NUMERO: #20230161

Título: DEEP LEARNING FOR ODONTOGENIC TUMORS' CLASSIFICATION

Nome do Apresentador: Daniela Giraldo ROLDAN

Categoria do Trabalho: Painel de pesquisa científica (PPC)

Área Temática: Patologia Oral

Resumo: Objective: To evaluate Deep Learning performance for the classification of ameloblastoma and ameloblastic carcinoma. Materials and methods: digital slides were manually annotated to assign regions with tumor in order to segment and further fragment into patches of 556x556pixels, according to the models' kernels. The total sample were divided into sets of training, validation and tests sets. In the training stage, different hyperparameters of the Resnet 50, VGG 16 and DenseNet networks were studied. Results and conclusion: Preliminary results revealed that the models have learning potential, but not generalization so far. The training and validation metrics did not show convergence, characterizing an overfit. In addition, the test assessment was unsatisfactory, with an average of 0.65, 0.67, 0.61, 0.68 and 0.63 for accuracy, precision, sensitivity, specificity and F1 score, respectively. These results may be influenced by the overlapping of characteristics between the tumors. Further experiments will improve the models and test other techniques of sampling, data augmentation, use of different classes of tumors and implementation of different convolutional neural networks.

Autor 1:  Daniela Giraldo ROLDAN

E-mail 1:  [email protected]

Autor 2:   Anna Luiza Damaceno ARAÚJO

E-mail 2:  [email protected]

Autor 3 :  Erin Crespo Cordeiro RIBEIRO

E-mail 3:  [email protected]

Autor 4:  Viviane Mariano DA SILVA

E-mail 4:  [email protected]

Autor 5:  Marcio Ajudate LOPES

E-mail 5:  [email protected]

Autor 6:  Matheus Cardoso MORAES

E-mail 6:  [email protected]

Autor 7:  Pablo Agustin Vargas.

E-mail 7:  [email protected]