11/29 2021

Automated Marker Detection of X-Rays Errors using Machine Learning

Laterality markers floating

Radiobotics has conducted a diagnostic of a Copenhagen hospital’s clinical knee x-ray production to analyse discrepancies in laterality markers which are used during x-ray acquisition.

RBfracture is a new AI-powered product now cleared for sale in the EU to assist in detecting fractures in Emergency Medicine. Globally, 178 million fractures were registered in 2019. At the same time, missed fractures account for the vast majority of diagnostic errors in the emergency setting. This is particularly where RBfracture can add value, assisting as an extra pair of eyes and safety net, aiding in detecting if the patient has suffered a fracture.
X-ray of knee
Figure. Standing side views of the knees are taken with a medial-lateral projection, thus the kneecap is expected to appear in the left side of the radiograph for an anatomical right knee. In this radiograph the kneecap is in the right part of the image, and thus the true expected anatomy is a left knee. However, the radiograph is marked with a “R” and thus highlighted by RBmarker as an inconsistency between expected anatomy and detected marker.
Radiobotics’ product, RBmarker, was used to conduct the analysis. RBmarker is a quality assurance tool which can be integrated into clinical Radiology departments to analyse errors in Radiographic markers, for both lead and digital markers. The tool can be useful at point of acquisition as a co-pilot for Radiographers and technicians. The product also gives a facility the ability to have a bird’s eye view of performance, including highlighting areas of excellence and areas which may need an intervention in order to lower discrepancies. This enables quality departments to better manage risk and unwanted clinical discrepancies. They can use this tool to do a case by case analysis and a second pair of eyes checking for any potential errors in marker placement. The results gave an indication of cases where no marker was placed on images, inconsistencies detected or where 2 contradicting markers were used in the same image. Over 2,000 images were analysed and the results are presented below:
Table showing statistics
This case study presents some important clinical considerations. Almost 1 in 5 images detected were not following agreed departmental protocols. Current manual analysis of these images did not raise errors of this magnitude, which is a clear strength of this tool. Presented with this kind of data gives departments a range of opportunities to reduce this unwanted variation from clinical protocols. We will present how this information can be used by Quality Management, Team Leaders and Practitioners.Firstly, for practitioners, this tool can be deployed on stations within the imaging console room, which allows practitioners to have in-the-moment feedback on their marker placement and can alert practitioners when they have misplaced markers or failed to add a marker to their images.Secondly, for quality management, this tool gives a holistic view of performance on the use of laterality markers at their facility. This tool can present trends and show the effect of on-the-ground interventions to evaluate if they are effective.Finally, for team leaders, this data arms them with data to alert where interventions should be carried out. Many of these interventions can be low-tech solutions and simple to execute.  These kinds of interventions range from raising this issue during team meetings or creating simple posters or similar measures to remind staff to double check they have placed a marker on the image before completing imaging post processing.To conclude this tool can be a useful co-pilot tool for Quality Management, Team Leads and Practitioners to understand when Radiographic markers are being used in conjunction with agreed clinical protocols.