Automated left ventricular dimension assessment using artificial intelligence

Abstract Background and purpose Artificial intelligence (AI) has the potential to greatly improve efficiency reproducibility of quantification in echocardiography, but gain widespread use it must both meet expert standards excellence have a transparent methodology. We developed an online platform enable multiple collaborators annotate medical images for training validating neural networks. Methods Using our collaborative 9 echocardiographers labelled 2056 that comprised dataset. They four points from where standard parasternal long axis (PLAX) measurements (interventricular septum, posterior wall, left ventricular dimension) would be made. these we trained 2d convolutional network replicate labels. Separately, curated external validation dataset systolic diastolic frames 100 PLAX acquisitions. Each were twice by 13 different experts, average 26 was taken as consensus standard. then compared individual experts AI on standard, calculated precision deviation (SD) signed differences Results For septum thickness, had SD 1.8 mm (ICC 0.81; 95% CI 0.73 0.97), with 2.0 0.64; 0.57 0.72). wall 1.4 0.54; 0.38 0.66), 2.2 0.37; 0.29 0.46). The AI's internal dimension 3.5 0.93, 0.90 0.94), 4.4mm 0.82, 0.78 0.95). Both performed better diastole than systole (precision 2.5mm vs 4.3mm, p<0.0001; 3.3mm 5.3mm, p<0.0001). Conclusions group echocardiography able perform which matched reference more closely any expert's own measurements. This open, approach may model development is explainable to, trusted clinicians. Funding Acknowledgement Type funding sources: Public grant(s) – National budget only. Main source(s): NIHR Imperil BRC ITMATDr Howard additionally funded Wellcome. Online platformResults