NUMERO: #20240201

Título: ARTIFICIAL INTELLIGENCE AND RADIOMICS FOR THE DIAGNOSIS OF BONE AND INTRAOSSEOUS LESIONS OF GNATHIC BONES

Nome do Apresentador: Daniela Giraldo Roldan

Categoria do Trabalho: Painel de pesquisa científica (PPC)

Área Temática: Áreas afins no caráter multidisciplinar

Resumo: Objective: This study aims to develop and assess AI and radiomics-based systems for diagnosing bone and intraosseous lesions, specifically focusing on distinguishing between benign and malignant lesions.

Methods: A data set of 77 panoramic radiographs containing intraosseous lesions were categorized into two groups: 50 benign and 27 malignant. Regions of interest (ROIs) were annotated using bounding boxes. Object detection and classification were performed using YOLOv8, with data augmentation techniques including horizontal inversion and rotation up to 10 degrees. All images were standardized to 640x640 size and underwent the same gray filter. The augmented dataset comprised 355 images, divided into training, validation, and testing sets, following a 70/20/10 ratio (240, 75 and 40 images respectively).

Results: Results metrics did not show a high performance; data augmentation process improved the resilience of the model but was unable to provide unrepresented contexts within the dataset. However, they are promising and have great opportunities for improvement, with an average of 0.27 and 0.29 for precision and recall.

Conclusion: The articulated use of AI and radiomics are promising, it is important to increase the volume of dataset to ensure the performance of the model. Exploring different architectures and techniques is key to enhancing study outcomes.

Autor 1:  Daniela Giraldo Roldan

E-mail 1:  [email protected]

Autor 2:   Matheus Henrique de Cerqueira Pinto

E-mail 2:  [email protected]

Autor 3 :  Anna Luíza Damaceno Araújo

E-mail 3:  [email protected]

Autor 4:  Marcio Ajudarte Lopes

E-mail 4:  [email protected]

Autor 5:  André Carlos Ponce de Leon Ferreira de Carvalho

E-mail 5:  [email protected]

Autor 6:  Pablo Agustin Vargas.

E-mail 6:  [email protected]