RESUMEN
Fringe projection profilometry (FPP) is the most commonly used structured light approach for 3D object profiling. Traditional FPP algorithms have multistage procedures that can lead to error propagation. Deep-learning-based end-to-end models currently have been developed to mitigate this error propagation and provide faithful reconstruction. In this paper, we propose LiteF2DNet, a lightweight deep-learning framework to estimate the depth profile of objects, given reference and deformed fringes. The proposed framework has dense connections in the feature extraction module to aid better information flow. The parameters in the framework are 40% less than those in the base model, which also means less inference time and limited memory requirements, making it suitable for real-time 3D reconstruction. To circumvent the tedious process of collecting real samples, synthetic sample training was adopted in this work using Gaussian mixture models and computer-aided design objects. The qualitative and quantitative results presented in this work demonstrate that the proposed network performs well compared to other standard methods in the literature. Various analysis plots also illustrate the model's superior performance at high dynamic ranges, even with low-frequency fringes and high noise. Moreover, the reconstruction results on real samples show that the proposed model can predict 3D profiles of real objects with synthetic sample training.
RESUMEN
In the imbalance data scenarios, Deep Neural Networks (DNNs) fail to generalize well on minority classes. In this letter, we propose a simple and effective learning function i.e, Visually Interpretable Space Adjustment Learning (VISAL) to handle the imbalanced data classification task. VISAL's objective is to create more room for the generalization of minority class samples by bringing in both the angular and euclidean margins into the cross-entropy learning strategy. When evaluated on the imbalanced versions of CIFAR, Tiny ImageNet, COVIDx and IMDB reviews datasets, our proposed method outperforms the state of the art works by a significant margin.
Asunto(s)
Algoritmos , Redes Neurales de la Computación , Aprendizaje Automático , Aprendizaje , Generalización PsicológicaRESUMEN
BACKGROUND: We retrospectively data-mined the case records of Reverse Transcription Polymerase Chain Reaction (RT-PCR) confirmed COVID-19 patients hospitalized to a tertiary care centre to derive mortality predictors and formulate a risk score, for prioritizing admission. METHODS AND FINDINGS: Data on clinical manifestations, comorbidities, vital signs, and basic lab investigations collected as part of routine medical management at admission to a COVID-19 tertiary care centre in Chengalpattu, South India between May and November 2020 were retrospectively analysed to ascertain predictors of mortality in the univariate analysis using their relative difference in distribution among 'survivors' and 'non-survivors'. The regression coefficients of those factors remaining significant in the multivariable logistic regression were utilised for risk score formulation and validated in 1000 bootstrap datasets. Among 746 COVID-19 patients hospitalised [487 "survivors" and 259 "non-survivors" (deaths)], there was a slight male predilection [62.5%, (466/746)], with a higher mortality rate observed among 40-70 years age group [59.1%, (441/746)] and highest among diabetic patients with elevated urea levels [65.4% (68/104)]. The adjusted odds ratios of factors [OR (95% CI)] significant in the multivariable logistic regression were SaO2<95%; 2.96 (1.71-5.18), Urea ≥50 mg/dl: 4.51 (2.59-7.97), Neutrophil-lymphocytic ratio (NLR) >3; 3.01 (1.61-5.83), Age ≥50 years;2.52 (1.45-4.43), Pulse Rate ≥100/min: 2.02 (1.19-3.47) and coexisting Diabetes Mellitus; 1.73 (1.02-2.95) with hypertension and gender not retaining their significance. The individual risk scores for SaO2<95-11, Urea ≥50 mg/dl-15, NLR >3-11, Age ≥50 years-9, Pulse Rate ≥100/min-7 and coexisting diabetes mellitus-6, acronymed collectively as 'OUR-ARDs score' showed that the sum of scores ≥ 25 predicted mortality with a sensitivity-90%, specificity-64% and AUC of 0.85. CONCLUSIONS: The 'OUR ARDs' risk score, derived from easily assessable factors predicting mortality, offered a tangible solution for prioritizing admission to COVID-19 tertiary care centre, that enhanced patient care but without unduly straining the health system.
