VP34.18: Managing expectations in the use of artificial intelligence for diagnosis support in maternal and fetal imaging: a scoping review

Abstract

To assess the use of artificial intelligence (AI) in maternal and fetal imaging. We conducted a scoping review on the use of AI in PubMed/MEDLINE database, in obstetrics and gynecology publications during the last 10 years in English. The paper characteristics analysis were the type of AI methods, the source of data and the result of the AI process. The initial systematic review yield 419 PubMed citations in obstetrics and gynecology indexed with “artificial intelligence” in MeSH. After title and abstract relevance screening, 47 citations were deemed relevant for maternal and fetal imaging. Five citations were withdrawn, because no AI was involved. In the 42 selected papers, AI methods were machine learning (n = 35, with 57% of Artificial Neural Networks), semantic reasoning (n = 3), expert system (n = 1) and not specified (n = 3). The AI processes were applied on US image datasets (2D, 3D, video) in 27/42 papers, on MRI image datasets in 5/47 and on other datasets (biometric values, Doppler signal, EHR, annotations) in 15/42 papers. Prospective clinical validation was conducted in one paper, prospective validation was performed in a simulation setting in one other paper and internal validity was assessed on retrospective data in 27 papers. Overall, the contributions for maternal and fetal imaging were 2 fully functional applications, 3 prototypes and 37 preclinical AI methods (image segmentation, feature and plane recognition, automated measurements, models). Clinical outcomes of AI are generally limited in maternal and fetal imaging. Most AI methods are machine learning applied on image, volume and video processing with promising results awaiting clinical setting validation. Actual decision support for diagnosis assistance are covered only by AI methods based on semantic reasoning and expert systems.

Ähnliche Arbeiten