RESUMO
OBJECTIVES@#To explore the application values of diatom artificial intelligence (AI) search system in the diagnosis of drowning.@*METHODS@#The liver and kidney tissues of 12 drowned corpses were taken and were performed with the diatom test, the view images were obtained by scanning electron microscopy (SEM). Diatom detection and forensic expert manual identification were carried out under the thresholds of 0.5, 0.7 and 0.9 of the diatom AI search system, respectively. Diatom recall rate, precision rate and image exclusion rate were used to detect and compare the efficiency of diatom AI search system.@*RESULTS@#There was no statistical difference between the number of diatoms detected in the target marked by the diatom AI search system and the number of diatoms identified manually (P>0.05); the recall rates of the diatom AI search system were statistically different under different thresholds (P<0.05); the precision rates of the diatom AI system were statistically different under different thresholds(P<0.05), and the highest precision rate was 53.15%; the image exclusion rates of the diatom AI search system were statistically different under different thresholds (P<0.05), and the highest image exclusion rate was 99.72%. For the same sample, the time taken by the diatom AI search system to identify diatoms was only 1/7 of that of manual identification.@*CONCLUSIONS@#Diatom AI search system has a good application prospect in drowning cases. Its automatic diatom search ability is equal to that of experienced forensic experts, and it can greatly reduce the workload of manual observation of images.
Assuntos
Humanos , Inteligência Artificial , Diatomáceas , Afogamento/diagnóstico , Fígado , Pulmão , Microscopia Eletrônica de VarreduraRESUMO
Se comparó una serie tradicional de pruebas bioquímicas con una de alta resolución para identificar 500 cepas de enterobacterias utilizando un método probabilístico para interpretar los resultados. La serie tradicional estuvo formada por 10 pruebas (producción de ornitina descarboxilasa, lisina descarboxilasa, lisina desaminasa, ácido sulfhídrico, indol y gas, hidrólisis de urea, utilización de citrato y de malonato, y movilidad). La serie de alta resolución, también con 10 pruebas, se integró con las primeras 4 mencionadas en la serie tradicional y 6 de fermentación de hidratos de carbono (adonitol, L-arabinosa, celobiosa, L-ramnosa, rafinosa y sorbitol). Con la serie de alta resolución se asignaron identidades únicas a 445 cepas (351 con probabilidad de 1,0 y 94 con probabilidades entre 0,010 y 0,999), y de las restantes 55 cepas, a 53 y 2 se asignaron dos y tres identidades probables respectivamente. Con la serie tradicional se asignaron identidades únicas a 306 cepas (110 con probabilidad de 1,0 y 196 con probabilidades entre 0,001 y 0,999) y a 179 y 5 se asignaron dos y tres identidades probables respectivamente. Diez cepas no se pudieron identificar. Todos los indicadores analizados revelaron la superioridad de la serie de alta resolución. El método probabilístico permitió la comparación objetiva de ambas series.
A traditional series of biochemical tests-was compared to a high-resolution one in order to identify 500 strains of enterobacteria, using a probabilistic method for the interpretation of experimental results. The traditional series was formed by 10 tests (ornithine decarboxylase, lysine decarboxylase, lysine deaminase, sulfhydric acid, indol and gas production, urea hydrolysis, citrate and malonate utilization, and motility). The high-resolution one was also formed by 10 tests, including the first 4 tests mentioned above and 6 carbohydrate fermentation tests (adonitol, L-arabinose, cellobiose, L-rhamnose, raffinose, and sorbitol). With the high-resolution series, single identities were assigned to 445 strains (351 with a probability of 1.0 and 94 with probabilities in the range 0.010-0.999), and for the remaining strains two and three probable identities were assigned to 53 and 2 strains, respectively. With the traditional series, single identities were assigned to 306 strains (110 with a probability of 1.0 and 196 with probabilities in the 0.001-0.999 range), two and three probable identities were assigned to 179 and 5 strains respectively; 10 strains turned out to be non-identifiable. Every parameter of comparison used revealed the superiority of the high-resolution series. The probabilistic method for interpretation of experimental results allowed an objective comparison of both series.
Foi comparada uma série tradicional de testes bioquímicos com uma outra de alta resolução para identificar 500 cepas de enterobactérias usando uma abordagem probabilística para a interpretação dos resultados. A série tradicional consistiu em 10 testes (produção de ornitina descarboxilase, lisina descarboxilase, lisina desaminase, ácido sulfídrico, indol e gás, hidrólise da ureia, utilização de citrato e de malonato, e mobilidade). A série de alta resolução, também com 10 ensaios, integrou-se com as primeiras 4 mencionadas na série tradicional e 6 de fermentação de hidratos de carbono (adonitol, L-arabinose, celobiose, L-ramnose, rafinose e sorbitol). Com a série de alta resolução foram atribuídas identidades únicas a 445 cepas (351, com probabilidade 1,0, e 94, com probabilidade entre 0,010 e 0,999), e das restantes 55 cepas, a 53 e 2 foram atribuídas duas e três identidades possíveis, respectivamente. Com a série tradicional foram atribuídas identidades únicas a 306 cepas (110 com probabilidade de 1,0 e 196 com probabilidades entre 0,001 e 0,999) e a 179 e 5 foram atribuídas duas e três identidades prováveis, respectivamente. Dez cepas não puderam ser identificadas. Todos os indicadores analisados demonstraram a superioridade da série de alta resolução. O método probabilístico permitiu a comparação objetiva de ambas as séries.