicited muscle synergies of the hand, measured at rest, reflect those found in voluntary behavior involving finger fractionation. This work provides a basis for future work using TMS to investigate muscle activation modularity in the human motor system.

Geometric Analysis of Uncertainty Sampling for Dense Neural Network Layer.

For model adaptation of fully connected neural network layers, we provide an information geometric and sample behavioral active learning uncertainty sampling objective analysis. We identify conditions under which several uncertainty-based methods have the same performance and show that such conditions are more likely to appear in the early stages of learning. We define riskier samples for adaptation, and demonstrate that, as the set of labeled samples increases, margin-based sampling outperforms other uncertainty sampling methods by preferentially selecting these risky samples. We support our derivations and illustrations with experiments using Meta-Dataset, a benchmark for few-shot learning. We compare uncertainty-based active learning objectives using features produced by SimpleCNAPS (a state-of-the-art few-shot classifier) as input for a fully-connected adaptation layer. Our results indicate that margin-based uncertainty sampling achieves similar performance as other uncertainty based sampling methods with fewer labelled samples as discussed in the novel geometric analysis.

Stochastic Mutual Information Gradient Estimation for Dimensionality Reduction Networks.

Feature ranking and selection is a widely used approach in various applications of supervised dimensionality reduction in discriminative machine learning. Nevertheless there exists significant evidence on feature ranking and selection algorithms based on any criterion leading to potentially sub-optimal solutions for class separability. In that regard, we introduce emerging information theoretic feature transformation protocols as an end-to-end neural network training approach. We present a dimensionality reduction network (MMINet) training procedure based on the stochastic estimate of the mutual information gradient. The network projects high-dimensional features onto an output feature space where lower dimensional representations of features carry maximum mutual information with their associated class labels. Furthermore, we formulate the training objective to be estimated non-parametrically with no distributional assumptions. We experimentally evaluate our method with applications to high-dimensional biological data sets, and relate it to conventional feature selection algorithms to form a special case of our approach.

Disentangled Adversarial Autoencoder for Subject-Invariant Physiological Feature Extraction.

Recent developments in biosignal processing have enabled users to exploit their physiological status for manipulating devices in a reliable and safe manner. One major challenge of physiological sensing lies in the variability of biosignals across different users and tasks. To address this issue, we propose an adversarial feature extractor for transfer learning to exploit disentangled universal representations. We consider the trade-off between task-relevant features and user-discriminative information by introducing additional adversary and nuisance networks in order to manipulate the latent representations such that the learned feature extractor is applicable to unknown users and various tasks. Results on cross-subject transfer evaluations exhibit the benefits of the proposed framework, with up to 8.8% improvement in average accuracy of classification, and demonstrate adaptability to a broader range of subjects.

Adversarial Deep Learning in EEG Biometrics.

Deep learning methods for person identification based on electroencephalographic (EEG) brain activity encounters the problem of exploiting the temporally correlated structures or recording session specific variability within EEG. Furthermore, recent methods have mostly trained and evaluated based on single session EEG data. We address this problem from an invariant representation learning perspective. We propose an adversarial inference approach to extend such deep learning models to learn session-invariant person-discriminative representations that can provide robustness in terms of longitudinal usability. Using adversarial learning within a deep convolutional network, we empirically assess and show improvements with our approach based on longitudinally collected EEG data for person identification from half-second EEG epochs.

On Analysis of Active Querying for Recursive State Estimation.

In stochastic linear/non-linear active dynamic systems, states are estimated with the evidence through recursive measurements in response to queries of the system about the state to be estimated. Therefore, query selection is essential for such systems to improve state estimation accuracy and time. Query selection is conventionally achieved by minimization of the evidence variance or optimization of various information theoretic objectives. It was shown that optimization of mutual information-based objectives and variance-based objectives arrive at the same solution. However, existing approaches optimize approximations to the intended objectives rather than solving the exact optimization problems. To overcome these shortcomings, we propose an active querying procedure using mutual information maximization in recursive state estimation. First we show that mutual information generalizes variance based query selection methods and show the equivalence between objectives if the evidence likelihoods have unimodal distributions. We then solve the exact optimization problem for query selection and propose a query (measurement) selection algorithm. We specifically formulate the mutual information maximization for query selection as a combinatorial optimization problem and show that the objective is sub-modular, therefore can be solved efficiently with guaranteed convergence bounds through a greedy approach. Additionally, we analyze the performance of the query selection algorithm by testing it through a brain computer interface typing system.

Optimal Query Selection Using Multi-Armed Bandits.

