NUMERO: #20240659

Título: Risk assessment of OPMD using individual dysplasia features and combined systems using machine learning

Nome do Apresentador: Marcelo SPERANDIO

Categoria do Trabalho: Painel de pesquisa científica (PPC)

Área Temática: Estomatologia

Resumo: Assessing the risk of malignant transformation (MT) in oral potentially malignant disorders (OPMD) is challenging due to limitations in current dysplasia grading systems, such as poor reproducibility and high costs. Machine learning (ML) offers promise in improving risk assessment, as shown in previous studies on oral leukoplakia (OL). This study aimed to develop ML models combining demographic data, dysplasia features, and DNA analysis to enhance MT risk assessment in OPMD, potentially improving clinical outcomes.
The investigation included 98 archived oral mucosal incisional biopsies diagnosed clinically and histologically as OL, of which 18 underwent malignant transformation. Dysplasia features were reviewed using the WHO classification, and nuclei suspensions were prepared for flow cytometry. A multilayer perceptron (MLP) model was developed and evaluated using accuracy, precision, and F1-score, with cross-validation. Significant correlations were found between MT and attributes like 4cER, G2, dysplasia grading, loss of epithelial cohesion, inflammatory infiltrate, and increased size and number of nucleoli. Visual analysis of KDE distributions suggested potential for differentiation between transforming and non-transforming subgroups. MLP modelling yielded accuracy (91%), sensitivity (73.3%), specificity (96.25%), precision (78.33%) and F1-score (74.28%). Conventional and ML-based strategies for classifying high-risk and low-risk lesions demonstrated accurate predictions, especially when MLP models were applied.

Autor 1:  Marcelo SPERANDIO

E-mail 1:  [email protected]

Autor 2:   Guilherme Iani Pontes

E-mail 2:  [email protected]

Autor 3 :  Andresa Borges Soares

E-mail 3:  [email protected]

Autor 4:  Saman Warnakulasuriya

E-mail 4:  [email protected]

Autor 5:  André Luis Santana de Freitas

E-mail 5:  [email protected]

Autor 6:  Paulo de Camargo Moraes

E-mail 6:  [email protected]

Autor 7:  Matheus Cardoso Moraes

E-mail 7:  [email protected]