Classification of gastric emptying and orocaecal transit through artificial neural networks

Nenhuma Miniatura disponível

Data

2021-01-01

Autores

Bezerra, Anibal Thiago
Pinto, Leonardo Antonio [UNESP]
Rodrigues, Diego Samuel
Bittencourt, Gabriela Nogueira [UNESP]
de Arruda Mancera, Paulo Fernando [UNESP]
de Arruda Miranda, José Ricardo [UNESP]

Título da Revista

ISSN da Revista

Título de Volume

Editor

Resumo

Classical quantification of gastric emptying (GE) and orocaecal transit (OCT) based on half-life time T50, mean gastric emptying time (MGET), orocaecal transit time (OCTT) or mean caecum arrival time (MCAT) can lead to misconceptions when analyzing irregularly or noisy data. We show that this is the case for gastrointestinal transit of control and of diabetic rats. Addressing this limitation, we present an artificial neural network (ANN) as an alternative tool capable of discriminating between control and diabetic rats through GE and OCT analysis. Our data were obtained via biological experiments using the alternate current biosusceptometry (ACB) method. The GE results are quantified by T50 and MGET, while the OCT is quantified by OCTT and MCAT. Other than these classical metrics, we employ a supervised training to classify between control and diabetes groups, accessing sensitivity, specificity, f1 score, and AUROC from the ANN. For GE, the ANN sensitivity is 88%, its specificity is 83%, and its f1 score is 88%. For OCT, the ANN sensitivity is 100%, its specificity is 75%, and its f1 score is 85%. The area under the receiver operator curve (AUROC) from both GE and OCT data is about 0.9 in both training and validation, while the AUCs for classical metrics are 0.8 or less. These results show that the supervised training and the binary classification of the ANN was successful. Classical metrics based on statistical moments and ROC curve analyses led to contradictions, but our ANN performs as a reliable tool to evaluate the complete profile of the curves, leading to a classification of similar curves that are barely distinguished using statistical moments or ROC curves. The reported ANN provides an alert that the use of classical metrics can lead to physiological misunderstandings in gastrointestinal transit processes. This ANN capability of discriminating diseases in GE and OCT processes can be further explored and tested in other applications.

Descrição

Palavras-chave

Artificial intelligence, Deep learning, Experimental diabetes mellitus, Gastric emptying, Orocaecal transit

Como citar

Mathematical Biosciences and Engineering, v. 18, n. 6, p. 9511-9524, 2021.

Coleções