Asunto(s)
COVID-19/epidemiología , COVID-19/mortalidad , Mortalidad Hospitalaria , Hospitalización , SARS-CoV-2/genética , Atención Terciaria de Salud/métodos , Adulto , Anciano , COVID-19/virología , Comorbilidad , Femenino , Humanos , India/epidemiología , Masculino , Persona de Mediana Edad , Pronóstico , Estudios Retrospectivos , Factores de Riesgo , Índice de Severidad de la Enfermedad , Centros de Atención TerciariaRESUMEN
Fringe projection profilometry (FPP) is a widely used non-contact optical method for 3D profiling of objects. The commonly used linear fringe pattern in FPP has periodic intensity variations along the lateral direction. As a result, the linear fringe pattern used in FPP cannot uniquely represent the lateral shift induced by the objects having surface discontinuities. Thus, unambiguous surface profiling of objects, especially with surface discontinuities, using a single linear fringe image having a single fringe frequency, is unfeasible. This paper proposes using a radially symmetric circular fringe pattern as the structured light pattern for accurate unambiguous surface profiling of sudden height-discontinuous objects. To the best of our knowledge, this is the only method that can reconstruct discontinuous height profiles with the help of a single fringe image having a single frequency. The performance of the proposed algorithm is evaluated on several synthetic and real objects having smooth variations and discontinuities. Compared to the well-known fringe projection methods, the results depict that for a tolerable range of error, the proposed method can be applied for the reconstruction of objects with 4 times higher dynamic range and even at much lower fringe frequencies.
RESUMEN
BACKGROUND: Through the Health Information Technology for Economic and Clinical Health Act of 2009, the federal government invested $26 billion in electronic health records (EHRs) to improve physician performance and patient safety; however, these systems have not met expectations. One of the cited issues with EHRs is the human-computer interaction, as exhibited by the excessive number of interactions with the interface, which reduces clinician efficiency. In contrast, real-time location systems (RTLS)-technologies that can track the location of people and objects-have been shown to increase clinician efficiency. RTLS can improve patient flow in part through the optimization of patient verification activities. However, the data collected by RTLS have not been effectively applied to optimize interaction with EHR systems. OBJECTIVES: We conducted a pilot study with the intention of improving the human-computer interaction of EHR systems by incorporating a RTLS. The aim of this study is to determine the impact of RTLS on process metrics (i.e., provider time, number of rooms searched to find a patient, and the number of interactions with the computer interface), and the outcome metric of patient identification accuracy METHODS: A pilot study was conducted in a simulated emergency department using a locally developed camera-based RTLS-equipped EHR that detected the proximity of subjects to simulated patients and displayed patient information when subjects entered the exam rooms. Ten volunteers participated in 10 patient encounters with the RTLS activated (RTLS-A) and then deactivated (RTLS-D). Each volunteer was monitored and actions recorded by trained observers. We sought a 50% improvement in time to locate patients, number of rooms searched to locate patients, and the number of mouse clicks necessary to perform those tasks. RESULTS: The time required to locate patients (RTLS-A = 11.9 ± 2.0 seconds vs. RTLS-D = 36.0 ± 5.7 seconds, p < 0.001), rooms searched to find patient (RTLS-A = 1.0 ± 1.06 vs. RTLS-D = 3.8 ± 0.5, p < 0.001), and number of clicks to access patient data (RTLS-A = 1.0 ± 0.06 vs. RTLS-D = 4.1 ± 0.13, p < 0.001) were significantly reduced with RTLS-A relative to RTLS-D. There was no significant difference between RTLS-A and RTLS-D for patient identification accuracy. CONCLUSION: This pilot demonstrated in simulation that an EHR equipped with real-time location services improved performance in locating patients and reduced error compared with an EHR without RTLS. Furthermore, RTLS decreased the number of mouse clicks required to access information. This study suggests EHRs equipped with real-time location services that automates patient location and other repetitive tasks may improve physician efficiency, and ultimately, patient safety.