Search results for: accuracy

Authors: M. A. C. S. Sampath Fernando, James M. Curran, Renate Meyer

Abstract:

Model assessment, in the Bayesian context, involves evaluation of the goodness-of-fit and the comparison of several alternative candidate models for predictive accuracy and improvements. In posterior predictive checks, the data simulated under the fitted model is compared with the actual data. Predictive model accuracy is estimated using information criteria such as the Akaike information criterion (AIC), the Bayesian information criterion (BIC), the Deviance information criterion (DIC), and the Watanabe-Akaike information criterion (WAIC). The goal of an information criterion is to obtain an unbiased measure of out-of-sample prediction error. Since posterior checks use the data twice; once for model estimation and once for testing, a bias correction which penalises the model complexity is incorporated in these criteria. Cross-validation (CV) is another method used for examining out-of-sample prediction accuracy. Leave-one-out cross-validation (LOO-CV) is the most computationally expensive variant among the other CV methods, as it fits as many models as the number of observations. Importance sampling (IS), truncated importance sampling (TIS) and Pareto-smoothed importance sampling (PSIS) are generally used as approximations to the exact LOO-CV and utilise the existing MCMC results avoiding expensive computational issues. The reciprocals of the predictive densities calculated over posterior draws for each observation are treated as the raw importance weights. These are in turn used to calculate the approximate LOO-CV of the observation as a weighted average of posterior densities. In IS-LOO, the raw weights are directly used. In contrast, the larger weights are replaced by their modified truncated weights in calculating TIS-LOO and PSIS-LOO. Although, information criteria and LOO-CV are unable to reflect the goodness-of-fit in absolute sense, the differences can be used to measure the relative performance of the models of interest. However, the use of these measures is only valid under specific circumstances. This study has developed 11 models using normal, log-normal, gamma, and student’s t distributions to improve the PCR stutter prediction with forensic data. These models are comprised of four with profile-wide variances, four with locus specific variances, and three which are two-component mixture models. The mean stutter ratio in each model is modeled as a locus specific simple linear regression against a feature of the alleles under study known as the longest uninterrupted sequence (LUS). The use of AIC, BIC, DIC, and WAIC in model comparison has some practical limitations. Even though, IS-LOO, TIS-LOO, and PSIS-LOO are considered to be approximations of the exact LOO-CV, the study observed some drastic deviations in the results. However, there are some interesting relationships among the logarithms of pointwise predictive densities (lppd) calculated under WAIC and the LOO approximation methods. The estimated overall lppd is a relative measure that reflects the overall goodness-of-fit of the model. Parallel log-likelihood profiles for the models conditional on equal posterior variances in lppds were observed. This study illustrates the limitations of the information criteria in practical model comparison problems. In addition, the relationships among LOO-CV approximation methods and WAIC with their limitations are discussed. Finally, useful recommendations that may help in practical model comparisons with these methods are provided.

Keywords: cross-validation, importance sampling, information criteria, predictive accuracy

Procedia PDF Downloads 392

Authors: Rafael Alvarenga, Hubert Herbaux, Laurent Linguet

Abstract:

The uncertainty concerning the power production of intermittent renewable energy is one of the main barriers to the integration of such assets into the power grid. Efforts have thus been made to develop methods to quantify this uncertainty, allowing producers to ensure more reliable and profitable engagements related to their future power delivery. Even though a diversity of probabilistic approaches was proposed in the literature giving promising results, the added value of adopting such methods for scheduling intermittent power plants is still unclear. In this study, the profits obtained by a decision-making model used to optimally schedule an existing PV power plant connected to batteries are compared when the model is fed with deterministic and probabilistic forecasts generated with two of the most recent methods proposed in the literature. Moreover, deterministic forecasts with different accuracy levels were used in the experiments, testing the utility and the capability of probabilistic methods of modeling the progressively increasing uncertainty. Even though probabilistic approaches are unquestionably developed in the recent literature, the results obtained through a study case show that deterministic forecasts still provide the best performance if accurate, ensuring a gain of 14% on final profits compared to the average performance of probabilistic models conditioned to the same forecasts. When the accuracy of deterministic forecasts progressively decreases, probabilistic approaches start to become competitive options until they completely outperform deterministic forecasts when these are very inaccurate, generating 73% more profits in the case considered compared to the deterministic approach.

Keywords: PV power forecasting, uncertainty quantification, optimal scheduling, power systems

Procedia PDF Downloads 87

Authors: Davide Gotti, Milan Prodanovic, Sergio Pinilla, David Muñoz-Torrero

Abstract:

Since its proposal, the Doyle-Fuller-Newman (DFN) lithium-ion battery model has gained popularity in the electrochemical field. In fact, this model provides the user with theoretical support for designing the lithium-ion battery parameters, such as the material particle or the diffusion coefficient adjustment direction. However, the model is mathematically complex as it is composed of several partial differential equations (PDEs) such as Fick’s law of diffusion, the MacInnes and Ohm’s equations, among other phenomena. Thus, to efficiently use the model in a time-domain simulation environment, the selection of the discretization technique is of a pivotal importance. There are several numerical methods available in the literature that can be used to carry out this task. In this study, a comparison between the explicit Euler, Crank-Nicolson, and Chebyshev discretization methods is proposed. These three methods are compared in terms of accuracy, stability, and computational times. Firstly, the explicit Euler discretization technique is analyzed. This method is straightforward to implement and is computationally fast. In this work, the accuracy of the method and its stability properties are shown for the electrolyte diffusion partial differential equation. Subsequently, the Crank-Nicolson method is considered. It represents a combination of the implicit and explicit Euler methods that has the advantage of being of the second order in time and is intrinsically stable, thus overcoming the disadvantages of the simpler Euler explicit method. As shown in the full paper, the Crank-Nicolson method provides accurate results when applied to the DFN model. Its stability does not depend on the integration time step, thus it is feasible for both short- and long-term tests. This last remark is particularly important as this discretization technique would allow the user to implement parameter estimation and optimization techniques such as system or genetic parameter identification methods using this model. Finally, the Chebyshev discretization technique is implemented in the DFN model. This discretization method features swift convergence properties and, as other spectral methods used to solve differential equations, achieves the same accuracy with a smaller number of discretization nodes. However, as shown in the literature, these methods are not suitable for handling sharp gradients, which are common during the first instants of the charge and discharge phases of the battery. The numerical results obtained and presented in this study aim to provide the guidelines on how to select the adequate discretization technique for the DFN model according to the type of application to be performed, highlighting the pros and cons of the three methods. Specifically, the non-eligibility of the simple Euler method for longterm tests will be presented. Afterwards, the Crank-Nicolson and the Chebyshev discretization methods will be compared in terms of accuracy and computational times under a wide range of battery operating scenarios. These include both long-term simulations for aging tests, and short- and mid-term battery charge/discharge cycles, typically relevant in battery applications like grid primary frequency and inertia control and electrical vehicle breaking and acceleration.

