The quality of chest radiographs is a practical issue because deviations from quality standards cost readers' time, may lead to misdiagnosis and hold legal risks. Inadequate patient positioning is one of the frequently encountered image quality shortcomings [1]. Automatic and reproducible assessment of the most important quality figures on every acquisition is an enabler to measure, maintain, and improve quality rates in a department on an everyday basis. An artificial intelligence-based software to measure and evaluate the following critical aspects of patient positioning in chest...

Artificial intellicence (a combination of convolutional neural networks and probabilistic anatomical atlases) is used to detect anatomical features robustly and consistently even in images not showing diagnostic quality: The convolutional neural networks are trained to detect the lung field border, the posterior ribs, the diaphragm, the clavicle heads, and the spinous processes. Probabilistic anatomical atlases are used to relate these anatomies to each other, to extrapolate incomplete data, and to reject inconsistant false detections (see fig 2). Based on these robustly detected landmarks several quality...

Technical evaluation of the three modules demonstrated quantitative accuracy and qualitative robustness with respect to landmark localization also for challenging cases. On a test set of 142 unseen cases the following errors have been measured: error on median μ σ cranial collimation Δc [mm] abdominal collimation Δd [mm] lateral collimation Δl, Δr [mm] 2.5 5.7 2.9 3.0 6.6 3.6 2.7 4.5 2.9 clavicle-head position [mm] rotation asymmetry α 2.5 0.06 4.1 0.11 4.7 0.12 inhalation rib index cR,cL [#ribs] 0.01 0.27 0.44 In on-going clinical...

The artificial intelligence modules enable timely, objective, reproducible,and quantitative quality feedback.They can be automatically applied on an everyday basis to every image. Technical evaluation demonstrated quantitative accuracy in measuring aspects of positioning in a retrospective analysis. Comparisonof these quality figures with expert ratings at multiple radiological departments are reported separately [2,3]. The modules enable automated indexing and sorting with respect to positioning issues, e.g. "missing apex" The availabilitysuch a tool for image assessment is expected to change quality management in radiography.

N. Wieberneit; Hamburg/DE - Employee at Philips Medical Systems DMC GmbH J. von Berg; Hamburg/DE - Employee at Philips Research D. Bystrov; Hamburg/DE - Employee at Philips Research S. Krönke; Hamburg/DE - Employee at Philips Research A. Gooßen; Hamburg/DE - Employee at Philips Research M. Brück; Hamburg/DE - Employee at Philips Research T. Harder; Hamburg/DE - Employee at Philips Research S. Young; Hamburg/DE - Employee at Philips Research

[1] K.J. Little et al., Unified Database for Rejected Image Analysis Across Multiple Vendors in radiography, J Am Coll Radiol 14 (2017), 208-216 [2] M. Englmeier et al, Inter-rater variability in the assessment of positioning quality in chest x-ray images, Poster C-05601ECR 2020 [3]S. Young et al., Measuring patient positioning quality in clinical chest radiographs, Poster C-12328 ECR 2020