EP01.19: Performance of an AI software in automatic detection of fetal ultrasound planes

Abstract

Fetal ultrasound practice is a demanding clinical practice because of workforce shortage combined with Increased standards, for example ISUOG guidelines increased recommended standard planes from 32 to 42 in 11 years. In such a context, quality control of images captured appears key to monitor quality of delivered care. Particularly, artificial intelligence algorithms could provide automatic and objective quality assessment. The objective was to assess the performance of an artificial intelligence (AI) software (under development, not approved) for fetal ultrasound (US) exams, by automatic identification and labelling of standard planes. 692 fetal US exams, performed in all trimesters (T1, T2, T3), were collected from 36 North American and European centres. The sensitivity of the AI software (see table 1) was assessed by comparing the algorithm's blinded assessment to experts' assessment. The concordances between experts' and the algorithm's assessments for the presence of a standard plane was defined as true positive for the corresponding plane. Reported in table 1. Demonstrating high performance on enlarged scope, such an AI software appears relevant in supporting fetal ultrasound practitioners in their routine clinical practice by automatically detecting planes captured and directly evaluating exam completeness during live-scan. Profile Abdominal circumference Foot CRL Transthalamic/Transventricular Hand 4 Chambers 0.914 (0.906-0.921) Profile Abdominal circumference Foot Transthalamic/Transventricular Hand 4 Chambers Sagittal Spine Lips and Nose Long bone LVOT Transcerebellar Orbits 3 vessels/3 vessels and trachea Axial Bladder RVOT Axial Kidneys External genitalia (female and male) Placenta insertion 0.937 (0.933-0.940)

Ähnliche Arbeiten