A carregar...
Publicação
Análise comparativa da precisão do diagnóstico de um software de inteligência artificial e médicos dentistas na avaliação de cáries (utilizando o índice CPO-D) e lesões periapicais em radiografias panorâmicas
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.92 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Introdução: A inteligência artificial (IA) deixou de ser uma promessa distante e começa a ganhar espaço ao lado do clínico na análise de ortopantomografias (radiografias panorâmicas). Contudo, antes de a adoptarmos, é imperativo demonstrar, segundo padrões da medicina dentária baseada na evidência, que a balança pende a favor do benefício clínico e não dos custos ou dos riscos éticos. Objetivos: Comparar a precisão de diagnóstico de cárie dentária e lesões apicais entre um software de inteligência artificial (IA) e médicos dentistas com mais de 5 anos de experiência.
Materiais e Métodos: Foram analisadas 200 ortopantomografias digitais de adultos, com qualidade técnica elevada, obtidas entre setembro de 2024 e dezembro de 2024. Três médicos dentistas experientes avaliaram cada imagem e definiram por consenso a referência diagnóstica. As mesmas imagens foram processadas pelo software WeDiagnostix (modos Sensível – WD-S e Ótimo – WD-O), sendo calculadas sensibilidade, especificidade, valores preditivos, precisão, F1-score, AUC e κ de Cohen.
Resultados: Os médicos dentistas mostraram concordância substancial (κ = 0,75). Para cáries, o WD-S atingiu 85,3 % de sensibilidade e 88,7 % de especificidade (AUC 0,870). O WD-O trocou sensibilidade (55,5 %) por especificidade quase total (99,0 %), mantendo exactidão de 96,2 % (AUC 0,773). No reconhecimento de lesões apicais, o WD-S registou 88,2 % de sensibilidade e 95,9 % de especificidade (AUC 0,920), enquanto o WD-O apresentou 75,0 % e 99,2 %, respectivamente, com exactidão de 98,9 % (AUC 0,871).
Conclusão: O software de IA, especialmente no modo Ótimo (WD-O), apresentou elevada precisão, especificidade e valor preditivo positivo, sendo comparável ao desempenho de médicos dentistas experientes. Já o modo Sensível (WD-S) revelou maior sensibilidade, à custa de mais falsos positivos. Assim, sugere-se que o software tem precisão diagnóstica equivalente ou superior à dos médicos experientes na deteção de cáries e lesões apicais.
Introduction: Artificial intelligence (AI) is no longer a distant promise and is beginning to establish its place alongside clinicians in the interpretation of orthopantomograms (panoramic radiographs). However, before its adoption, it is essential to demonstrate—according to evidence-based dentistry standards—that clinical benefit outweighs costs or ethical risks. Objetives: To compare the diagnostic accuracy of an AI-based software and dentists with over 5 years of experience in detecting dental caries and apical lesions. Material and Methods: A total of 200 high-quality digital panoramic radiographs from adult patients (September 2024–December 2024) were analysed. Three experienced dentists reviewed each image and reached a consensus diagnosis used as reference. The same images were processed by the WeDiagnostix software (Sensitive (WD-S) and Optimum (WD-O) modes), and sensitivity, specificity, predictive values, accuracy, F1-score, AUC and Cohen’s κ were calculated. Results: Dentists showed substantial agreement (κ = 0.75). For caries, WD-S achieved 85.3% sensitivity and 88.7% specificity (AUC 0.870). WD-O traded sensitivity (55.5%) for near-total specificity (99.0%) with 96.2% accuracy (AUC 0.773). For apical lesions, WD-S reached 88.2% sensitivity and 95.9% specificity (AUC 0.920), while WD-O achieved 75.0% and 99.2%, respectively, with 98.9% accuracy (AUC 0.871). Conclusion: The AI software, particularly in Optimum mode (WD-O), demonstrated high precision, specificity, and positive predictive value, comparable to experienced dentists. The Sensitive mode (WD-S) showed higher sensitivity but generated more false positives. These findings suggest that the software provides diagnostic accuracy equivalent or superior to that of experienced clinicians in detecting caries and apical lesions.
Introduction: Artificial intelligence (AI) is no longer a distant promise and is beginning to establish its place alongside clinicians in the interpretation of orthopantomograms (panoramic radiographs). However, before its adoption, it is essential to demonstrate—according to evidence-based dentistry standards—that clinical benefit outweighs costs or ethical risks. Objetives: To compare the diagnostic accuracy of an AI-based software and dentists with over 5 years of experience in detecting dental caries and apical lesions. Material and Methods: A total of 200 high-quality digital panoramic radiographs from adult patients (September 2024–December 2024) were analysed. Three experienced dentists reviewed each image and reached a consensus diagnosis used as reference. The same images were processed by the WeDiagnostix software (Sensitive (WD-S) and Optimum (WD-O) modes), and sensitivity, specificity, predictive values, accuracy, F1-score, AUC and Cohen’s κ were calculated. Results: Dentists showed substantial agreement (κ = 0.75). For caries, WD-S achieved 85.3% sensitivity and 88.7% specificity (AUC 0.870). WD-O traded sensitivity (55.5%) for near-total specificity (99.0%) with 96.2% accuracy (AUC 0.773). For apical lesions, WD-S reached 88.2% sensitivity and 95.9% specificity (AUC 0.920), while WD-O achieved 75.0% and 99.2%, respectively, with 98.9% accuracy (AUC 0.871). Conclusion: The AI software, particularly in Optimum mode (WD-O), demonstrated high precision, specificity, and positive predictive value, comparable to experienced dentists. The Sensitive mode (WD-S) showed higher sensitivity but generated more false positives. These findings suggest that the software provides diagnostic accuracy equivalent or superior to that of experienced clinicians in detecting caries and apical lesions.
Descrição
Dissertação para obtenção do grau de Mestre no Instituto Universitário Egas Moniz
Palavras-chave
Inteligência artifical Cárie dentária Diagnóstico Lesões periapicais