Portal de Periódicos da CAPES

OP01.10: Auditing the quality of ultrasound images using an AI solution: ScanNav® for fetal second trimester ultrasound scans

2019; Wiley; Volume: 54; Issue: S1 Linguagem: Inglês

10.1002/uog.20656

1469-0705

Mohammad Yaqub, K. Cook, Kim Cocks, Z. Chen, B. Chikkanna, Nicholas Sleep, J. Alison Noble, Aris T. Papageorghiou,

Fetal and Pediatric Neurological Disorders

ScanNav (Intelligent Ultrasound, UK) uses artificial intelligence models for real-time automatic ultrasound image quality audit by assessing frozen ultrasound images. We investigate ScanNav performance in a) image categorisation, and b) image adherence to quality standards. Prospective study of routine obstetric sonograms, following the UK second-trimester Fetal Anomaly Scanning Programme (FASP) protocol, were captured at St George's Hospital, London, UK. These were assessed by ScanNav. Five experienced sonographers (the panel) independently assessed ScanNav image categorisation; and agreement and kappa statistics on adherence to objective quality standards for the 7 FASP protocol views (head transventricular (TV), head transcerebellar (TC), abdomen (AC), femur, spine (SP), coronal face (FC), heart 4 chamber view (4CH)) plus an “Other” category for non-protocol view images. ScanNav was considered correct when the panel was in unanimous (5:0) or majority (4:1) agreement with ScanNav or when the panel disagreed with split 3:2 (or 2:3). 1,853 images from 89 subjects were assessed. ScanNav's agreement with the panel for image categorisation was 99% (95% Confidence interval (CI) 98-100%). Mean agreement between ScanNav and the panel for adherence to quality was 83% and kappa was 0.62 (95% CI 0.57 – 0.67), identical to sonographer intra-agreement of 83% and kappa 0.60 (95% CI 0.55-0.65). ScanNav's determination of protocol adherance agreed well with each sonographer for all views. Agreement was 100%, 95%, 94%, 91%, 84%, 84% and 78% for FL, SP, TC, TV, FC, 4CH, AC respectively. Importantly, no protocol adherent images were miscategorised by ScanNav. ScanNav accurately categorises obstetric images into protocol views. Performance of categorisation and quality assessment is similar to a panel of experts, and appears to relate to the clinical complexity of the view (e.g., 4CH and FC have lower agreement). Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.