NUMERO: #20231122

Título: MACHINE LEARNING ALGORITHMS ENHANCE PREDICTION OF MALIGNANT TRANSFORMATION USING INDIVIDUAL DYSPLASIA FEATURES

Nome do Apresentador: Marcelo SPERANDIO

Categoria do Trabalho: Painel de pesquisa científica (PPC)

Área Temática: Patologia Oral

Resumo: This study aimed to evaluate the accuracy of artificial intelligence (AI) algorithms based on individual epithelial dysplasia (ED) features in predicting malignant transformation (MT) in oral leukoplakia (OL). The microscopic features of cell architecture and atypia were analyzed in 99 cases of OL to identify individual predictors of MT. The associations between each dysplasia feature and MT was evaluated by the Spearman´s test, area under the curve (AUC), as well as accuracy. AI algorithms were developed using Random Forest models. The results showed that loss of cohesion, increased number and size of nucleoli, and inflammation were the single most important predictors of MT. When combined, these 3 features generated AUC=0.73 with a positive predictive value (PPV) of 35% (61% sensitivity, 75% specificity). The AI algorithms identified six key predictors and had a 95% PPV (100% sensitivity and 71% specificity). In conclusion, 3 individual ED features were key predictors, and AI algorithms were significantly superior to traditional methods in establishing MT risk in oral leukoplakia.

Autor 1:  Marcelo SPERANDIO

E-mail 1:  [email protected]

Autor 2:   Leticia da Silveira Terra JUNQUEIRA

E-mail 2:  [email protected]

Autor 3 :  Vera Cavalcanti de ARAÚJO

E-mail 3:  [email protected]

Autor 4:  Victor Angelo Martins MONTALLI

E-mail 4:  [email protected]

Autor 5:  Paulo de Camargo MORAES

E-mail 5:  [email protected]

Autor 6:  André Luis Santana de FREITAS

E-mail 6:  [email protected]

Autor 7:  Andresa Borges SOARES

E-mail 7:  [email protected]