1075. The MET Project: Distinguishing Multisystem Inflammatory Syndrome in Children from Typhus Using Artificial Intelligence

Background

Multisystem inflammatory syndrome in children (MIS-C) following SARS-CoV-2 infection shares features with other inflammatory states, notably Kawasaki Disease. The rickettsial infection murine typhus is also in the differential for MIS-C in endemic areas. As the therapeutic approaches differ, it is essential to distinguish these disorders soon after presentation, well before confirmatory serologic testing results. Our objective was to develop an algorithm to accurately predict MIS-C versus typhus.

Methods

Retrospective review extracted demographic, clinical, and laboratory features available within 6 hours of presentation for 133 MIS-C and 87 typhus patients. 33 features were broken into 44 inputs and passed through an attention module to compute importance. Inputs were then entered into machine learning algorithms as MIS-C or typhus. Patients were divided into training and test cohorts respecting proportions in the dataset. An equation was built to calculate the “MET” (MIS-C versus endemic typhus) score.

Results

MIS-C patients were younger (8.4 v 11.2 years, p< 0.0001) and the majority (71%) presented on day 4-6 of fever; most typhus patients (84%) presented with ≥6 days (mean 4.9 v 7.3 days, p< 0.0001). Typhus patients were more likely to have rash (86% v 51%, p< 0.0001) and MIS-C patients red eyes (71% v 36%, p< 0.0001), other features were similar. MIS-C patients had higher C-reactive protein levels (17.7 v 9.8 mg/dL), procalcitonin (14.0 v 0.48 ng/mL), fibrinogen (558 v 394 mg/dL) and neutrophil-to-lymphocyte ratio (12 v 3.5), all p< 0.0001, other parameters were similar. MIS-C patients were also more likely to have elevated troponin (0.48 v 0.01 ng/mL, p< 0.0001) and require intensive care (66% v 6%, p< 0.0001). A long short term memory network outperformed 6 other models (99% accuracy using all 33 elements). The MET score predicted MIS-C versus typhus with 90% accuracy using only 10 features (sensitivity 90%, specificity 90%).

Conclusion

The clinical and laboratory similarities between typhus and MIS-C present challenges, but they can be reliably distinguished using artificial intelligence with as little as 10 features. Our ongoing interprofessional collaboration aims to make the MET score readily available to clinicians for use in patient encounters.

Disclosures

Flor M. Munoz, MD, MSc, Gilead: Grant/Research Support|Moderna: DSMB|Pfizer: DSMB Tiphanie Vogel, MD, PhD, Moderna: Advisor/Consultant|Novartis: Advisor/Consultant|Pfizer: Advisor/Consultant.

Competencies for the Use of Artificial Intelligence in Primary Care

The artificial intelligence (AI) revolution has arrived for the health care sector and is finally penetrating the far-reaching but perpetually underfinanced primary care platform. While AI has the potential to facilitate the achievement of the Quintuple Aim (better patient outcomes, population health, and health equity at lower costs while preserving clinician well-being), inattention to primary care training in the use of AI-based tools risks the opposite effects, imposing harm and exacerbating inequalities. The impact of AI-based tools on these aims will depend heavily on the decisions and skills of primary care clinicians; therefore, appropriate medical education and training will be crucial to maximize potential benefits and minimize harms. To facilitate this training, we propose 6 domains of competency for the effective deployment of AI-based tools in primary care: (1) foundational knowledge (what is this tool?), (2) critical appraisal (should I use this tool?), (3) medical decision making (when should I use this tool?), (4) technical use (how do I use this tool?), (5) patient communication (how should I communicate with patients regarding the use of this tool?), and (6) awareness of unintended consequences (what are the "side effects" of this tool?). Integrating these competencies will not be straightforward because of the breadth of knowledge already incorporated into family medicine training and the constantly changing technological landscape. Nonetheless, even incremental increases in AI-relevant training may be beneficial, and the sooner these challenges are tackled, the sooner the primary care workforce and those served by it will begin to reap the benefits.

