Postoperative delirium prediction using machine learning models and preoperative electronic health record data

Background Accurate, pragmatic risk stratification for postoperative delirium (POD) is necessary to target preventative resources toward high-risk patients. Machine learning (ML) offers a novel approach to leveraging electronic health record (EHR) data for POD prediction. We sought to develop and internally validate a ML-derived POD risk prediction model using preoperative risk features, and to compare its performance to models developed with traditional logistic regression. Methods This was a retrospective analysis of preoperative EHR data from 24,885 adults undergoing a procedure requiring anesthesia care, recovering in the main post-anesthesia care unit, and staying in the hospital at least overnight between December 2016 and December 2019 at either of two hospitals in a tertiary care health system. One hundred fifteen preoperative risk features including demographics, comorbidities, nursing assessments, surgery type, and other preoperative EHR data were used to predict postoperative delirium (POD), defined as any instance of Nursing Delirium Screening Scale ≥2 or positive Confusion Assessment Method for the Intensive Care Unit within the first 7 postoperative days. Two ML models (Neural Network and XGBoost), two traditional logistic regression models (“clinician-guided” and “ML hybrid”), and a previously described delirium risk stratification tool (AWOL-S) were evaluated using the area under the receiver operating characteristic curve (AUC-ROC), sensitivity, specificity, positive likelihood ratio, and positive predictive value. Model calibration was assessed with a calibration curve. Patients with no POD assessments charted or at least 20% of input variables missing were excluded. Results POD incidence was 5.3%. The AUC-ROC for Neural Net was 0.841 [95% CI 0. 816–0.863] and for XGBoost was 0.851 [95% CI 0.827–0.874], which was significantly better than the clinician-guided (AUC-ROC 0.763 [0.734–0.793], p < 0.001) and ML hybrid (AUC-ROC 0.824 [0.800–0.849], p < 0.001) regression models and AWOL-S (AUC-ROC 0.762 [95% CI 0.713–0.812], p < 0.001). Neural Net, XGBoost, and ML hybrid models demonstrated excellent calibration, while calibration of the clinician-guided and AWOL-S models was moderate; they tended to overestimate delirium risk in those already at highest risk. Conclusion Using pragmatically collected EHR data, two ML models predicted POD in a broad perioperative population with high discrimination. Optimal application of the models would provide automated, real-time delirium risk stratification to improve perioperative management of surgical patients at risk for POD.

AVALIAÇÃO DA ACURÁCIA DE ALGORITMOS DE CLASSIFICAÇÃO DE IMAGENS ORBITAIS NA BAÍA DA BABITONGA, NORDESTE DE SANTA CATARINA

O objetivo deste trabalho foi a avaliação da acurácia de algoritmos de classificação do uso e cobertura do solo, quando aplicados a uma imagem orbital de média resolução espacial. Para esse estudo foram utilizadas as bandas espectrais da faixa do visível e infravermelho próximo, do sensor Operational Land Imager – OLI na Baía da Babitonga/SC. Foram propostas nove classes de cobertura do solo, que serviram como controle para testar 11 algoritmos classificadores: Binary Encoding, Example Based Feature Extraction, IsoData, K-Means, Mahalanobis Distance, Maximum Likelihood, Minimum Distance, Neural Net, Parallelepiped, Spectral Angle Mapper e Spectral Information Divergence. O classificador Maximum Likelihood foi o que apresentou o melhor desempenho, obtendo um índice Kappa de 0,89 e acurácia global de 95,5%, sendo capaz de distinguir as nove classes de cobertura do solo propostas. Evaluation of the Accuracy of Orbital Image Classification Algorithms in Babitonga Bay, northeast of Santa Catarina A B S T R A C TThe objective of this work was to evaluate the classification algorithms accuracy of the soil use and cover when applied to a spatial mean orbital image. For this study we used the visible and near infrared spectral bands of the Operational Land Imager - OLI sensor in Babitonga Bay / SC. Nine classes of soil cover were proposed, which served as control to test 11 classifier algorithms: Binary Encoding, Example Based Feature Extraction, IsoData, K-Means, Mahalanobis Distance, Maximum Likelihood, Minimum Distance, Neural Net, Parallelepiped, Spectral Angle Mapper and Spectral Information Divergence. The Maximum Likelihood classifier presented the best performance, obtaining a Kappa index of 0.89 and a global accuracy of 95.5%, being able to distinguish the nine proposed classes of soil cover.Keywords: Algorithms Accuracy, Babitonga Bay, Orbital image, Remote sensing, Soil Use and Cover.

Predicción de la calidad en revestimientos moldeados para puertas mediante el uso de minería de datos

Un revestimiento moldeado para puertas es un tablero de madera de alta densidad que es utilizado como el principal componente en la fabricación de puertas. Para asegurar su comercialización, se debe cumplir con exigentes normas de calidad, siendo la principal norma aquella que mide la fuerza necesaria para desprender el revestimiento de la estructura de una puerta. Los ensayos de calidad son realizados cada dos horas y sus resultados son obtenidos luego de aproximadamente cinco horas. Si los resultados muestran que los revestimientos están fuera del estándar de calidad exigido, se generan pérdidas económicas debido a este tiempo de espera. Esta investigación propone el uso de minería de datos mediante técnicas de machine learning para predecir en forma continua esta medida de calidad y reducir las pérdidas económicas asociadas a la espera de los resultados. Para la aplicación de minería de datos, se creó una base de datos en base al registro histórico de las variables del proceso productivo y de los ensayos de calidad. La metodología empleada es el descubrimiento de conocimiento en bases de datos KDD (Knowledge Discovery in Databases). La aplicación de esta metodología permitió identificar las principales variables que afectan la calidad de los revestimientos y entrenar cuatro algoritmos de machine learning para predecir su calidad. Los resultados muestran que el algoritmo que mejor predice la calidad es Neural Net y permiten demostrar que la implementación del algoritmo Neural Net reducirá las pérdidas económicas asociadas a la espera de los resultados de los ensayos de calidad.

