Banca de QUALIFICAÇÃO: CAMILA FRANZON CHINI

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : CAMILA FRANZON CHINI
DATE: 27/11/2024
TIME: 09:00
LOCAL: Plataforma Teams
TITLE:

MACHINE LEARNING ALGORITHMS FOR DETECTION OF CAROTID ATHEROMA CALCIFICATION IN PANORAMIC RADIOGRAPHS 

KEY WORDS:

Artificial Intelligence; Carotid Arteries; Oral Diagnosis; Radiography Panoramic; Stroke; Vascular Calcification." 

PAGES: 100
BIG AREA: Ciências da Saúde
AREA: Odontologia
SUMMARY:

Stroke is caused by the accumulation of atherosclerotic plaques, which, when calcified, can be identified in panoramic radiographs (PR). PR serves as an auxiliary tool for identifying these calcifications, and in this context, the use of Artificial Intelligence (AI)based methods can aid the process. The primary aim of this study was to develop and validate an automated system for the identification of carotid artery calcification (CAC) in PR using AI algorithms. Specific objectives involved analyzing the influence of AI on clinical decision-making among different groups of professionals and the development of a digital repository. The research retrospectively evaluated 19,205 PR images, and a database was created, supplemented with information from medical and dental records. For the primary objective, 372 images were selected, and 574 CACs were manually detected. The models used were FastViT for classification, AttentionNet for detection, and UNet for segmentation. Model results varied, with FastViT achieving precision and accuracy of 87%. AttentionNet obtained an average precision of 41% and an average sensitivity of 55% for medium-sized CAC. UNet achieved a precision of 61%, sensitivity of 73%, and AUC of 80%. A comparison was made between the influence of classification on detection and segmentation, showing that the introduction of the classification step significantly increased precision (from 0.48 to 0.93), accuracy (from 0.79 to 0.97), and specificity (from 0.75 to 0.98). The research demonstrated the capability to develop machine learning algorithms geared towards CAC detection and segmentation in PR. Although the model is in the initial phase, it already shows good detection and segmentation capabilities; however, there remains room for improvement. In future steps, a comparative study will be applied to assess the impact of AIassisted clinical decision-making among different expertise groups. The image database will also be consolidated into a digital repository. ”

COMMITTEE MEMBERS:
Externo à Instituição - FREDERICO SAMPAIO NEVES - UFBA
Externo ao Programa - 2518570 - MARCOS FAGUNDES CAETANO - nullPresidente - 2315081 - NILCE SANTOS DE MELO
Interno - 3475960 - PAULO TADEU DE SOUZA FIGUEIREDO