This article sought to investigate the potential of generative models in the field of MRI of the spine, and did so by performing clinically relevant benchmark cases. The interest in generative models, which are computer programs that are able to generate novel data, as opposed to classifying or processing existing data, is due to the fact that considerable technological innovations have been able to improve their performance, thus creating more applications for this technology within MRI. Through the tests conducted by the authors, they were able to conclude that deep learning-based generative methods have the potential to make an impact on the future of musculoskeletal radiology.
“Generative models aimed at creating high-quality images; for example, photorealistic faces of non-existing people have made striking advances in recent times,” say authors Dr. Fabio Galbusera and Dr. Luca Maria Sconfienza from Milan. “We believe that such models will find practical applications in musculoskeletal radiology in the near future, such as synthetic re-slicing for high-quality visualization in the non-acquired orientations and the automated translation between different imaging modalities. Clinical routine may benefit from shorter acquisition times for MR examinations with generative models filling the gaps.”
Key Points:
- Deep learning-based generative models are able to generate convincing synthetic images of the spine
- Generative models provide a promising improvement of the level of detail in MRI images of the spine, with limitations requiring further research
- The availability of large radiological datasets is a key factor in improving the performance of deep learning models
Authors: Fabio Galbusera, Tito Bassani, Gloria Casaroli, Salvatore Gitto, Edoardo Zanchetta, Francesco Costa and Luca Maria Sconfienza