NUMERO: #20230416

Título: DEEP LEARNING MAY DIFFERENTIATE HEAD AND NECK HIGH-GRADE LYMPHOMAS

Nome do Apresentador: Lucas Lacerda de SOUZA

Categoria do Trabalho: Painel de pesquisa científica (PPC)

Área Temática: Patologia Oral

Resumo: Objective: The diagnosis of high-grade lymphomas remains challenging for pathologists. This study aimed to implement a deep learning-based model (DLBM) to assist pathologists in differentiating diffuse large B-cell lymphoma (DLBCL), Burkitt lymphoma (BL), and plasmablastic lymphoma (PL).
Study Design: Whole slide images from 30 patients (10 with DLBCL, 10 with BL, and 10 with PL) were annotated, segmented, and fragmented into 41,227 patches (DLBCL=13,813, BL=14,705 and PL=12,709) of 299×299 pixels to use in training the DLBM for classification of high-grade lymphomas. Pre-trained models VGG16, Xception and ResNet101 and other open-source libraries for machine learning and image processing were used.
Results: The pre-trained models VGG16, Xception, and ResNet101 achieved accuracies of 91.66%, 95.24%, and 94.41%, respectively, with a training ratio of 50%. F1-score for VGG16 was 0.91, for Xception was 0.95 and for ResNet101 was 0.94. The ROC curve analysis showed that VGG16 had a fine class separation ability of 93.68%, Xception had 96.39%, and ResNet101 had 95.76%.
Conclusion: The DLBM used in this study is feasible to differentiate DLBCL, BL and PL. Taken together, these results suggest that although the three pre-trained models performed well in differentiating the high-grade lymphomas, Xception was the best among them.

Autor 1:  Lucas Lacerda de SOUZA

E-mail 1:  [email protected]

Autor 2:   Alan Roger SANTOS-SILVA

E-mail 2:  [email protected]

Autor 3 :  Marcio Ajudarte LOPES

E-mail 3:  [email protected]

Autor 4:  Ayyub ALZAHEM

E-mail 4:  [email protected]

Autor 5:  Ibrahim OMARA

E-mail 5:  [email protected]

Autor 6:  Ahmed HAGAG

E-mail 6:  [email protected]

Autor 7:  Pablo Agustin VARGAS

E-mail 7:  [email protected]