NUMERO: #20231085

Título: Comparison between human and computerized analysis in cytopathology. An experimental tool for oral cancer screening

Nome do Apresentador: Igor Cavalcante Guedes

Categoria do Trabalho: Painel de pesquisa científica (PPC)

Área Temática: Patologia Oral

Resumo: Early detection of squamous cell carcinoma (OSCC) is essential for a better prognosis to patients. This early detection may be reached by cytopathology of epithelial cells of oral mucosa. Cytopathology is time consuming and present different levels of accuracy considering human evaluation. Artificial Intelligence (AI) can be employed as an auxiliary method to improve the accuracy standards. This study compares the effectiveness of qualitative evaluation performed by human and computerized system (CNN) for oral cell smears stained by the Papanicolaou technique. Oral mucosa smears were collected from 5 patients from the following groups: control subjects, exposed group to risk factors for OSCC, patients with Oral Potentially Malignant Disorders (OPMD), and patients with OSCC. The smears were stained with the Papanicolaou technique and at least 100 fields were photographed at 400x magnification. The images were analyzed by human and CNN system developed by our research group. The comparison between human and AI analyses had a percentage of agreement of 60.7%, and disclosed difficulty in analyze aggregated cells (clusters). Despite being a developing tool, AI showed promise in automated assistance for cytopathology with the possibility to improve this exam accuracy contributing for OSCC screening strategies.

Autor 1:  Igor Cavalcante Guedes

E-mail 1:  [email protected]

Autor 2:   Manuel Menezes de Oliveira Neto

E-mail 2:  [email protected]

Autor 3 :  Fernanda Visioli

E-mail 3:  [email protected]

Autor 4:  Maikel Maciel Rönnau

E-mail 4:  [email protected]

Autor 5:  Tatiana Wannmacher Lepper

E-mail 5:  [email protected]

Autor 6:  Ana Laura Espinosa

E-mail 6:  [email protected]

Autor 7:  Pantelis Varvaki Rados

E-mail 7:  [email protected]