Background: Routine mid-trimester anomaly scans require the acquisition of numerous standard views and biometric measurements, and their duration and completeness are influenced by maternal, fetal, technical, and operator-related factors. Artificial intelligence (AI) has been shown in research settings to shorten anomaly scans, but whether this benefit translates to routine practice is unknown. Aim: To evaluate the use of AI during routine second trimester anomaly scans without prior AI-specific training or changes to routine workflow, and to explore its association with scan duration and recall rate. Materials and methods: We conducted a prospective service evaluation from 1 December 2025 to 31 May 2026 in the maternity ultrasound departments of Queen Charlotte’s and Chelsea Hospital and St. Mary's Hospital, Imperial College Healthcare NHS Trust, London, including anomaly scans performed between 18+0 and 20+6 weeks of gestation. Sonographers recorded whether AI was used (for biometry and standard plane recognition) and the need for recall to complete the examination. Operators used the AI technology without altering their scanning technique or standard anomaly scan imaging protocols, and without dedicated training. Three ultrasound systems were used: two AI-capable systems, used for all AI-assisted scans and for standard scans with AI disabled, and one older non-AI-capable system, used for standard scans only. The outcomes were scan duration and recall rate, compared between AI-assisted and non-AI-assisted scans. Additional variables, such as gestational age, BMI, maternal age, presence of fibroids, and parity, were also recorded. Results were adjusted for maternal and pregnancy characteristics; to account for differences between ultrasound systems, a further analysis was restricted to scans performed on the two AI-capable systems and adjusted for system. Results: A total of 2,997 scans were analysed: 393 were AI-assisted and 2,604 non-AI. Median scan duration was 28 minutes in the AI group and 30 minutes in the non-AI group (p<0.001). However, nearly half of standard scans were performed on an older, non-AI-capable system, which had the longest median duration. When the analysis was restricted to the 1,716 scans performed on AI-capable systems and adjusted for system, AI was no longer associated with scan duration (adjusted β −0.61 minutes, 95% CI −1.60 to 0.38; p=0.225). Obesity, older maternal age, nulliparity, and presence of fibroids were independently associated with longer scans. Recall rates were 9.7% without AI and 8.7% with AI (p=0.519), and AI was not associated with a change in recall rate. Obesity and earlier gestational age were associated with higher odds of recall. Conclusions: In this real-world evaluation of untutored use of AI embedded in commercial ultrasound systems, AI-assisted scans were slightly shorter, but this association was attenuated and no longer significant when comparison was restricted to the AI-capable systems, suggesting it was at least partly explained by differences between ultrasound systems. Scan duration was also associated with maternal factors: obesity, older maternal age, and the presence of fibroids were each associated with longer scans. AI was not associated with recall rate. Further evaluation after targeted sonographer training on AI use is planned.
Can artificial intelligence enhance routine mid-trimester scans? Findings from a prospective service evaluation.
CATAPANO, MARIASILVIA
2024/2025
Abstract
Background: Routine mid-trimester anomaly scans require the acquisition of numerous standard views and biometric measurements, and their duration and completeness are influenced by maternal, fetal, technical, and operator-related factors. Artificial intelligence (AI) has been shown in research settings to shorten anomaly scans, but whether this benefit translates to routine practice is unknown. Aim: To evaluate the use of AI during routine second trimester anomaly scans without prior AI-specific training or changes to routine workflow, and to explore its association with scan duration and recall rate. Materials and methods: We conducted a prospective service evaluation from 1 December 2025 to 31 May 2026 in the maternity ultrasound departments of Queen Charlotte’s and Chelsea Hospital and St. Mary's Hospital, Imperial College Healthcare NHS Trust, London, including anomaly scans performed between 18+0 and 20+6 weeks of gestation. Sonographers recorded whether AI was used (for biometry and standard plane recognition) and the need for recall to complete the examination. Operators used the AI technology without altering their scanning technique or standard anomaly scan imaging protocols, and without dedicated training. Three ultrasound systems were used: two AI-capable systems, used for all AI-assisted scans and for standard scans with AI disabled, and one older non-AI-capable system, used for standard scans only. The outcomes were scan duration and recall rate, compared between AI-assisted and non-AI-assisted scans. Additional variables, such as gestational age, BMI, maternal age, presence of fibroids, and parity, were also recorded. Results were adjusted for maternal and pregnancy characteristics; to account for differences between ultrasound systems, a further analysis was restricted to scans performed on the two AI-capable systems and adjusted for system. Results: A total of 2,997 scans were analysed: 393 were AI-assisted and 2,604 non-AI. Median scan duration was 28 minutes in the AI group and 30 minutes in the non-AI group (p<0.001). However, nearly half of standard scans were performed on an older, non-AI-capable system, which had the longest median duration. When the analysis was restricted to the 1,716 scans performed on AI-capable systems and adjusted for system, AI was no longer associated with scan duration (adjusted β −0.61 minutes, 95% CI −1.60 to 0.38; p=0.225). Obesity, older maternal age, nulliparity, and presence of fibroids were independently associated with longer scans. Recall rates were 9.7% without AI and 8.7% with AI (p=0.519), and AI was not associated with a change in recall rate. Obesity and earlier gestational age were associated with higher odds of recall. Conclusions: In this real-world evaluation of untutored use of AI embedded in commercial ultrasound systems, AI-assisted scans were slightly shorter, but this association was attenuated and no longer significant when comparison was restricted to the AI-capable systems, suggesting it was at least partly explained by differences between ultrasound systems. Scan duration was also associated with maternal factors: obesity, older maternal age, and the presence of fibroids were each associated with longer scans. AI was not associated with recall rate. Further evaluation after targeted sonographer training on AI use is planned.| File | Dimensione | Formato | |
|---|---|---|---|
|
Catapano_Mariasilvia.pdf
Accesso riservato
Dimensione
878.6 kB
Formato
Adobe PDF
|
878.6 kB | 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/20.500.12608/120807