Query selection for latent variable estimation is conventionally performed by opting for observations with low noise or optimizing information theoretic objectives related to reducing the level of estimated uncertainty based on the current best estimate. In these approaches, typically the system makes a decision by leveraging the current available information about the state. However, trusting the current best estimate results in poor query selection when truth is far from the current estimate, and this negatively impacts the speed and accuracy of the latent variable estimation procedure. We introduce a novel sequential adaptive action value function for query selection using the multi-armed bandit (MAB) framework which allows us to find a tractable solution. For this adaptive-sequential query selection method, we analytically show: (i) performance improvement in the query selection for a dynamical system, (ii) the conditions where the model outperforms competitors. We also present favorable empirical assessments of the performance for this method, compared to alternative methods, both using Monte Carlo simulations and human-in-the-loop experiments with a brain computer interface (BCI) typing system where the language model provides the prior information.

Effects of Catalytic Action and Ligand Binding on Conformational Ensembles of Adenylate Kinase.

Crystal structures of adenylate kinase (AdK) from Escherichia coli capture two states: an "open" conformation (apo) obtained in the absence of ligands and a "closed" conformation in which ligands are bound. Other AdK crystal structures suggest intermediate conformations that may lie on the transition pathway between these two states. To characterize the transition from open to closed states in solution, X-ray solution scattering data were collected from AdK in the apo form and with progressively increasing concentrations of five different ligands. Scattering data from apo AdK are consistent with scattering predicted from the crystal structure of AdK in the open conformation. In contrast, data from AdK samples saturated with Ap5A do not agree with that calculated from AdK in the closed conformation. Using cluster analysis of available structures, we selected representative structures in five conformational states: open, partially open, intermediate, partially closed, and closed. We used these structures to estimate the relative abundances of these states for each experimental condition. X-ray solution scattering data obtained from AdK with AMP are dominated by scattering from AdK in the open conformation. For AdK in the presence of high concentrations of ATP and ADP, the conformational ensemble shifts to a mixture of partially open and closed states. Even when AdK is saturated with Ap5A, a significant proportion of AdK remains in a partially open conformation. These results are consistent with an induced-fit model in which the transition of AdK from an open state to a closed state is initiated by ATP binding.

An Active RBSE Framework to Generate Optimal Stimulus Sequences in a BCI for Spelling.

A class of brain computer interfaces (BCIs) employs noninvasive recordings of electroencephalography (EEG) signals to enable users with severe speech and motor impairments to interact with their environment and social network. For example, EEG based BCIs for typing popularly utilize event related potentials (ERPs) for inference. Presentation paradigm design in current ERP-based letter by letter typing BCIs typically query the user with an arbitrary subset characters. However, the typing accuracy and also typing speed can potentially be enhanced with more informed subset selection and flash assignment. In this manuscript, we introduce the active recursive Bayesian state estimation (active-RBSE) framework for inference and sequence optimization. Prior to presentation in each iteration, rather than showing a subset of randomly selected characters, the developed framework optimally selects a subset based on a query function. Selected queries are made adaptively specialized for users during each intent detection. Through a simulation-based study, we assess the effect of active-RBSE on the performance of a language-model assisted typing BCI in terms of typing speed and accuracy. To provide a baseline for comparison, we also utilize standard presentation paradigms namely, row and column matrix presentation paradigm and also random rapid serial visual presentation paradigms. The results show that utilization of active-RBSE can enhance the online performance of the system, both in terms of typing accuracy and speed. Moreover, we conduct real time experiments with human participants to study the human-in-the-loop effect on the performance of the proposed active-RBSE framework and consistent with the simulation results, the results of these experiments show improvement both in typing speed and accuracy.

Plus Disease in Retinopathy of Prematurity: A Continuous Spectrum of Vascular Abnormality as a Basis of Diagnostic Variability.