Keywords: Doyle-Fuller-Newman battery model, partial differential equations, discretization, numerical methods

Procedia PDF Downloads 23

Authors: Peifei Zhu, Zisheng Li, Yasuki Kakishita, Mayumi Suzuki, Tomoaki Chono

Abstract:

Segmentation of left ventricle (LV) from cardiac ultrasound images provides a quantitative functional analysis of the heart to diagnose disease. Active Shape Model (ASM) is a widely used approach for LV segmentation but suffers from the drawback that initialization of the shape model is not sufficiently close to the target, especially when dealing with abnormal shapes in disease. In this work, a two-step framework is proposed to improve the accuracy and speed of the model-based segmentation. Firstly, a robust and efficient detector based on Hough forest is proposed to localize cardiac feature points, and such points are used to predict the initial fitting of the LV shape model. Secondly, to achieve more accurate and detailed segmentation, ASM is applied to further fit the LV shape model to the cardiac ultrasound image. The performance of the proposed method is evaluated on a dataset of 800 cardiac ultrasound images that are mostly of abnormal shapes. The proposed method is compared to several combinations of ASM and existing initialization methods. The experiment results demonstrate that the accuracy of feature point detection for initialization was improved by 40% compared to the existing methods. Moreover, the proposed method significantly reduces the number of necessary ASM fitting loops, thus speeding up the whole segmentation process. Therefore, the proposed method is able to achieve more accurate and efficient segmentation results and is applicable to unusual shapes of heart with cardiac diseases, such as left atrial enlargement.

Keywords: hough forest, active shape model, segmentation, cardiac left ventricle

Procedia PDF Downloads 339

Authors: M. Aghili, S. Tabarestani, C. Freytes, M. Shojaie, M. Cabrerizo, A. Barreto, N. Rishe, R. E. Curiel, D. Loewenstein, R. Duara, M. Adjouadi

Abstract:

A major challenge in medical studies, especially those that are longitudinal, is the problem of missing measurements which hinders the effective application of many machine learning algorithms. Furthermore, recent Alzheimer's Disease studies have focused on the delineation of Early Mild Cognitive Impairment (EMCI) and Late Mild Cognitive Impairment (LMCI) from cognitively normal controls (CN) which is essential for developing effective and early treatment methods. To address the aforementioned challenges, this paper explores the potential of using the eXtreme Gradient Boosting (XGBoost) algorithm in handling missing values in multiclass classification. We seek a generalized classification scheme where all prodromal stages of the disease are considered simultaneously in the classification and decision-making processes. Given the large number of subjects (1631) included in this study and in the presence of almost 28% missing values, we investigated the performance of XGBoost on the classification of the four classes of AD, NC, EMCI, and LMCI. Using 10-fold cross validation technique, XGBoost is shown to outperform other state-of-the-art classification algorithms by 3% in terms of accuracy and F-score. Our model achieved an accuracy of 80.52%, a precision of 80.62% and recall of 80.51%, supporting the more natural and promising multiclass classification.

Keywords: eXtreme gradient boosting, missing data, Alzheimer disease, early mild cognitive impairment, late mild cognitive impair, multiclass classification, ADNI, support vector machine, random forest

Procedia PDF Downloads 188

Authors: Areej Salaymeh, Loren Schwiebert, Stephen Remias, Jonathan Waddell

Abstract:

Extracting information from new data sources has emerged as a crucial task in many traffic planning processes, such as identifying traffic patterns, route planning, traffic forecasting, and locating infrastructure improvements. Given the advanced technologies used to collect Global Positioning System (GPS) data from dedicated GPS devices, GPS equipped phones, and navigation tools, intelligent data analysis methodologies are necessary to mine this raw data. In this research, an automatic detection framework is proposed to help identify and classify the locations of stopped GPS waypoints into two main categories: signalized intersections or highway congestion. The Delaunay triangulation is used to perform this assessment in the clustering phase. While most of the existing clustering algorithms need assumptions about the data distribution, the effectiveness of the Delaunay triangulation relies on triangulating geographical data points without such assumptions. Our proposed method starts by cleaning noise from the data and normalizing it. Next, the framework will identify stoppage points by calculating the traveled distance. The last step is to use clustering to form groups of waypoints for signalized traffic and highway congestion. Next, a binary classifier was applied to find distinguish highway congestion from signalized stop points. The binary classifier uses the length of the cluster to find congestion. The proposed framework shows high accuracy for identifying the stop positions and congestion points in around 99.2% of trials. We show that it is possible, using limited GPS data, to distinguish with high accuracy.

Keywords: Delaunay triangulation, clustering, intelligent transportation systems, GPS data

Procedia PDF Downloads 275

Authors: Christina Kirsch, Adam Hatzigiannis

Abstract:

Rail operators and maintainers worldwide are increasingly replacing walking patrols in the rail corridor with mechanized track patrols -essentially data capture on trains- and on-screen reviews of track infrastructure in centralized review facilities. The benefit is that infrastructure workers are less exposed to the dangers of the rail corridor. The impact is a significant change in work design from walking track sections and direct observation in the real world to sedentary jobs in the review facility reviewing captured data on screens. Defects in rail infrastructure can have catastrophic consequences. Reviewer performance regarding accuracy and efficiency of reviews within the available time frame is essential to ensure safety and operational performance. Rail operators must optimize workload and resource loading to transition to on-screen reviews successfully. Therefore, they need to know what workload assessment methodologies will provide reliable and valid data to optimize resourcing for on-screen reviews. This paper compares objective workload measures, including track difficulty ratings and review distance covered per hour, and subjective workload assessments (NASA TLX) and analyses the link between workload and reviewer performance, including sensitivity, precision, and overall accuracy. An experimental study was completed with eight on-screen reviewers, including infrastructure workers and engineers, reviewing track sections with different levels of track difficulty over nine days. Each day the reviewers completed four 90-minute sessions of on-screen inspection of the track infrastructure. Data regarding the speed of review (km/ hour), detected defects, false negatives, and false positives were collected. Additionally, all reviewers completed a subjective workload assessment (NASA TLX) after each 90-minute session and a short employee engagement survey at the end of the study period that captured impacts on job satisfaction and motivation. The results showed that objective measures for tracking difficulty align with subjective mental demand, temporal demand, effort, and frustration in the NASA TLX. Interestingly, review speed correlated with subjective assessments of physical and temporal demand, but to mental demand. Subjective performance ratings correlated with all accuracy measures and review speed. The results showed that subjective NASA TLX workload assessments accurately reflect objective workload. The analysis of the impact of workload on performance showed that subjective mental demand correlated with high precision -accurately detected defects, not false positives. Conversely, high temporal demand was negatively correlated with sensitivity and the percentage of detected existing defects. Review speed was significantly correlated with false negatives. With an increase in review speed, accuracy declined. On the other hand, review speed correlated with subjective performance assessments. Reviewers thought their performance was higher when they reviewed the track sections faster, despite the decline in accuracy. The study results were used to optimize resourcing and ensure that reviewers had enough time to review the allocated track sections to improve defect detection rates in accordance with the efficiency-thoroughness trade-off. Overall, the study showed the importance of a multi-method approach to workload assessment and optimization, combining subjective workload assessments with objective workload and performance measures to ensure that recommendations for work system optimization are evidence-based and reliable.