Population stratification enables modeling effects of reopening policies on mortality and hospitalization rates

OBJECTIVE

Study the impact of local policies on near-future hospitalization and mortality rates.

MATERIALS AND METHODS

We introduce a novel risk-stratified SIR-HCD model that introduces new variables to model the dynamics of low-contact (e.g., work from home) and high-contact (e.g., work on-site) subpopulations while sharing parameters to control their respective R₀(t) over time. We test our model on data of daily reported hospitalizations and cumulative mortality of COVID-19 in Harris County, Texas, from May 1, 2020, until October 4, 2020, collected from multiple sources (USA FACTS, U.S. Bureau of Labor Statistics, Southeast Texas Regional Advisory Council COVID-19 report, TMC daily news, and Johns Hopkins University county-level mortality reporting).

RESULTS

We evaluated our model's forecasting accuracy in Harris County, TX (the most populated county in the Greater Houston area) during Phase-I and Phase-II reopening. Not only does our model outperform other competing models, but it also supports counterfactual analysis to simulate the impact of future policies in a local setting, which is unique among existing approaches.

DISCUSSION

Mortality and hospitalization rates are significantly impacted by local quarantine and reopening policies. Existing models do not directly account for the effect of these policies on infection, hospitalization, and death rates in an explicit and explainable manner. Our work is an attempt to improve prediction of these trends by incorporating this information into the model, thus supporting decision-making.

CONCLUSION

Our work is a timely effort to attempt to model the dynamics of pandemics under the influence of local policies.

A Layered Mounting Method for Extended Time-Lapse Confocal Microscopy of Whole Zebrafish Embryos

Dynamics of development can be followed by confocal time-lapse microscopy of live transgenic zebrafish embryos expressing fluorescence in specific tissues or cells. A difficulty with imaging whole embryo development is that zebrafish embryos grow substantially in length. When mounted as regularly done in 0.3-1% low melt agarose, the agarose imposes growth restriction, leading to distortions in the soft embryo body. Yet, to perform confocal time-lapse microscopy, the embryo must be immobilized. This article describes a layered mounting method for zebrafish embryos that restrict the motility of the embryos while allowing for the unrestricted growth. The mounting is performed in layers of agarose at different concentrations. To demonstrate the usability of this method, whole embryo vascular, neuronal and muscle development was imaged in transgenic fish for 55 consecutive hours. This mounting method can be used for easy, low-cost imaging of whole zebrafish embryos using inverted microscopes without requirements of molds or special equipment.

Automated, foot-bone registration using subdivision-embedded atlases for spatial mapping of bone mineral density

We present an atlas-based registration method for bones segmented from quantitative computed tomography (QCT) scans, with the goal of mapping their interior bone mineral densities (BMDs) volumetrically. We introduce a new type of deformable atlas, called subdivision-embedded atlas, which consists of a control grid represented as a tetrahedral subdivision mesh and a template bone surface embedded within the grid. Compared to a typical lattice-based deformation grid, the subdivision control grid possesses a relatively small degree of freedom tailored to the shape of the bone, which allows efficient fitting onto subjects. Compared with previous subdivision atlases, the novelty of our atlas lies in the addition of the embedded template surface, which further increases the accuracy of the fitting. Using this new atlas representation, we developed an efficient and fully automated pipeline for registering atlases of 12 tarsal and metatarsal bones to a segmented QCT scan of a human foot. Our evaluation shows that the mapping of BMD enabled by the registration is consistent for bones in repeated scans, and the regional BMD automatically computed from the mapping is not significantly different from expert annotations. The results suggest that our improved subdivision-based registration method is a reliable, efficient way to replace manual labor for measuring regional BMD in foot bones in QCT scans.

Frames-Based Denoising in 3D Confocal Microscopy Imaging

In this paper, we propose a novel denoising method for 3D confocal microscopy data based on robust edge detection. Our approach relies on the construction of a non-separable frame system in 3D that incorporates the Sobel operator in dual spatial directions. This multidirectional set of digital filters is capable of robustly detecting edge information by ensemble thresholding of the filtered data. We demonstrate the application of our method to both synthetic and real confocal microscopy data by comparing it to denoising methods based on separable 3D wavelets and 3D median filtering, and report very encouraging results.