PURPOSE: To identify patterns of interexpert discrepancy in plus disease diagnosis in retinopathy of prematurity (ROP). DESIGN: We developed 2 datasets of clinical images as part of the Imaging and Informatics in ROP study and determined a consensus reference standard diagnosis (RSD) for each image based on 3 independent image graders and the clinical examination results. We recruited 8 expert ROP clinicians to classify these images and compared the distribution of classifications between experts and the RSD. PARTICIPANTS: Eight participating experts with more than 10 years of clinical ROP experience and more than 5 peer-reviewed ROP publications who analyzed images obtained during routine ROP screening in neonatal intensive care units. METHODS: Expert classification of images of plus disease in ROP. MAIN OUTCOME MEASURES: Interexpert agreement (weighted κ statistic) and agreement and bias on ordinal classification between experts (analysis of variance [ANOVA]) and the RSD (percent agreement). RESULTS: There was variable interexpert agreement on diagnostic classifications between the 8 experts and the RSD (weighted κ, 0-0.75; mean, 0.30). The RSD agreement ranged from 80% to 94% for the dataset of 100 images and from 29% to 79% for the dataset of 34 images. However, when images were ranked in order of disease severity (by average expert classification), the pattern of expert classification revealed a consistent systematic bias for each expert consistent with unique cut points for the diagnosis of plus disease and preplus disease. The 2-way ANOVA model suggested a highly significant effect of both image and user on the average score (dataset A: P < 0.05 and adjusted R2 = 0.82; and dataset B: P < 0.05 and adjusted R2 = 0.6615). CONCLUSIONS: There is wide variability in the classification of plus disease by ROP experts, which occurs because experts have different cut points for the amounts of vascular abnormality required for presence of plus and preplus disease. This has important implications for research, teaching, and patient care for ROP and suggests that a continuous ROP plus disease severity score may reflect more accurately the behavior of expert ROP clinicians and may better standardize classification in the future.

Plus Disease in Retinopathy of Prematurity: Improving Diagnosis by Ranking Disease Severity and Using Quantitative Image Analysis.

PURPOSE: To determine expert agreement on relative retinopathy of prematurity (ROP) disease severity and whether computer-based image analysis can model relative disease severity, and to propose consideration of a more continuous severity score for ROP. DESIGN: We developed 2 databases of clinical images of varying disease severity (100 images and 34 images) as part of the Imaging and Informatics in ROP (i-ROP) cohort study and recruited expert physician, nonexpert physician, and nonphysician graders to classify and perform pairwise comparisons on both databases. PARTICIPANTS: Six participating expert ROP clinician-scientists, each with a minimum of 10 years of clinical ROP experience and 5 ROP publications, and 5 image graders (3 physicians and 2 nonphysician graders) who analyzed images that were obtained during routine ROP screening in neonatal intensive care units. METHODS: Images in both databases were ranked by average disease classification (classification ranking), by pairwise comparison using the Elo rating method (comparison ranking), and by correlation with the i-ROP computer-based image analysis system. MAIN OUTCOME MEASURES: Interexpert agreement (weighted κ statistic) compared with the correlation coefficient (CC) between experts on pairwise comparisons and correlation between expert rankings and computer-based image analysis modeling. RESULTS: There was variable interexpert agreement on diagnostic classification of disease (plus, preplus, or normal) among the 6 experts (mean weighted κ, 0.27; range, 0.06-0.63), but good correlation between experts on comparison ranking of disease severity (mean CC, 0.84; range, 0.74-0.93) on the set of 34 images. Comparison ranking provided a severity ranking that was in good agreement with ranking obtained by classification ranking (CC, 0.92). Comparison ranking on the larger dataset by both expert and nonexpert graders demonstrated good correlation (mean CC, 0.97; range, 0.95-0.98). The i-ROP system was able to model this continuous severity with good correlation (CC, 0.86). CONCLUSIONS: Experts diagnose plus disease on a continuum, with poor absolute agreement on classification but good relative agreement on disease severity. These results suggest that the use of pairwise rankings and a continuous severity score, such as that provided by the i-ROP system, may improve agreement on disease severity in the future.

Constrained Maximum Likelihood Estimation of Relative Abundances of Protein Conformation in a Heterogeneous Mixture from Small Angle X-Ray Scattering Intensity Measurements.

In this paper, we describe a model for maximum likelihood estimation (MLE) of the relative abundances of different conformations of a protein in a heterogeneous mixture from small angle X-ray scattering (SAXS) intensities. To consider cases where the solution includes intermediate or unknown conformations, we develop a subset selection method based on k-means clustering and the Cramér-Rao bound on the mixture coefficient estimation error to find a sparse basis set that represents the space spanned by the measured SAXS intensities of the known conformations of a protein. Then, using the selected basis set and the assumptions on the model for the intensity measurements, we show that the MLE model can be expressed as a constrained convex optimization problem. Employing the adenylate kinase (ADK) protein and its known conformations as an example, and using Monte Carlo simulations, we demonstrate the performance of the proposed estimation scheme. Here, although we use 45 crystallographically determined experimental structures and we could generate many more using, for instance, molecular dynamics calculations, the clustering technique indicates that the data cannot support the determination of relative abundances for more than 5 conformations. The estimation of this maximum number of conformations is intrinsic to the methodology we have used here.

Quantitative change of EEG and respiration signals during mindfulness meditation.

