Background: 

The field of radiology has changed drastically over the years with the introduction of artificial intelligence (AI). In order to develop accurate AI algorithms, large, annotated datasets are often required. Unfortunately, the formation of accurate cohorts proves to be difficult as the medical imaging files may be missing crucial metadata. One such type of metadata that may be missing or inaccurate is the body part scanned tag.

Methods: 

We propose the use of a deep learning framework for enriching the metadata concerning which body parts have been scanned in MRI volumes. A method has been previously developed [Schuhegger 2021] for computed tomography (CT) volumes, but has not been expanded for other types of radiological scans. We propose to expand this network by also allowing for the multiple types of MR scan contrasts. 

Results:

As a pilot project, a small cohort of MRI volumes has been created and the deep learning network trained to detect the prostate region. The expansion to other areas of the body and the improvement of the network for multiple MRI scan parameters is currently under development. 

Conclusions:

The use of a deep learning framework shows promise in enriching the metadata in medical imaging volumes.