Introduction: Breast cancer is the most common malignancy among women worldwide, and mammography screening programs have been shown to significantly reduce the mortality associated to this malignancy. Despite their proven effectiveness, suboptimal sensitivity and overdiagnosis are still an issue. There are still few studies in the literature, that have demonstrated the usefulness of artificial intelligence (AI) in improving diagnostic accuracy and reducing the workload in screening, most of which were not conducted within an official mammography screening program. Objective: The aim of this study was to evaluate the effectiveness of AI as a decision support tool in the mammography screening program of Treviso. This study compared the diagnostic performance of standard digital mammography with AI-supported double reading. The main objective was to assess whether AI can improve the performance of mammography screening when used as a decision support tool, according to European guidelines. Materials and Methods: A retrospective study was conducted, including asymptomatic women aged 49 to 75 years, who underwent biennial screening mammography between 2019 and 2023. The first biennium (2019-2021) involved standard digital mammography with independent readings by two readers, while the second biennium (2021-2023) incorporated AI as a decision support tool. Results: The results showed a significant increase in the detection rate (DR) with the introduction of AI (8.2 vs. 6.9; p = 0.001), with no statistically significant increase in the recall rate (RR). Additionally, there was an increase in the positive predictive value (PPV) of the screening mammography (28.4 vs. 24.6; p = 0.002). AI contributed to the identification of a higher number of infiltrating lobular and non-special type tumors. No significant differences were found in the proportion of in situ tumors, suggesting a potentially reduced risk of overdiagnosis. Conclusions: These results support the use of AI as a decision support tool in mammography screening programs, improving diagnostic efficacy. The study findings also confirm the utility of AI as a support tool for radiologists, enabling more accurate identification of breast malignacies.
Introduzione: il tumore al seno è la neoplasia più frequente tra le donne a livello globale; i programmi di screening mammografico hanno dimostrato di ridurre significativamente la mortalità ad esso associata. Nonostante la loro efficacia, persistono problematiche quali sensibilità subottimale e sovradiagnosi. In merito all’utilizzo dell’intelligenza artificiale (IA), in letteratura sono presenti ancora pochi studi che abbiano dimostrato l’utilità di questo strumento nel migliorare l’accuratezza diagnostica e nel ridurre il carico di lavoro nello screening, la maggior parte dei quali, inoltre, non eseguita all’interno di un vero e proprio programma di screening mammografico. Scopo: valutare l'efficacia dell’IA come strumento di supporto decisionale nel programma di screening mammografico dell’Ulss 2. Questo studio ha confrontato la performance diagnostica della mammografia digitale standard con la doppia lettura IA-supportata. L'obiettivo è verificare se l'IA, utilizzata come strumento di supporto decisionale, possa migliorare le performance dello screening mammografico, in accordo con le raccomandazioni delle linee guida europee. Materiali e Metodi: è stato condotto uno studio retrospettivo, includendo donne asintomatiche di età compresa tra 49 e 75 anni, sottoposte a screening biennale nel periodo 2019-2023. Il primo biennio (2019-2021) ha utilizzato la sola mammografia digitale standard refertata da due lettori in maniera indipendente, mentre il secondo biennio (2021-2023) ha integrato l’IA come strumento di supporto decisionale. Risultati: i risultati hanno evidenziato un incremento significativo del tasso di identificazione (detection rate, DR) con l’introduzione dell’IA (8,2 vs. 6,9; p = 0,001), senza un aumento statisticamente significativo dei tassi di richiamo (recall rate, RR). Inoltre, è stato riscontrato un incremento del valore predittivo positivo della mammografia di screening (VPP) (28,4 vs. 24,6; p = 0,002). L’IA ha contribuito all'identificazione di un maggior numero di tumori infiltranti di istotipo lobulare e di istotipo non speciale. Non sono emerse differenze significative nella proporzione dei tumori in situ, suggerendo un potenziale ridotto rischio di sovradiagnosi. Conclusioni: Questi risultati suffragano l’utilizzo dell’IA come strumento di supporto nei programmi di screening mammografico, migliorando l’efficacia diagnostica. I risultati dello studio confermano, inoltre, l’utilità dell’IA come strumento di supporto per i radiologi, permettendo di identificare in modo più accurato i tumori mammari.
Impatto dell'intelligenza artificiale (IA) nel programma di screening mammografico di Treviso: risultati della doppia lettura IA supportata.
CAVALLERI, CRISTINA
2022/2023
Abstract
Introduction: Breast cancer is the most common malignancy among women worldwide, and mammography screening programs have been shown to significantly reduce the mortality associated to this malignancy. Despite their proven effectiveness, suboptimal sensitivity and overdiagnosis are still an issue. There are still few studies in the literature, that have demonstrated the usefulness of artificial intelligence (AI) in improving diagnostic accuracy and reducing the workload in screening, most of which were not conducted within an official mammography screening program. Objective: The aim of this study was to evaluate the effectiveness of AI as a decision support tool in the mammography screening program of Treviso. This study compared the diagnostic performance of standard digital mammography with AI-supported double reading. The main objective was to assess whether AI can improve the performance of mammography screening when used as a decision support tool, according to European guidelines. Materials and Methods: A retrospective study was conducted, including asymptomatic women aged 49 to 75 years, who underwent biennial screening mammography between 2019 and 2023. The first biennium (2019-2021) involved standard digital mammography with independent readings by two readers, while the second biennium (2021-2023) incorporated AI as a decision support tool. Results: The results showed a significant increase in the detection rate (DR) with the introduction of AI (8.2 vs. 6.9; p = 0.001), with no statistically significant increase in the recall rate (RR). Additionally, there was an increase in the positive predictive value (PPV) of the screening mammography (28.4 vs. 24.6; p = 0.002). AI contributed to the identification of a higher number of infiltrating lobular and non-special type tumors. No significant differences were found in the proportion of in situ tumors, suggesting a potentially reduced risk of overdiagnosis. Conclusions: These results support the use of AI as a decision support tool in mammography screening programs, improving diagnostic efficacy. The study findings also confirm the utility of AI as a support tool for radiologists, enabling more accurate identification of breast malignacies.File | Dimensione | Formato | |
---|---|---|---|
tesi definitiva.pdf
accesso riservato
Dimensione
1.66 MB
Formato
Adobe PDF
|
1.66 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/81531