Keywords: automation, efficiency-thoroughness trade-off, human factors, job design, NASA TLX, performance optimization, subjective workload assessment, workload analysis

Procedia PDF Downloads 121

Authors: Helene Thieblemont, Fariborz Haghighat

Abstract:

Energy storage systems play a crucial role in decreasing building energy consumption during peak periods and expand the use of renewable energies in buildings. To provide a high building thermal performance, the energy storage system has to be properly controlled to insure a good energy performance while maintaining a satisfactory thermal comfort for building’s occupant. In the case of passive discharge storages, defining in advance the required amount of energy is required to avoid overheating in the building. Consequently, anticipatory supervisory control strategies have been developed forecasting future energy demand and production to coordinate systems. Anticipatory supervisory control strategies are based on some predictions, mainly of the weather forecast. However, if the forecasted hourly outdoor temperature may be found online with a high accuracy, solar radiations predictions are most of the time not available online. To estimate them, this paper proposes an advanced approach based on the forecast of weather conditions. Several methods to correlate hourly weather conditions forecast to real hourly solar radiations are compared. Results show that using weather conditions forecast allows estimating with an acceptable accuracy solar radiations of the next day. Moreover, this technique allows obtaining hourly data that may be used for building models. As a result, this solar radiation prediction model may help to implement model-based controller as Model Predictive Control.

Keywords: anticipatory control, model predictive control, solar radiation forecast, thermal storage

Procedia PDF Downloads 271

Authors: A. Taheri Tizro, Rojin Fasihi

Abstract:

Geoelectrical method is one of the most effective tools in determining subsurface lithological layers. The electromagnetic method is also a newer method that can play an important role in determining and separating subsurface layers with acceptable accuracy. In the present research, 10 electromagnetic soundings were collected in the upstream of 5 karstic springs of Famaseb, Faresban, Ghale Baroodab, Gian and Gonbad kabood in Nahavand plain of Hamadan province. By using the emerging data, the belectromagnetic logs were prepared at different depths and compared with 5 logs of the geoelectric method. The comparison showed that the value of NRMSE in the geoelectric method for the 5 springs of Famaseb, Faresban, Ghale Baroodab, Gian and Gonbad kabood were 7.11, 7.50, respectively. It is 44.93, 3.99, and 2.99, and in the electromagnetic method, the value of this coefficient for the investigated springs is about 1.4, 1.1, 1.2, 1.5, and 1.3, respectively. In addition to the similarity of the results of the two methods, it is found that, the accuracy of the electromagnetic method based on the NRMSE value is higher than the geoelectric method. The advantage of the electromagnetic method compared to geoelectric is on less time consuming and its cost prohibitive. The depth to water table is the final result of this research work , which showed that in the springs of Famaseb, Faresban, Ghale Baroodab, Gian and Gonbad kabood, having depth of about 6, 20, 10, 2 36 meters respectively. The maximum thickness of the aquifer layer was estimated in Gonbad kabood spring (36 meters) and the lowest in Gian spring (2 meters). These results can be used to identify the water potential of the region in order to better manage water resources.

Keywords: karst spring, geoelectric, aquifer layers, nahavand

Procedia PDF Downloads 70

Authors: Sara A. Ben Lashihar

Abstract:

The heritage building information modelling (HBIM) of the historical masonry buildings has expanded lately to meet the urgent needs for conservation and structural analysis. The masonry structures are unique features for ancient building architectures worldwide that have special cultural, spiritual, and historical significance. However, there is a research gap regarding the reliability of the HBIM modeling process of these structures. The HBIM modeling process of the masonry structures faces significant challenges due to the inherent complexity and uniqueness of their structural systems. Most of these processes are based on tracing the point clouds and rarely follow documents, archival records, or direct observation. The results of these techniques are highly abstracted models where the accuracy does not exceed LOD 200. The masonry assemblages, especially curved elements such as arches, vaults, and domes, are generally modeled with standard BIM components or in-place models, and the brick textures are graphically input. Hence, future investigation is necessary to establish a methodology to generate automatically parametric masonry components. These components are developed algorithmically according to mathematical and geometric accuracy and the validity of the survey data. The main aim of this paper is to provide a comprehensive review of the state of the art of the existing researches and papers that have been conducted on the HBIM modeling of the masonry structural elements and the latest approaches to achieve parametric models that have both the visual fidelity and high geometric accuracy. The paper reviewed more than 800 articles, proceedings papers, and book chapters focused on "HBIM and Masonry" keywords from 2017 to 2021. The studies were downloaded from well-known, trusted bibliographic databases such as Web of Science, Scopus, Dimensions, and Lens. As a starting point, a scientometric analysis was carried out using VOSViewer software. This software extracts the main keywords in these studies to retrieve the relevant works. It also calculates the strength of the relationships between these keywords. Subsequently, an in-depth qualitative review followed the studies with the highest frequency of occurrence and the strongest links with the topic, according to the VOSViewer's results. The qualitative review focused on the latest approaches and the future suggestions proposed in these researches. The findings of this paper can serve as a valuable reference for researchers, and BIM specialists, to make more accurate and reliable HBIM models for historic masonry buildings.

