E!113653 Apollo: real-time AI for brain MRI radiology workflow optimization

The main aim of the Apollo project was to evaluate an AI-based software (Apollo) for real-time analysis of magnetic resonance images (MRI) for automated detection of a range of brain pathologies (stroke, bleeds, tumors, etc) in order to adapt the acquisition protocol prospectively according to the detected pathology (or absence of pathology). This is termed 'smart' MR protocols. It turned out to be very challenging to obtain approval for the integration of Apollo into the radiological workflow at OUS due to the very extensive and time-consuming documentation requirements - which were incompatible with a 2-year research project. We, therefore, decided to test Apollo in retrospective data (i.e. MR data already available in the OUS database) in order to assess the diagnostic accuracy of the Apollo software in classifying brain pathology from MRIs. This assessment has been carried out in a large dataset of MR images acquired at Ullevål Hospital during the last five years - with promising results. We are now in the final stages of comparing the results from Apollo with the corresponding radiological diagnosis obtained from an experienced radiologist in the same material. Finally, we are also looking into the possibility of using Apollo specifically for better characterization of acute strokes. Here, we have promising results in using Apollo for better assessment of the extent of irreversible brain tissue damage from MRIs acquired in the acute phase.

Due to regulatory barriers at Oslo University Hospital, it was impossible to fully implement the proposed software platform (Apollo) for testing in the clinical workflow. We have still been able to test the platform's ability to detect different brain pathologies in retrospective data automatically. We found the platform particularly useful for acute stroke diagnostics and have pursued this indication specifically in the latter part of the project period.

This project supports the development of the deep learning-based Apollo software, providing real-time AI medical guidance and decision support in brain MRI scanning to improve the workflow in radiology and ensure that attention is brought to images with potential indications of pathologies. The key functions of Apollo (Smart Protocol, Image Reporter and Triage Advisor) will be developed to improve productivity and reach a clinically relevant accuracy across 95% of all brain diseases diagnosed.