BACKGROUND: This study investigates measures of mindfulness meditation (MM) as a mental practice, in which a resting but alert state of mind is maintained. A population of older people with high stress level participated in this study, while electroencephalographic (EEG) and respiration signals were recorded during a MM intervention. The physiological signals during meditation and control conditions were analyzed with signal processing. METHODS: EEG and respiration data were collected and analyzed on 34 novice meditators after a 6-week meditation intervention. Collected data were analyzed with spectral analysis, phase analysis and classification to evaluate an objective marker for meditation. RESULTS: Different frequency bands showed differences in meditation and control conditions. Furthermore, we established a classifier using EEG and respiration signals with a higher accuracy (85%) at discriminating between meditation and control conditions than a classifier using the EEG signal only (78%). CONCLUSION: Support vector machine (SVM) classifier with EEG and respiration feature vector is a viable objective marker for meditation ability. This classifier should be able to quantify different levels of meditation depth and meditation experience in future studies.

Manifold Learning by Preserving Distance Orders.

Nonlinear dimensionality reduction is essential for the analysis and the interpretation of high dimensional data sets. In this manuscript, we propose a distance order preserving manifold learning algorithm that extends the basic mean-squared error cost function used mainly in multidimensional scaling (MDS)-based methods. We develop a constrained optimization problem by assuming explicit constraints on the order of distances in the low-dimensional space. In this optimization problem, as a generalization of MDS, instead of forcing a linear relationship between the distances in the high-dimensional original and low-dimensional projection space, we learn a non-decreasing relation approximated by radial basis functions. We compare the proposed method with existing manifold learning algorithms using synthetic datasets based on the commonly used residual variance and proposed percentage of violated distance orders metrics. We also perform experiments on a retinal image dataset used in Retinopathy of Prematurity (ROP) diagnosis.

M2M-InvNet: Human Motor Cortex Mapping From Multi-Muscle Response Using TMS and Generative 3D Convolutional Network.

Transcranial magnetic stimulation (TMS) is often applied to the motor cortex to stimulate a collection of motor evoked potentials (MEPs) in groups of peripheral muscles. The causal interface between TMS and MEP is the selective activation of neurons in the motor cortex; moving around the TMS 'spot' over the motor cortex causes different MEP responses. A question of interest is whether a collection of MEP responses can be used to identify the stimulated locations on the cortex, which could potentially be used to then place the TMS coil to produce chosen sets of MEPs. In this work we leverage our previous report on a 3D convolutional neural network (CNN) architecture that predicted MEPs from the induced electric field, to tackle an inverse imaging task in which we start with the MEPs and estimate the stimulated regions on the motor cortex. We present and evaluate five different inverse imaging CNN architectures, both conventional and generative, in terms of several measures of reconstruction accuracy. We found that one architecture, which we propose as M2M-InvNet, consistently achieved the best performance.

Advancing post-traumatic seizure classification and biomarker identification: Information decomposition based multimodal fusion and explainable machine learning with missing neuroimaging data.

A late post-traumatic seizure (LPTS), a consequence of traumatic brain injury (TBI), can potentially evolve into a lifelong condition known as post-traumatic epilepsy (PTE). Presently, the mechanism that triggers epileptogenesis in TBI patients remains elusive, inspiring the epilepsy community to devise ways to predict which TBI patients will develop PTE and to identify potential biomarkers. In response to this need, our study collected comprehensive, longitudinal multimodal data from 48 TBI patients across multiple participating institutions. A supervised binary classification task was created, contrasting data from LPTS patients with those without LPTS. To accommodate missing modalities in some subjects, we took a two-pronged approach. Firstly, we extended a graphical model-based Bayesian estimator to directly classify subjects with incomplete modality. Secondly, we explored conventional imputation techniques. The imputed multimodal information was then combined, following several fusion and dimensionality reduction techniques found in the literature, and subsequently fitted to a kernel- or a tree-based classifier. For this fusion, we proposed two new algorithms: recursive elimination of correlated components (RECC) that filters information based on the correlation between the already selected features, and information decomposition and selective fusion (IDSF), which effectively recombines information from decomposed multimodal features. Our cross-validation findings showed that the proposed IDSF algorithm delivers superior performance based on the area under the curve (AUC) score. Ultimately, after rigorous statistical comparisons and interpretable machine learning examination using Shapley values of the most frequently selected features, we recommend the two following magnetic resonance imaging (MRI) abnormalities as potential biomarkers: the left anterior limb of internal capsule in diffusion MRI (dMRI), and the right middle temporal gyrus in functional MRI (fMRI).