Keywords: HBIM, masonry, structure, modeling, automatic, approach, parametric

Procedia PDF Downloads 165

Authors: Mehrnoosh Omati, Mahmod Reza Sahebi

Abstract:

The information on land use/land cover changing plays an essential role for environmental assessment, planning and management in regional development. Remotely sensed imagery is widely used for providing information in many change detection applications. Polarimetric Synthetic aperture radar (PolSAR) image, with the discrimination capability between different scattering mechanisms, is a powerful tool for environmental monitoring applications. This paper proposes a new boundary-based segmentation algorithm as a fundamental step for land cover change detection. In this method, first, two PolSAR images are segmented using integration of marker-controlled watershed algorithm and coupled Markov random field (MRF). Then, object-based classification is performed to determine changed/no changed image objects. Compared with pixel-based support vector machine (SVM) classifier, this novel segmentation algorithm significantly reduces the speckle effect in PolSAR images and improves the accuracy of binary classification in object-based level. The experimental results on Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) polarimetric images show a 3% and 6% improvement in overall accuracy and kappa coefficient, respectively. Also, the proposed method can correctly distinguish homogeneous image parcels.

Keywords: coupled Markov random field (MRF), environment, object-based analysis, polarimetric SAR (PolSAR) images

Procedia PDF Downloads 218

Authors: Joel Ballesteros, Harriet Jane Caleja, Florian Del Mundo, Cherrie Pascual

Abstract:

Diffraction-based sensing was utilized in the quantification of human ferritin in blood serum to provide an alternative to label-based immunoassays currently used in clinical diagnostics and researches. The diffraction intensity was measured by the diffractive optics technology or dotLab™ system. Two methods were evaluated in this study: direct immunoassay and direct sandwich immunoassay. In the direct immunoassay, human ferritin was captured by human ferritin antibodies immobilized on an avidin-coated sensor while the direct sandwich immunoassay had an additional step for the binding of a detector human ferritin antibody on the analyte complex. Both methods were repeatable with coefficient of variation values below 15%. The direct sandwich immunoassay had a linear response from 10 to 500 ng/mL which is wider than the 100-500 ng/mL of the direct immunoassay. The direct sandwich immunoassay also has a higher calibration sensitivity with value 0.002 Diffractive Intensity (ng mL-1)-1) compared to the 0.004 Diffractive Intensity (ng mL-1)-1 of the direct immunoassay. The limit of detection and limit of quantification values of the direct immunoassay were found to be 29 ng/mL and 98 ng/mL, respectively, while the direct sandwich immunoassay has a limit of detection (LOD) of 2.5 ng/mL and a limit of quantification (LOQ) of 8.2 ng/mL. In terms of accuracy, the direct immunoassay had a percent recovery of 88.8-93.0% in PBS while the direct sandwich immunoassay had 94.1 to 97.2%. Based on the results, the direct sandwich immunoassay is a better diffraction-based immunoassay in terms of accuracy, LOD, LOQ, linear range, and sensitivity. The direct sandwich immunoassay was utilized in the determination of human ferritin in blood serum and the results are validated by Chemiluminescent Magnetic Immunoassay (CMIA). The calculated Pearson correlation coefficient was 0.995 and the p-values of the paired-sample t-test were less than 0.5 which show that the results of the direct sandwich immunoassay was comparable to that of CMIA and could be utilized as an alternative analytical method.

Keywords: biosensor, diffraction, ferritin, immunoassay

Procedia PDF Downloads 354

Authors: Ahmed ElAzab, Amira M. Idrees, Mahmoud A. Mahmoud, Hesham Hefny

Abstract:

Social networking sites such as Twitter and Facebook attracts over 500 million users across the world, for those users, their social life, even their practical life, has become interrelated. Their interaction with social networking has affected their life forever. Accordingly, social networking sites have become among the main channels that are responsible for vast dissemination of different kinds of information during real time events. This popularity in Social networking has led to different problems including the possibility of exposing incorrect information to their users through fake accounts which results to the spread of malicious content during life events. This situation can result to a huge damage in the real world to the society in general including citizens, business entities, and others. In this paper, we present a classification method for detecting fake accounts on Twitter. The study determines the minimized set of the main factors that influence the detection of the fake accounts on Twitter, then the determined factors have been applied using different classification techniques, a comparison of the results for these techniques has been performed and the most accurate algorithm is selected according to the accuracy of the results. The study has been compared with different recent research in the same area, this comparison has proved the accuracy of the proposed study. We claim that this study can be continuously applied on Twitter social network to automatically detect the fake accounts, moreover, the study can be applied on different Social network sites such as Facebook with minor changes according to the nature of the social network which are discussed in this paper.

Keywords: fake accounts detection, classification algorithms, twitter accounts analysis, features based techniques

Procedia PDF Downloads 416

Authors: Pongsakorn Komutpol, Pravej Serichetaphongse, Soontra Panmekiate, Atiphan Pimkhaokham

Abstract:

Statement of Problem: The important parameter of esthetic assessment in anterior maxillary implant include pink esthetic of gingiva and white esthetic of restoration. While the 3 dimensional (3D) implant position are recently concerned as a key for succeeding in implant treatment. However, to our knowledge, the authors did not come across any publication that demonstrated the relations of esthetic outcome and 3D implant position. Objectives: To investigate the correlation between positional accuracy of single-tooth implant restoration (STIR) in all 3 dimensions and their esthetic outcomes. Materials and Methods: 17 patients’ data who had a STIR at central incisor with pristine contralateral tooth were included in this study. Intraoral photographs, dental models, and cone beam computed tomography (CBCT) images were retrieved. The esthetic outcome was assessed in accordance with pink esthetic score and white esthetic score (PES/WES). While the number of correct position in each dimension (mesiodistal, labiolingual, apicocoronal) of the implant were evaluated and defined as 'right' or 'wrong' according to ITI consensus conference by one investigator using CBCT data. The different mean score between right and wrong position in all dimensions was analyzed by Mann-Whitney U test with 0.05 was the significant level of the study. Results: The average score of PES/WES was 15.88 ± 1.65 which was considered as clinically acceptable. The average PES/WES score in 1, 2 and 3 right dimension of the implant position were 16.71, 15.75 and 15.17 respectively. None of the implants placed wrongly in all three dimensions. Statistically significant difference of the PES/WES score was found between the implants that placed right in 3 dimensions and 1 dimension (p = 0.041). Conclusion: This study supported the principle of 3D position of implant. The more properly implant was placed, the higher esthetic outcome was found.