Performance Evaluation of Abdominal Fat Burden Quantification in CT

Abdominal fat accumulation is an important cardiovascular risk ovascular risk factor. In clinical practice, delineation of subcutaneous and visceral fat is performed manually by an expert. This procedure is labor intensive, time consuming, and subject to inter-and intra-observer variability. In this paper, we present an extension of our previous work on automatic fat burden quantification and classification. Our improved method automatically differentiates abdominal fat into subcutaneous and visceral fat components and removes equipment-related artifacts. We evaluated the performance of our method using data from 40 subjects with very encouraging results.

Computer-aided non-contrast CT-based quantification of pericardial and thoracic fat and their associations with coronary calcium and Metabolic Syndrome

INTRODUCTION

Pericardial fat is emerging as an important parameter for cardiovascular risk stratification. We extended previously developed quantitation of thoracic fat volume (TFV) from non-contrast coronary calcium (CC) CT scans to also quantify pericardial fat volume (PFV) and investigated the associations of PFV and TFV with CC and the Metabolic Syndrome (METS).

METHODS

TFV is quantified automatically from user-defined range of CT slices covering the heart. Pericardial fat contours are generated by spline interpolation between 5-7 control points, placed manually on the pericardium within this cardiac range. Contiguous fat voxels within the pericardium are identified as pericardial fat. PFV and TFV were measured from non-contrast CT for 201 patients. In 105 patients, abdominal visceral fat area (VFA) was measured from an additional single-slice CT. In 26 patients, images were quantified by two readers to establish inter-observer variability. TFV and PFV were examined in relation to Body Mass Index (BMI), waist circumference and VFA, standard coronary risk factors (RF), CC (Agatston score >0) and METS.

RESULTS

PFV and TFV showed excellent correlation with VFA (R=0.79, R=0.89, p<0.0001), and moderate correlation with BMI (R=0.49, R=0.48, p<0.0001). In 26 scans, the inter-observer variability was greater for PFV (8.0+/-5.3%) than for TFV (4.4+/-3.9%, p=0.001). PFV and TFV, but not RF, were associated with CC [PFV: p=0.04, Odds Ratio 3.1; TFV: p<0.001, OR 7.9]. PFV and TFV were also associated with METS [PFV: p<0.001, OR 6.1; TFV p<0.001, OR 5.7], unlike CC [OR=1.0 p=NS] or RF. PFV correlated with low-HDL and high-glucose; TFV correlated with low-HDL, low-adiponectin, and high glucose and triglyceride levels.

CONCLUSIONS

PFV and TFV can be obtained easily and reproducibly from routine CC scoring scans, and may be important for risk stratification and monitoring.

Mortality incidence and the severity of coronary atherosclerosis assessed by computed tomography angiography

OBJECTIVES

This study investigated whether cardiac computed tomography angiography (CTA) can predict all-cause mortality in symptomatic patients.

BACKGROUND

Noninvasive coronary angiography is being increasingly performed by CTA to assess for obstructive coronary artery disease (CAD), and minimal outcome data exist for coronary CTA. We have utilized a cohort of symptomatic patients who underwent electron beam tomography to allow for longer follow-up (up to 12 years) than currently available with newer 64-slice multidetector-row computed tomography studies.

METHODS

In all, 2,538 consecutive patients who underwent CTA by electron beam tomography (age 59 +/- 14 years, 70% males) without known CAD were studied. Computed tomographic angiography results were categorized as significant CAD (> or =50% luminal narrowing), mild CAD (<50% stenosis), and normal coronary arteries. Multivariable Cox proportional hazards models were developed to predict all-cause mortality. Risk-adjusted models incorporated traditional risk factors for coronary disease and coronary artery calcification (CAC).

