Diagnosis of patients with blood cell count for COVID-19: An explainable artificial intelligence approach / Diagnóstico de pacientes com hemograma para COVID-19: Uma abordagem com inteligência artificial explicável / Diagnóstico de pacientes con hemograma para COVID-19: un enfoque explicable de inteligencia artificial

Objective: Present an explainable artificial intelligence (AI) approach for COVID-19 diagnosis with blood cell count. Methods: Five AI algorithms were evaluated: Logistic Regression, Random Forest, Support Vector Machine, Gradient Boosting and eXtreme Gradient Boosting. A Bayesian optimization with 5-Fold cross-validation was used to hyper-parameters tuning. The model selection evaluated three results: cross validation performance, test set prediction performance and a backtest: performance on identifying patients negative for COVID-19, but positive for others respiratory pathologies. Shapley Additive explanations (SHAP) was used to explain the chosen model. Results: A Random Forest model was obtained with 77.7% F1-Score (IC95%:57.1;92.3), 85.9% AUC (IC95%:73.7;95.9), 74.4% Sensitivity (IC95%:50.0;92.1) and 97.5% Specificity (IC95%:93.6;100.0). The main features were leukocytes, platelets and eosinophils. Conclusion: The research highlights the importance of model interpretability, demonstrating blood cell count as a possibility for COVID-19 diagnosis. The methodological structure developed, using TRIPOD's guidelines, can be extrapolated to other pathologies.

Objetivo: Propor uma abordagem com inteligência artificial explicável para diagnóstico de COVID-19 com hemograma. Métodos: Cinco algoritmos de IA foram testados: Regressão Logística, Florestas Aleatórias, Máquina de Vetores de Suporte, Gradient Boosting e eXtreme Gradient Boosting. Os hiper-parâmetros foram definidos através da otimização bayesiana com validação cruzada 5-Fold. A seleção de modelo utilizou três resultados de desempenho para definir o melhor modelo: validação cruzada, conjunto de teste e rendimento na identificação de pacientes negativos para COVID-19, porém positivos para outras patologias respiratórias (backtest). Ao final, Shapley Additive explanations (SHAP) foi utilizado para explicar o modelo escolhido. Resultados: Obteve-se um modelo Random Forest com F1-Score de 77.7% (IC95%:57.1;92.3), AUC de 85.9% (IC95%:73.7;95.9), Sensibilidade de 74.4% (IC95%:50.0;92.1) e Especificidade de 97.5% (IC95%:93.6;100.0). As principais variáveis foram leucócitos, plaquetas e eosinófilos. Conclusão: A pesquisa destaca a importância da interpretabilidade do modelo, demonstrando o hemograma como uma possibilidade para diagnosticar COVID-19. A estrutura metodológica desenvolvida no estudo, utilizando as diretrizes do TRIPOD, pode ser extrapolada para detecção de outras patologias.

Objetivo: Proponer un enfoque explicable de inteligencia artificial (IA) para el diagnóstico de COVID-19 con el uso de hemograma. Métodos: Cinco modelos de IA fueron evaluados: Logistic Regression, Random Forest, Support Vector Machine, Gradient Boosting e eXtreme Gradient Boosting. Los hiper-parámetros fueron definidos a través de optimización bayesiana con validación cruzada 5-Folds. La selección del modelo se utilizó tres resultados: rendimiento del validación cruzada, rendimento en conjunto de pruebas y el análisis de desempeño en identificación de pacientes negativos para COVID-19, pero positivos para otras patologías respiratorias (backtest). Shapley Additive explanations (SHAP) fue utilizado para explicar el modelo elegido. Resultados: Se obtuvo un modelo Random Forest con F1-Score de 77.7% (IC95%:57.1;92.3), AUC de 85.9% (IC95%:73.7;95.9), Sensibilidad de 74.4% (IC95%:50.0;92.1) y Especificidad de 97.5% (IC95%:93.6;100.0). Las principales variables fueron leucocitos, plaquetas y eosinófilos. Conclusión: La investigación presenta la importancia de la interpretabilidad del modelo, demostrando el uso de hemograma como posibilidad para diagnosticar COVID-19. La estructura elaborada, siguiendo las directrices de TRIPOD, puede ser extrapolar para otras patologías.

Simulation of aerated lagoon using artificial neural networks and multivariate regression techniques.

The aim of this study was to develop an empirical model that provides accurate predictions of the biochemical oxygen demand of the output stream from the aerated lagoon at International Paper of Brazil, one of the major pulp and paper plants in Brazil. Predictive models were calculated from functional link neural networks (FLNNs), multiple linear regression, principal components regression, and partial least-squares regression (PLSR). Improvement in FLNN modeling capability was observed when the data were preprocessed using the PLSR technique. PLSR also proved to be a powerful linear regression technique for this problem, which presents operational data limitations.