Keywords: accuracy, dental implant, esthetic, 3D implant position

Procedia PDF Downloads 179

Authors: Yen Chia-Ju, Cheng Ding-Ruei

Abstract:

This study took case office building as subject and explored the responsive work order repair request of facilities and equipment in offices and public areas by gray theory, with the purpose of providing for future related office building owners, executive managers, property management companies, mechanical and electrical companies as reference for deciding and assessing forecast model. Important conclusions of this study are summarized as follows according to the study findings: 1. Grey Relational Analysis discusses the importance of facilities repair number of six categories, namely, power systems, building systems, water systems, air conditioning systems, fire systems and manpower dispatch in order. In terms of facilities maintenance importance are power systems, building systems, water systems, air conditioning systems, manpower dispatch and fire systems in order. 2. GM (1,N) and regression method took maintenance hours as dependent variables and repair number, leased area and tenants number as independent variables and conducted single month forecast based on 12 data from January to December 2011. The mean absolute error and average accuracy of GM (1,N) from verification results were 6.41% and 93.59%; the mean absolute error and average accuracy of regression model were 4.66% and 95.34%, indicating that they have highly accurate forecast capability.

Keywords: rey theory, forecast model, Taipei 101, office buildings, property management, facilities, equipment

Procedia PDF Downloads 444

Authors: Seydou Sinde

Abstract:

The aims of this paper are to formulate mathematical expressions that can be used to estimate the standpipe pressure (SPP). The developed formulas take into account the main factors that, directly or indirectly, affect the behavior of SPP values. Fluid rheology and well hydraulics are some of these essential factors. Mud Plastic viscosity, yield point, flow power, consistency index, flow rate, drillstring, and annular geometries are represented by the frictional pressure (Pf), which is one of the input independent parameters and is calculated, in this paper, using Herschel-Bulkley rheological model. Other input independent parameters include the rate of penetration (ROP), applied load or weight on the bit (WOB), bit revolutions per minute (RPM), bit torque (TRQ), and hole inclination and direction coupled in the hole curvature or dogleg (DL). The technique of repeating parameters and Buckingham PI theorem are used to reduce the number of the input independent parameters into the dimensionless revolutions per minute (RPMd), the dimensionless torque (TRQd), and the dogleg, which is already in the dimensionless form of radians. Multivariable linear and polynomial regression technique using PTC Mathcad Prime 4.0 is used to analyze and determine the exact relationships between the dependent parameter, which is SPP, and the remaining three dimensionless groups. Three models proved sufficiently satisfactory to estimate the standpipe pressure: multivariable linear regression model 1 containing three regression coefficients for vertical wells; multivariable linear regression model 2 containing four regression coefficients for deviated wells; and multivariable polynomial quadratic regression model containing six regression coefficients for both vertical and deviated wells. Although that the linear regression model 2 (with four coefficients) is relatively more complex and contains an additional term over the linear regression model 1 (with three coefficients), the former did not really add significant improvements to the later except for some minor values. Thus, the effect of the hole curvature or dogleg is insignificant and can be omitted from the input independent parameters without significant losses of accuracy. The polynomial quadratic regression model is considered the most accurate model due to its relatively higher accuracy for most of the cases. Data of nine wells from the Middle East were used to run the developed models with satisfactory results provided by all of them, even if the multivariable polynomial quadratic regression model gave the best and most accurate results. Development of these models is useful not only to monitor and predict, with accuracy, the values of SPP but also to early control and check for the integrity of the well hydraulics as well as to take the corrective actions should any unexpected problems appear, such as pipe washouts, jet plugging, excessive mud losses, fluid gains, kicks, etc.

Keywords: standpipe, pressure, hydraulics, nondimensionalization, parameters, regression

Procedia PDF Downloads 84

Authors: Parthiban Srinivasan

Abstract:

With recent advances in computational power and access to enough data in biosciences, artificial intelligence methods are increasingly being used in drug discovery research. These methods are essentially a series of advanced statistics based exercises that review the past to indicate the likely future. Our goal is to develop a model that accurately predicts biological activity and toxicity parameters for novel compounds. We have compiled a robust library of over 150,000 chemical compounds with different pharmacological properties from literature and public domain databases. The compounds are stored in simplified molecular-input line-entry system (SMILES), a commonly used text encoding for organic molecules. We utilize an automated process to generate an array of numerical descriptors (features) for each molecule. Redundant and irrelevant descriptors are eliminated iteratively. Our prediction engine is based on a portfolio of machine learning algorithms. We found Random Forest algorithm to be a better choice for this analysis. We captured non-linear relationship in the data and formed a prediction model with reasonable accuracy by averaging across a large number of randomized decision trees. Our next step is to apply deep neural network (DNN) algorithm to predict the biological activity and toxicity properties. We expect the DNN algorithm to give better results and improve the accuracy of the prediction. This presentation will review all these prominent machine learning and deep learning methods, our implementation protocols and discuss these techniques for their usefulness in biomedical and health informatics.

Keywords: deep learning, drug discovery, health informatics, machine learning, toxicity prediction

Procedia PDF Downloads 356

Authors: Shang-Sin Shiu, Shu-I Chin, Hsiu-Ju Chen, Ru-Yu Lien

Abstract:

The Hospice and Palliative Care Act has undergone three revisions, making it less challenging for terminal patients to withdraw life support systems. However, the adequacy of care before withdraw is a crucial factor in end-of-life medical treatment. The author observed that intensive care unit (ICU) nursing staff often rely on simple flowcharts or word of mouth, leading to inadequate preparation and failure to meet patient needs before withdraw. This results in confusion or hesitation among those executing the process. Therefore, there is a motivation to improve the withdraw of patient care processes, establish standardized procedures, ensure the accuracy of removal execution, enhance end-of-life care self-efficacy for nursing staff, and improve the overall quality of care. The investigation identified key issues: the lack of applicable guidelines for ICU care for withdraw from life-sustaining, insufficient education and training on withdraw and end-of-life care, scattered locations of withdraw-related tools, and inadequate self-efficacy in withdraw from life-sustaining care. Solutions proposed include revising withdraw care processes and guidelines, integrating tools and locations, conducting educational courses, and forming support groups. After the project implementation, the accuracy of removal cognition improved from 78% to 96.5%, self-efficacy in end-of-life care after removal increased from 54.7% to 93.1%, and the correctness of care behavior progressed from 27.7% to 97.8%. It is recommended to regularly conduct courses on removing life support system care and grief consolation to enhance the quality of end-of-life care.