Automation of Carbonate Rock Thin Section Description Using Cognitive Image Recognition

Simulation Engineers and Geomodelers rely on reservoir rock geological descriptions to help identify baffles, barriers and pathways to fluid flow critical to accurate reservoir performance predictions. Part of the reservoir modelling process involves Petrographers laboriously describing rock thin sections to interpret the depositional environment and diagenetic processes controlling rock quality, which along with pressure differences, controls fluid movement and influences ultimate oil recovery. Supervised Machine Learning and a rock fabric labelled data set was used to train a neural net to recognize Modified Durham classification reservoir rock thin section images and their individual components (fossils and pore types) plus predict rock quality. The image recognition program's accuracy was tested on an unseen thin section image database.

DEEP LEARNING TECHNOLOGY IN PATTERN RECOGNITION

The article discusses the methods of face recognition based on convolutional neural net-works, the problems of face recognition in the presence of interference or face masking, the main stages of training neural networks and the process of actual recognition.

Integrating Pore Geometrical Characteristics for Permeability Prediction of Tight Carbonates Utilizing Artificial Intelligence

Accurate permeability estimation in tight carbonates is a key reservoir characterization challenge, more pronounced with heterogeneous pore structures. Experiments on large volumes of core samples are required to precisely characterize permeability in such reservoirs which means investment of large amounts of time and capital. Therefore, it is imperative that an integrated model exists that can predict field-wide permeability for un-cored sections to optimize reservoir strategies. Various studies exist with a scope to address this challenge, however, most of them lack universality in application or do not consider important carbonate geometrical features. Accordingly, this work presents a novel correlation to determine permeability of tight carbonates as a function of carbonate pore geometry utilizing a combination of machine learning and optimization algorithms. Primarily, a Deep Learning Neural Network (NN) is constructed and further optimized to produce a data-driven permeability predictor. Customization of the model to tight-heterogenous pore-scale features is accomplished by considering key geometrical carbonate topologies, porosity, formation resistivity, pore cementation representation, characteristic pore throat diameter, pore diameter, and grain diameter. Multiple realizations are conducted spanning from a perceptron-based model to a multi-layered neural net with varying degrees of activation and transfer functions. Next, a physical equation is derived from the optimized model to provide a stand-alone equation for permeability estimation. Validation of the proposed model is conducted by graphical and statistical error analysis of model testing on unseen dataset. A major outcome of this study is the development of a physical mathematical equation which can be used without diving into the intricacy of artificial intelligence algorithms. To evaluate performance of the new correlation, an error metric comprising of average absolute percentage error (AAPE), root mean squared error (RMSE), and correlation coefficient (CC) was used. The proposed correlation performs with low error values and gives CC more than 0.95. A possible reason for this outcome is that the machine learning algorithms can construct relationship between various non-linear inputs (for e.g., carbonate heterogeneity) and output (permeability) parameters through its inbuilt complex interaction of transfer and activation function methodologies.

Development of Droplet Microfluidics Capable of Quantitative Estimation of Single-Cell Multiplex Proteins

This study presented constriction microchannel based droplet microfluidics realizing quantitative measurements of multiplex types of single-cell proteins with high throughput. Cell encapsulation with evenly distributed fluorescence labelled antibodies stripped from targeted proteins by proteinase K was injected into the constriction microchannel with the generated fluorescence signals captured and translated into protein numbers leveraging the equivalent detection volume formed by constriction microchannels in both droplet measurements and fluorescence calibration. In order to form the even distribution of fluorescence molecules within each droplet, the stripping effect of proteinase K to decouple binding forces between targeted proteins and fluorescence labelled antibodies was investigated and optimized. Using this microfluidic system, binding sites for beta-actin, alpha-tubulin, and beta-tubulin were measured as 1.15±0.59×106, 2.49±1.44×105, and 2.16±1.01×105 per cell of CAL 27 (N cell=15486), 0.98±0.58×106, 1.76±1.34×105 and 0.74±0.74×105 per cell of Hep G2 (N cell=18266). Neural net pattern recognition was used to differentiate CAL 27 and Hep G2 cells, producing successful rates of 59.4% (beta-actin), 64.9% (alpha-tubulin), 88.8% (beta-tubulin), and 93.0% in combination, validating the importance of quantifying multiple types of proteins. As a quantitative tool, this approach could provide a new perspective for single-cell proteomic analysis.

Satellite-based analysis of classification algorithms applied to the riparian zone of the Malaya Kokshaga river

The paper comparatively analyses the accuracy of land cover classification in the riparian zone of the Malaya Kokshaga river in the Mari El Republic of Russia using Sentinel-2A satellite images with the algorithms of supervised classification: Maximum Likelihood (ML), Decision Tree (DT) and Neural Net (NN) in the ENVI-5.2 software package. Six main classes of land cover were identified based on field studies: coniferous, mixed (deciduous), shrublands, herbaceous, and water. The assessment of the area and the structure of land cover showed that forest covers 76% of the entire territory of the riparian area of the Malaya Kokshaga river. The analysis of the results of thematic mapping shows that the overall classification accuracy obtained by the ML algorithm is 96.09%, by NN - 94.51%, and by DT - 86.54%. The producer’s accuracy and user’s accuracy for most classes have the maximum value when the ML algorithm is used. For the NN algorithm, the maximum value of producer’s accuracy is observed for the mixed (deciduous) class, while for the DT algorithm – for the coniferous. When classified using all three algorithms the water and bare land classes were mixed, which requires more detailed work when estimating riparian forest ecosystems.