Multimodal fusion of EMG and vision for human grasp intent inference in prosthetic hand control.

Objective: For transradial amputees, robotic prosthetic hands promise to regain the capability to perform daily living activities. Current control methods based on physiological signals such as electromyography (EMG) are prone to yielding poor inference outcomes due to motion artifacts, muscle fatigue, and many more. Vision sensors are a major source of information about the environment state and can play a vital role in inferring feasible and intended gestures. However, visual evidence is also susceptible to its own artifacts, most often due to object occlusion, lighting changes, etc. Multimodal evidence fusion using physiological and vision sensor measurements is a natural approach due to the complementary strengths of these modalities. Methods: In this paper, we present a Bayesian evidence fusion framework for grasp intent inference using eye-view video, eye-gaze, and EMG from the forearm processed by neural network models. We analyze individual and fused performance as a function of time as the hand approaches the object to grasp it. For this purpose, we have also developed novel data processing and augmentation techniques to train neural network components. Results: Our results indicate that, on average, fusion improves the instantaneous upcoming grasp type classification accuracy while in the reaching phase by 13.66% and 14.8%, relative to EMG (81.64% non-fused) and visual evidence (80.5% non-fused) individually, resulting in an overall fusion accuracy of 95.3%. Conclusion: Our experimental data analyses demonstrate that EMG and visual evidence show complementary strengths, and as a consequence, fusion of multimodal evidence can outperform each individual evidence modality at any given time.

User Training With Error Augmentation for sEMG-Based Gesture Classification.

We designed and tested a system for real-time control of a user interface by extracting surface electromyographic (sEMG) activity from eight electrodes in a wristband configuration. sEMG data were streamed into a machine-learning algorithm that classified hand gestures in real-time. After an initial model calibration, participants were presented with one of three types of feedback during a human-learning stage: veridical feedback, in which predicted probabilities from the gesture classification algorithm were displayed without alteration; modified feedback, in which we applied a hidden augmentation of error to these probabilities; and no feedback. User performance was then evaluated in a series of minigames, in which subjects were required to use eight gestures to manipulate their game avatar to complete a task. Experimental results indicated that relative to the baseline, the modified feedback condition led to significantly improved accuracy. Class separation also improved, though this trend was not significant. These findings suggest that real-time feedback in a gamified user interface with manipulation of feedback may enable intuitive, rapid, and accurate task acquisition for sEMG-based gesture recognition applications.

Oxygenation Fluctuations Associated with Severe Retinopathy of Prematurity: Insights from a Multimodal Deep Learning Approach.

Purpose: Retinopathy of prematurity (ROP) is one of the leading causes of blindness in children. Although the role of oxygen in the pathophysiology of ROP is well established, a precise understanding of the dynamic relationship between oxygen exposure ROP incidence and severity is lacking. The purpose of this study was to evaluate the correlation between time-dependent oxygen variables and the onset of ROP. Design: Retrospective cohort study. Participants: Two hundred thirty infants who were born at a single academic center and met the inclusion criteria were included. Infants are mainly born between January 2011 and October 2022. Methods: Patient data were extracted from electronic health records (EHRs), with sufficient time-dependent oxygen data. Clinical outcomes for ROP were recorded as none/mild or moderate/severe (defined as type II or worse). Mixed-effects linear models were used to compare the 2 groups in terms of dynamic oxygen variables, such as daily average and the coefficient of variation (COV) fraction of inspired oxygen (FiO2). Support vector machine (SVM) and long-short-term memory (LSTM)-based multimodal models were trained with fivefold cross-validation to predict which infants would develop moderate/severe ROP. Gestational age (GA), birth weight, and time-dependent oxygen variables were used to develop predictive models. Main Outcome Measures: Model cross-validation performance was evaluated by computing the mean area under the receiver operating characteristic (AUROC) curve, precision, recall, and F1 score. Results: We found that both daily average and COV of FiO2 were associated with more severe ROP (adjusted P < 0.001). With fivefold cross-validation, the multimodal LSTM models had higher performance than the best static models (SVM using GA and 3 average FiO2 features) and SVM models trained on GA alone (mean AUROC = 0.89 ± 0.04 vs. 0.86 ± 0.05 vs. 0.83 ± 0.04). Conclusions: The development of severe ROP might not only be influenced by oxygen exposure but also by its fluctuation, which provides direction for future study of pathophysiological factors associated with severe ROP development. Additionally, we demonstrated that multimodal neural networks can be a method to extract useful information from time-series data, which may be a valuable methodology for the investigation of other diseases using EHR data. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.