Keywords: the intensive care unit (ICU) patients, nursing staff, withdraw life support systems, self-efficacy

Procedia PDF Downloads 51

Authors: Mohd Ruzlin Mohd Mokhtar

Abstract:

At present, most operators who are working on projects for utilities such as power, water, oil, gas, telecommunication and sewerage are using technologies e.g. Total station, Global Positioning System (GPS), Electromagnetic Locator (EML) and Ground Penetrating Radar (GPR) to perform underground utility mapping. With the increase in popularity of Horizontal Directional Drilling (HDD) method among the local authorities and asset owners, most of newly installed underground utilities need to use the HDD method. HDD method is seen as simple and create not much disturbance to the public and traffic. Thus, it was the preferred utilities installation method in most of areas especially in urban areas. HDDs were installed much deeper than exiting utilities (some reports saying that HDD is averaging 5 meter in depth). However, this impacts the accuracy or ability of existing underground utility mapping technologies. In most of Malaysia underground soil condition, those technologies were limited to maximum of 3 meter depth. Thus, those utilities which were installed much deeper than 3 meter depth could not be detected by using existing detection tools. The accuracy and reliability of existing underground utility mapping technologies or work procedure were in doubt. Thus, a mitigation action plan is required. While installing new utility using Horizontal Directional Drilling (HDD) method, a more accurate underground utility mapping can be achieved by using Gyro OMU compared to existing practice using e.g. EML and GPR. Gyro OMU is a method to accurately identify the location of HDD thus this mapping can be used or referred to avoid those cost of breakdown due to future HDD works which can be caused by inaccurate underground utility mapping.

Keywords: Gyro Orientation Measurement Unit (Gyro OMU), Horizontal Directional Drilling (HDD), Ground Penetrating Radar (GPR), Electromagnetic Locator (EML)

Procedia PDF Downloads 140

Authors: Ping Li, Bing Feng, Junpu Zhao, Xudong Xie, Dangpeng Xu, Kuixing Zheng, Qihua Zhu, Xiaofeng Wei

Abstract:

In laser driven inertial confinement fusion (ICF), the control of the temporal shape of the laser pulse is a key point to ensure an optimal interaction of laser-target. One of the main difficulties in controlling the temporal shape is the foot part control accuracy of high contrast pulse. Based on the analysis of pulse perturbation in the process of amplification and frequency conversion in high power lasers, an approach of beam spatio-temporal multiplexing is proposed to improve the control precision of high contrast pulse. In the approach, the foot and peak part of high contrast pulse are controlled independently, which propagate separately in the near field, and combine together in the far field to form the required pulse shape. For high contrast pulse, the beam area ratio of the two parts is optimized, and then beam fluence and intensity of the foot part are increased, which brings great convenience to the control of pulse. Meanwhile, the near field distribution of the two parts is also carefully designed to make sure their F-numbers are the same, which is another important parameter for laser-target interaction. The integrated calculation results show that for a pulse with a contrast of up to 500, the deviation of foot part can be improved from 20% to 5% by using beam spatio-temporal multiplexing approach with beam area ratio of 1/20, which is almost the same as that of peak part. The research results are expected to bring a breakthrough in power balance of high power laser facility.

Keywords: inertial confinement fusion, laser pulse control, beam spatio-temporal multiplexing, power balance

Procedia PDF Downloads 147

Authors: Andrea Treviño Gavito, Diego Klabjan, Sanjiv J. Shah

Abstract:

Echocardiograms serve as pivotal resources for clinicians in diagnosing cardiac conditions, offering non-invasive insights into a heart’s structure and function. When echocardiographic studies are conducted, no standardized labeling of the acquired views is performed. Employing machine learning algorithms for automated echocardiogram view detection has emerged as a promising solution to enhance efficiency in echocardiogram use for diagnosis. However, existing approaches predominantly rely on supervised learning, necessitating labor-intensive expert labeling. In this paper, we introduce a fully unsupervised echocardiographic view detection framework that leverages convolutional autoencoders to obtain lower dimensional representations and the K-means algorithm for clustering them into view-related groups. Our approach focuses on discriminative patches from echocardiographic frames. Additionally, we propose a trainable inverse average layer to optimize decoding of average operations. By integrating both public and proprietary datasets, we obtain a marked improvement in model performance when compared to utilizing a proprietary dataset alone. Our experiments show boosts of 15.5% in accuracy and 9.0% in the F-1 score for frame-based clustering, and 25.9% in accuracy and 19.8% in the F-1 score for view-based clustering. Our research highlights the potential of unsupervised learning methodologies and the utilization of open-sourced data in addressing the complexities of echocardiogram interpretation, paving the way for more accurate and efficient cardiac diagnoses.

Keywords: artificial intelligence, echocardiographic view detection, echocardiography, machine learning, self-supervised representation learning, unsupervised learning

Procedia PDF Downloads 32

Authors: Min Jindong, Wang Mingxia

Abstract:

With the widespread use of artificial intelligence in life, computer vision, especially deep convolutional neural network models, has developed rapidly. With the increase of the complexity of the real visual target detection task and the improvement of the recognition accuracy, the target detection network model is also very large. The huge deep neural network model is not conducive to deployment on edge devices with limited resources, and the timeliness of network model inference is poor. In this paper, knowledge distillation is used to compress the huge and complex deep neural network model, and the knowledge contained in the complex network model is comprehensively transferred to another lightweight network model. Different from traditional knowledge distillation methods, we propose a novel knowledge distillation that incorporates multi-faceted features, called M-KD. In this paper, when training and optimizing the deep neural network model for target detection, the knowledge of the soft target output of the teacher network in knowledge distillation, the relationship between the layers of the teacher network and the feature attention map of the hidden layer of the teacher network are transferred to the student network as all knowledge. in the model. At the same time, we also introduce an intermediate transition layer, that is, an intermediate guidance layer, between the teacher network and the student network to make up for the huge difference between the teacher network and the student network. Finally, this paper adds an exploration module to the traditional knowledge distillation teacher-student network model. The student network model not only inherits the knowledge of the teacher network but also explores some new knowledge and characteristics. Comprehensive experiments in this paper using different distillation parameter configurations across multiple datasets and convolutional neural network models demonstrate that our proposed new network model achieves substantial improvements in speed and accuracy performance.

