Year: 2024
Type: conference-abstract
Abstract: Early detection of spina bifida (SB) is crucial, to allow in utero surgery or pregnancy termination. The nuchal translucency (NT) view is recommended at the first trimester (ultrasound (US) scan and allows screening for SB with high specificity/sensitivity when the operator systematically visualises the brainstem (BS), intracranial translucency (IT) and cisterna magna (CM) in the same image. The objective of this study was to assess whether AI software (not approved) correctly identifies the absence of these structures, alerting the physician to possible SB. The validation dataset consisted of 95 images of NT scans of non-anomalous fetuses from USA and European centres. Each scan was reviewed by a fetal US expert who assessed the visualisation/non-visualisation of these structures (BS, IT, CM). The validation dataset was enriched with 11 NT images of fetuses with SB from 3 German ultrasound centres. The AI software was run on these images to evaluate its performance in identifying absent IT or CM. Among the 95 normal images, the simultaneous presence of both IT and CM was observed in 61 by the fetal US expert. The AI software detected the simultaneous presence of IT and CM in 54 out of these 61 examinations, achieving a specificity of 88.5%. All cases of SB raised an alert of the AI software, resulting in a sensitivity of 100% for SB. The AI software demonstrates excellent sensitivity for SB detection by detecting absence of IT or CM on the NT view. It was also demonstrated that in practice these structures are not always present on NT images because of variable examination quality. Quality control tools such as this AI software have the potential to improve examination quality and make first trimester SB screening, once reserved for experts, accessible to screeners on a routine basis.
DOI: 10.1002/uog.28027
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4402358286