NUMERO: #20260072

Título: MACHINE LEARNING-BASED PREDICTION OF EROSIVE CHANGES IN THE MANDIBULAR CORTEX IN EDENTULOUS PATIENTS

Nome do Apresentador: Fábio Wildson Gurgel COSTA

Categoria do Trabalho: Pôster - Investigação Original (IO)

Área Temática: Imaginologia dentomaxilofacial

Resumo: To assess the performance of machine learning models in predicting erosive changes of the mandibular cortex in edentulous patients. This retrospective investigation included 390 panoramic radiographs from edentulous individuals aged 51-80 years. Linear measurements were used as input features for different machine learning algorithms. The outcome was defined using a dichotomized Klemetti classification: C1 (normal cortex) versus C2/C3 (cortical erosion). Model performance was assessed using accuracy, precision, sensitivity (recall), and F1-score. Performance varied by model and evaluation metric. Gradient Boosting achieved the highest accuracy (0.67) and weighted F1-score (0.67), with better performance for C1, but low sensitivity for C2/C3 (0.25). Logistic Regression, KNN, Decision Tree, and SVM showed higher sensitivity for detecting erosion (0.75), although with low precision (0.23-0.27) and lower overall accuracy. Random Forest demonstrated the most balanced performance (accuracy=0.63; macro F1-score=0.56), while MLP showed intermediate results (sensitivity=0.69). Machine learning models based on radiomorphometric measurements demonstrated potential for identifying mandibular cortical alterations in edentulous patients, with a trade-off between overall accuracy and sensitivity for erosion detection. Funding: Ceará Foundation for Support to Scientific and Technological Development (FUNCAP; process no. FPD-0213-00294.01.00/23 and ITR-0214-00074.01.00/23).

Autor 1:  Fábio Wildson Gurgel COSTA

E-mail 1:  [email protected]

Autor 2:   Marcela Lima GURGEL

E-mail 2:  [email protected]

Autor 3 :  Esther Carneiro RIBEIRO

E-mail 3:  [email protected]

Autor 4:  Paulo Goberlânio de Barros SILVA

E-mail 4:  [email protected]

Autor 5:  Lúcia Helena Soares CEVIDANES

E-mail 5:  [email protected]

Autor 6:  João Victor de Oliveira RODRIGUES

E-mail 6:  [email protected]

Autor 7:  José Wellington Franco da SILVA

E-mail 7:  [email protected]