Keywords: object detection, knowledge distillation, convolutional network, model compression

Procedia PDF Downloads 278

Authors: Kenan Morani, Esra Kaya Ayana

Abstract:

This study aimed to develop a novel pipeline for COVID-19 detection using a large and rigorously annotated database of computed tomography (CT) images. The pipeline consists of UNet-based segmentation, lung extraction, and a classification part, with the addition of optional slice removal techniques following the segmentation part. In this work, a batch normalization was added to the original UNet model to produce lighter and better localization, which is then utilized to build a full pipeline for COVID-19 diagnosis. To evaluate the effectiveness of the proposed pipeline, various segmentation methods were compared in terms of their performance and complexity. The proposed segmentation method with batch normalization outperformed traditional methods and other alternatives, resulting in a higher dice score on a publicly available dataset. Moreover, at the slice level, the proposed pipeline demonstrated high validation accuracy, indicating the efficiency of predicting 2D slices. At the patient level, the full approach exhibited higher validation accuracy and macro F1 score compared to other alternatives, surpassing the baseline. The classification component of the proposed pipeline utilizes a convolutional neural network (CNN) to make final diagnosis decisions. The COV19-CT-DB dataset, which contains a large number of CT scans with various types of slices and rigorously annotated for COVID-19 detection, was utilized for classification. The proposed pipeline outperformed many other alternatives on the dataset.

Keywords: classification, computed tomography, lung extraction, macro F1 score, UNet segmentation

Procedia PDF Downloads 131

Authors: Yasar Samet Gokceoglu, Ayse Nur Incesu, Furkan Okatar, Berk Nimetoglu, Serkan Bayram, Turgut Akgul

Abstract:

Ankle syndesmosis injuries pose a significant challenge in orthopedic practice due to their potential for prolonged recovery and chronic ankle dysfunction. Accurate diagnosis and management of these injuries are essential for achieving optimal patient outcomes. The use of radiological methods, such as X-ray, computed tomography (CT), and magnetic resonance imaging (MRI), plays a vital role in the accurate diagnosis of syndesmosis injuries in the context of ankle fractures. Treatment options for ankle syndesmosis injuries vary, with surgical interventions such as screw fixation and suture-button implantation being commonly employed. The choice of treatment is influenced by the severity of the injury and the presence of associated fractures. Additionally, the mechanism of injury, such as pure syndesmosis injury or specific fracture types, can impact the stability and management of syndesmosis injuries. Ankle fractures with syndesmosis injury present a complex clinical scenario, requiring accurate diagnosis, appropriate reduction, and tailored management strategies. The interplay between the mechanism of injury, associated fractures, and treatment modalities significantly influences the outcomes of these challenging injuries. The long-term outcomes and patient satisfaction following ankle fractures with syndesmosis injury are crucial considerations in the field of orthopedics. Patient-reported outcome measures, such as the Foot and Ankle Outcome Score (FAOS), provide essential information about functional recovery and quality of life after these injuries. When diagnosing syndesmosis injuries, standard measurements, such as the medial clear space, tibiofibular overlap, tibiofibular clear space, anterior tibiofibular ratio (ATFR), and the anterior-posterior tibiofibular ratio (APTF), are assessed through radiographs and computed tomography (CT) scans. These parameters are critical in evaluating the presence and severity of syndesmosis injuries, enabling clinicians to choose the most appropriate treatment approach. Despite advancements in diagnostic imaging, challenges remain in accurately diagnosing and treating ankle syndesmosis injuries. Traditional diagnostic parameters, while beneficial, may not capture the full extent of the injury or provide sufficient information to guide therapeutic decisions. This gap highlights the need for exploring additional diagnostic parameters that could enhance the accuracy of syndesmosis injury diagnoses and inform treatment strategies more effectively. The primary goal of this research is to evaluate the usefulness of traditional radiographic measurements in comparison to new CT-based measurements for diagnosing ankle syndesmosis injuries. Specifically, this study aims to assess the accuracy of conventional parameters, including medial clear space, tibiofibular overlap, tibiofibular clear space, ATFR, and APTF, in contrast with the recently proposed CT-based measurements such as the delta and gamma angles. Moreover, the study intends to explore the relationship between these diagnostic parameters and functional outcomes, as measured by the Foot and Ankle Outcome Score (FAOS). Establishing a correlation between specific diagnostic measurements and FAOS scores will enable us to identify the most reliable predictors of functional recovery following syndesmosis injuries. This comparative analysis will provide valuable insights into the accuracy and dependability of CT-based measurements in diagnosing ankle syndesmosis injuries and their potential impact on predicting patient outcomes. The results of this study could greatly influence clinical practices by refining diagnostic criteria and optimizing treatment planning for patients with ankle syndesmosis injuries.

Keywords: ankle syndesmosis injury, diagnostic accuracy, computed tomography, radiographic measurements, Tibiofibular syndesmosis distance

Procedia PDF Downloads 73

Authors: Sudhir Kumar Singh, Debashish Chakravarty

Abstract:

Slope stability analysis is an important aspect in the field of geotechnical engineering. It is also important from safety, and economic point of view as any slope failure leads to loss of valuable lives and damage to property worth millions. This paper aims at mitigating the risk of slope failure by studying the effect of different joint sets on the stability of mine pit slopes under the influence of various external factors, namely degree of saturation, rainfall intensity, and seismic coefficients. Supervised machine learning approach has been utilized for making accurate and reliable predictions regarding the stability of slopes based on the value of Factor of Safety. Numerous cases have been studied for analyzing the stability of slopes using the popular Finite Element Method, and the data thus obtained has been used as training data for the supervised machine learning models. The input data has been trained on different supervised machine learning models, namely Random Forest, Decision Tree, Support vector Machine, and XGBoost. Distinct test data that is not present in training data has been used for measuring the performance and accuracy of different models. Although all models have performed well on the test dataset but Random Forest stands out from others due to its high accuracy of greater than 95%, thus helping us by providing a valuable tool at our disposition which is neither computationally expensive nor time consuming and in good accordance with the numerical analysis result.

Keywords: finite element method, geotechnical engineering, machine learning, slope stability

Procedia PDF Downloads 101

Authors: Sirilak Areerachakul, Nat Ployong, Supayothin Na Songkla

Abstract:

This research’s objective is to select the model with most accurate value by using Neural Network Technique as a way to filter potential students who enroll in IT course by electronic learning at Suan Suanadha Rajabhat University. It is designed to help students selecting the appropriate courses by themselves. The result showed that the most accurate model was 100 Folds Cross-validation which had 73.58% points of accuracy.