RESULTS

During a mean follow-up of 78 +/- 12 months, the death rate was 3.4% (86 deaths). The CTA-diagnosed CAD was an independent predictor of mortality in a multivariable model adjusted for age, gender, cardiac risk factors, and CAC (p < 0.0001). The addition of CAC to CTA-diagnosed CAD increased the concordance index significantly (0.69 for risk factors, 0.83 for the CTA-diagnosed CAD, and 0.89 for the addition of CAC to CAD, p < 0.0001). Risk-adjusted hazard ratios for CTA-diagnosed CAD were 1.7-, 1.8-, 2.3-, and 2.6-fold for 3-vessel nonobstructive, 1-vessel obstructive, 2-vessel obstructive, and 3-vessel obstructive CAD, respectively (p < 0.0001), when compared with the group who did not have CAD.

CONCLUSIONS

The primary results of our study reveal that the burden of angiographic disease detected by CTA provides both independent and incremental value in predicting all-cause mortality in symptomatic patients independent of age, gender, conventional risk factors, and CAC.

In-vivo imaging of carotid plaque neoangiogenesis with contrast-enhanced harmonic ultrasound

We describe a case where a mild carotid atherosclerotic plaque was assessed by contrast enhanced harmonic ultrasonography and image analysis. Quantitative indices of plaque echogenicity were determined prior and after the injection of microbubbles. Changes in plaque echogenicity were detected possibly due to the flow of microbubbles through vasa vasorum within the plaque and at the plaque base at the adventitial level. Future histological studies remain to be done to link the presence and the extent of plaque and adventitial neovascularization with the visual and quantitative findings derived by contrast enhanced harmonic ultrasound and image analysis.

Automated characterization of gene expression patterns with an atlas of the mouse brain

A spatio-temporal map of gene activity in the brain would be an important contribution to the understanding of brain development, disease, and function. Such a resource is now possible using high-throughput in situ hybridization, a method for transcriptome-wide acquisition of cellular resolution gene expression patterns in serial tissue sections. However, querying an enormous quantity of image data requires computational methods for describing and organizing gene expression patterns in a consistent manner. In addressing this, we have developed procedures for automated annotation of gene expression patterns in the postnatal mouse brain.

Towards automatic reconstruction of dendrite morphology from live neurons

The recent advent of optical imaging methods to capture both structural and functional data from living neurons holds the promise of improving our understanding of neuronal function. An important technical advance would be to use computer simulations to determine the optimal locations for high-speed functional imaging and electrical recording on an individual neuron. However, such simulations require a precise reconstruction of the dendritic morphology. Thus, the currently available time-consuming, semi-manual methods for morphological reconstruction prevent integration of cell-specific simulations into the experiments. Our work focuses on implementing a fast, robust method for extracting dendrite morphology from confocal or two-photon images to enable concurrent simulation and functional imaging of individual neurons. We discuss the current implementation of the morphological reconstruction, our approach to validation, and our initial results.

3D volume reconstruction of a mouse brain from histological sections using warp filtering

Sectioning tissues for optical microscopy often introduces upon the resulting sections distortions that make 3D reconstruction difficult. Here we present an automatic method for producing a smooth 3D volume from distorted 2D sections in the absence of any undistorted references. The method is based on pairwise elastic image warps between successive tissue sections, which can be computed by 2D image registration. Using a Gaussian filter, an average warp is computed for each section from the pairwise warps in a group of its neighboring sections. The average warps deform each section to match its neighboring sections, thus creating a smooth volume where corresponding features on successive sections lie close to each other. The proposed method can be used with any existing 2D image registration method for 3D reconstruction. In particular, we present a novel image warping algorithm based on dynamic programming that extends Dynamic Time Warping in 1D speech recognition to compute pairwise warps between high-resolution 2D images. The warping algorithm efficiently computes a restricted class of 2D local deformations that are characteristic between successive tissue sections. Finally, a validation framework is proposed and applied to evaluate the quality of reconstruction using both real sections and a synthetic volume.