Banca de DEFESA: LUANNY DE BRITO AVELINO CASSIANO

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : LUANNY DE BRITO AVELINO CASSIANO
DATE: 23/07/2025
TIME: 08:00
LOCAL: GOOGLE MEET
TITLE:

USE OF ARTIFICIAL INTELLIGENCE FOR CALCULATING THE LENGTH OF MINI-IMPLANTS USED IN THE MARPE TECHNIQUE ON COMPUTED TOMOGRAPHY

KEY WORDS:

Artificial intelligence; Deep learning; Palatal expansion technique; Orthodontic anchorage procedures; Cone-beam computed tomography.

PAGES: 58
BIG AREA: Ciências da Saúde
AREA: Odontologia
SUBÁREA: Ortodontia
SUMMARY:

Introduction: The mini-screw assisted rapid palatal expansion technique requires the selection of mini-screws with effective lengths that ensure good primary stability through bicortical engagement. This necessitates manual measurement of palatal thickness on maxillary computed tomography scans, which is time-consuming and subject to operator measurement errors. In this context, automating these measurements using artificial intelligence (AI) emerges as a promising approach. Objective: This study evaluated the applicability of AI for measuring the bone and mucosa of the palate in cone-beam computed tomography (CBCT) scans of the maxilla or face to determine the effective length of mini-screws. Materials and Methods: This is an observational, cross-sectional, descriptive, and inferential study. The sample consisted of 588 reference points corresponding to 12 points per scan (49 CBCT scans of the maxilla and face), which were manually annotated using the Create new point list tool in the 3DSlicer® software. These points served as the ground-truth for training a 3D U-Net neural network. The landmark detection model was trained using 3D patches, employing convolutional neural networks combined with lightweight attention mechanisms. Appropriate validation metrics were applied, including Mean Radial Error (MRE), Success Detection Rate (SDR), Intraclass Correlation Coefficient (ICC), and Bland-Altman plots. Results: The model achieved an MRE of 1.13 ± 0.47 mm, and SDRs of 85%, 92.5%, and 97.5% within radii of 2.0 mm, 2.5 mm, and 3.0 mm, respectively. The global ICC for all validation scans was 0.66, with 5 out of 10 scans reaching substantial to excellent agreement. The Bland-Altman analysis supported the ICC results and indicated that the AI model tends to slightly underestimate measurements compared to the manual method, considered the gold standard. Conclusion: The developed model showed promising results with a mean error below 2.0 mm and moderate agreement with the ground- truth. Although it demonstrates potential as a useful tool for measuring the bone and mucosal thickness of the palate, the orthodontist’s clinical judgment remains indispensable, as the precision of the AI model is not yet fully equivalent to human assessment.

COMMITTEE MEMBERS:
Presidente - 1009536 - SERGEI GODEIRO FERNANDES RABELO CALDAS
Interno - 1378325 - ARTHUR CESAR DE MEDEIROS ALVES
Externo à Instituição - ALEXANDRE ANTONIO RIBEIRO - ABO