Keywords: artificial neural network, classification, students, e-learning

Procedia PDF Downloads 426

Authors: Hassan Irshad Bhatti, Saravanan Yuvaraja, Xiaohang Li

Abstract:

GaN-based devices, such as high electron mobility transistors (HEMTs), offer superior characteristics for high-power, high-frequency, and high-temperature applications [1]. However, this exceptional electrical performance is compromised by undesirable self-heating effects under high-power applications [2, 3]. Some of the issues caused by self-heating are current collapse, thermal runway and performance degradation [4, 5]. Therefore, accurate and reliable methods for measuring the temperature of individual devices on a chip are needed to monitor and control the thermal behavior of GaN-based devices [6]. Temperature measurement at the micro/nanoscale is a challenging task that requires specialized techniques such as Infrared microscopy, Raman thermometry, and thermoreflectance. Recently, micro-thermocouples (MTCs) have attracted considerable attention due to their advantages of simplicity, low cost, high sensitivity, and compatibility with standard fabrication processes [7, 8]. A micro-thermocouple is a junction of two different metal thin films, which generates a Seebeck voltage related to the temperature difference between a hot and cold zone. Integrating MTC in a device allows local temperature to be measured with high sensitivity and accuracy [9]. This work involves the fabrication and integration of micro-thermocouples (MTCs) to measure the channel temperature of GaN HEMT. Our fabricated MTC (Platinum-Chromium junction) has shown a sensitivity of 16.98 µV/K and can measure device channel temperature with high precision and accuracy. The temperature information obtained using this sensor can help improve GaN-based devices and provide thermal engineers with useful insights for optimizing their designs.

Keywords: Electrical Engineering, Thermal engineering, Power Devices, Semiconuctors

Procedia PDF Downloads 19

Authors: Hsiao-Chuan Huang, King-Lung Kuo, Mei-Hsin Lo, Hsiao-Yun Chou, Yusen Lin

Abstract:

The most robust and economical method for laboratory diagnosis of TB is to identify mycobacterial bacilli (AFB) under acid-fast staining despite its disadvantages of low sensitivity and labor-intensive. Though digital pathology becomes popular in medicine, an automated microscopic system for microbiology is still not available. A new AI-assisted automated microscopic system, consisting of a microscopic scanner and recognition program powered by big data and deep learning, may significantly increase the sensitivity of TB smear microscopy. Thus, the objective is to evaluate such an automatic system for the identification of AFB. A total of 5,930 smears was enrolled for this study. An intelligent microscope system (TB-Scan, Wellgen Medical, Taiwan) was used for microscopic image scanning and AFB detection. 272 AFB smears were used for transfer learning to increase the accuracy. Referee medical technicians were used as Gold Standard for result discrepancy. Results showed that, under a total of 1726 AFB smears, the automated system's accuracy, sensitivity and specificity were 95.6% (1,650/1,726), 87.7% (57/65), and 95.9% (1,593/1,661), respectively. Compared to culture, the sensitivity for human technicians was only 33.8% (38/142); however, the automated system can achieve 74.6% (106/142), which is significantly higher than human technicians, and this is the first of such an automated microscope system for TB smear testing in a controlled trial. This automated system could achieve higher TB smear sensitivity and laboratory efficiency and may complement molecular methods (eg. GeneXpert) to reduce the total cost for TB control. Furthermore, such an automated system is capable of remote access by the internet and can be deployed in the area with limited medical resources.

Keywords: TB smears, automated microscope, artificial intelligence, medical imaging

Procedia PDF Downloads 229

Authors: Daohong Yang, Xiang Zhang, Lei Li, Wanting Zhou

Abstract:

Visual Simultaneous Localization and Mapping (VSLAM) is a technology that obtains information in the environment for self-positioning and mapping. It is widely used in computer vision, robotics and other fields. Many visual SLAM systems, such as OBSLAM3, employ a constant-speed motion model that provides the initial pose of the current frame to improve the speed and accuracy of feature matching. However, in actual situations, the constant velocity motion model is often difficult to be satisfied, which may lead to a large deviation between the obtained initial pose and the real value, and may lead to errors in nonlinear optimization results. Therefore, this paper proposed a motion model based on acceleration, which can be applied on most SLAM systems. In order to better describe the acceleration of the camera pose, we decoupled the pose transformation matrix, and calculated the rotation matrix and the translation vector respectively, where the rotation matrix is represented by rotation vector. We assume that, in a short period of time, the changes of rotating angular velocity and translation vector remain the same. Based on this assumption, the initial pose of the current frame is estimated. In addition, the error of constant velocity model was analyzed theoretically. Finally, we applied our proposed approach to the ORBSLAM3 system and evaluated two sets of sequences on the TUM dataset. The results showed that our proposed method had a more accurate initial pose estimation and the accuracy of ORBSLAM3 system is improved by 6.61% and 6.46% respectively on the two test sequences.

Keywords: error estimation, constant acceleration motion model, pose estimation, visual SLAM

Procedia PDF Downloads 94

Authors: Shahad Nagoor, Lucy Hederman, Kevin Koidl, Annalina Caputo

Abstract:

Doctors’ notes reflect their impressions, attitudes, clinical sense, and opinions about patients’ conditions and progress, and other information that is essential for doctors’ daily clinical decisions. Despite their value, clinical notes are insufficiently researched within the language processing community. Automatically extracting information from unstructured text data is known to be a difficult task as opposed to dealing with structured information such as vital physiological signs, images, and laboratory results. The aim of this research is to investigate how Natural Language Processing (NLP) techniques and machine learning techniques applied to clinician notes can assist in doctors’ decision-making in Intensive Care Unit (ICU) for coronavirus disease 2019 (COVID-19) patients. The hypothesis is that clinical outcomes like survival or mortality can be useful in influencing the judgement of clinical sentiment in ICU clinical notes. This paper introduces two contributions: first, we introduce COVID_ICU_BERT, a fine-tuned version of clinical transformer models that can reliably predict clinical sentiment for notes of COVID patients in the ICU. We train the model on clinical notes for COVID-19 patients, a type of notes that were not previously seen by clinicalBERT, and Bio_Discharge_Summary_BERT. The model, which was based on clinicalBERT achieves higher predictive accuracy (Acc 93.33%, AUC 0.98, and precision 0.96 ). Second, we perform data augmentation using clinical contextual word embedding that is based on a pre-trained clinical model to balance the samples in each class in the data (survived vs. deceased patients). Data augmentation improves the accuracy of prediction slightly (Acc 96.67%, AUC 0.98, and precision 0.92 ).

Keywords: BERT fine-tuning, clinical sentiment, COVID-19, data augmentation

Procedia PDF Downloads 206