RESUMO
BACKGROUND: High-sensitivity severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) antigen assays are desirable to mitigate false negative results. Limited data are available to quantify and track SARS-CoV-2 antigen burden in respiratory samples from different populations. METHODS: We developed the Microbubbling SARS-CoV-2 Antigen Assay (MSAA) with smartphone readout, with a limit of detection of 0.5 pg/mL (10.6 fmol/L) nucleocapsid antigen or 4000 copies/mL inactivated SARS-CoV-2 virus in nasopharyngeal (NP) swabs. We developed a computer vision and machine learning-based automatic microbubble image classifier to accurately identify positives and negatives and quantified and tracked antigen dynamics in intensive care unit coronavirus disease 2019 (COVID-19) inpatients and immunocompromised COVID-19 patients. RESULTS: Compared to qualitative reverse transcription-polymerase chain reaction methods, the MSAA demonstrated a positive percentage agreement of 97% (95% CI 92%-99%) and a negative percentage agreement of 97% (95% CI 94%-100%) in a clinical validation study with 372 residual clinical NP swabs. In immunocompetent individuals, the antigen positivity rate in swabs decreased as days-after-symptom-onset increased, despite persistent nucleic acid positivity. Antigen was detected for longer and variable periods of time in immunocompromised patients with hematologic malignancies. Total microbubble volume, a quantitative marker of antigen burden, correlated inversely with cycle threshold values and days-after-symptom-onset. Viral sequence variations were detected in patients with long duration of high antigen burden. CONCLUSIONS: The MSAA enables sensitive and specific detection of acute infections and quantification and tracking of antigen burden and may serve as a screening method in longitudinal studies to identify patients who are likely experiencing active rounds of ongoing replication and warrant close viral sequence monitoring.
Assuntos
Antígenos Virais/análise , Teste para COVID-19/métodos , COVID-19 , Smartphone , COVID-19/diagnóstico , Humanos , Aprendizado de Máquina , SARS-CoV-2 , Sensibilidade e EspecificidadeRESUMO
Background: Little is known about the dynamics of SARS-CoV-2 antigen burden in respiratory samples in different patient populations at different stages of infection. Current rapid antigen tests cannot quantitate and track antigen dynamics with high sensitivity and specificity in respiratory samples. Methods: We developed and validated an ultra-sensitive SARS-CoV-2 antigen assay with smartphone readout using the Microbubbling Digital Assay previously developed by our group, which is a platform that enables highly sensitive detection and quantitation of protein biomarkers. A computer vision-based algorithm was developed for microbubble smartphone image recognition and quantitation. A machine learning-based classifier was developed to classify the smartphone images based on detected microbubbles. Using this assay, we tracked antigen dynamics in serial swab samples from COVID patients hospitalized in ICU and immunocompromised COVID patients. Results: The limit of detection (LOD) of the Microbubbling SARS-CoV-2 Antigen Assay was 0.5 pg/mL (10.6 fM) recombinant nucleocapsid (N) antigen or 4000 copies/mL inactivated SARS-CoV-2 virus in nasopharyngeal (NP) swabs, comparable to many rRT-PCR methods. The assay had high analytical specificity towards SARS-CoV-2. Compared to EUA-approved rRT-PCR methods, the Microbubbling Antigen Assay demonstrated a positive percent agreement (PPA) of 97% (95% confidence interval (CI), 92-99%) in symptomatic individuals within 7 days of symptom onset and positive SARS-CoV-2 nucleic acid results, and a negative percent agreement (NPA) of 97% (95% CI, 94-100%) in symptomatic and asymptomatic individuals with negative nucleic acid results. Antigen positivity rate in NP swabs gradually decreased as days-after-symptom-onset increased, despite persistent nucleic acid positivity of the same samples. The computer vision and machine learning-based automatic microbubble image classifier could accurately identify positives and negatives, based on microbubble counts and sizes. Total microbubble volume, a potential marker of antigen burden, correlated inversely with Ct values and days-after-symptom-onset. Antigen was detected for longer periods of time in immunocompromised patients with hematologic malignancies, compared to immunocompetent individuals. Simultaneous detectable antigens and nucleic acids may indicate the presence of replicating viruses in patients with persistent infections. Conclusions: The Microbubbling SARS-CoV-2 Antigen Assay enables sensitive and specific detection of acute infections, and quantitation and tracking of antigen dynamics in different patient populations at various stages of infection. With smartphone compatibility and automated image processing, the assay is well-positioned to be adapted for point-of-care diagnosis and to explore the clinical implications of antigen dynamics in future studies.
RESUMO
Visual recognition systems mounted on autonomous moving agents face the challenge of unconstrained data, but simultaneously have the opportunity to improve their performance by moving to acquire new views at test time. In this work, we first show how a recurrent neural network-based system may be trained to perform end-to-end learning of motion policies suited for this "active recognition" setting. Further, we hypothesize that active vision requires an agent to have the capacity to reason about the effects of its motions on its view of the world. To verify this hypothesis, we attempt to induce this capacity in our active recognition pipeline, by simultaneously learning to forecast the effects of the agent's motions on its internal representation of the environment conditional on all past views. Results across three challenging datasets confirm both that our end-to-end system successfully learns meaningful policies for active category recognition, and that "learning to look ahead" further boosts recognition performance.
RESUMO
Standard computer vision systems assume access to intelligently captured inputs (e.g., photos from a human photographer), yet autonomously capturing good observations is a major challenge in itself. We address the problem of learning to look around: How can an agent learn to acquire informative visual observations? We propose a reinforcement learning solution, where the agent is rewarded for reducing its uncertainty about the unobserved portions of its environment. Specifically, the agent is trained to select a short sequence of glimpses, after which it must infer the appearance of its full environment. To address the challenge of sparse rewards, we further introduce sidekick policy learning, which exploits the asymmetry in observability between training and test time. The proposed methods learned observation policies that not only performed the completion task for which they were trained but also generalized to exhibit useful "look-around" behavior for a range of active perception tasks.