Reduced gadolinium dose in brain tumor imaging through deep learning model
Posted on: 2020-12-18
Do you think that deep learning can be used to produce high-quality synthetic post-contrast MRI brain images without the use of gadolinium-based contrast agents? That was the question posed at a presentation at RSNA by Gowtham Murugesan, PhD, Department of Radiology, University of Texas Southwestern Medical Center, Dallas – and, according to the presentation, gadolinium-free exams using deep learning are feasible.
Gadolinium is responsible for shortening T1 relaxation in tissues where the agent accumulates, resulting in a brighter signal from them in T1-weighted post-contrast images. In addition, gadolinium increases tissue contrast, accentuating areas where contrast agents have spilled over the blood-brain barrier.
However, a warning regarding gadolinium retention was issued by the U.S. Federal Drug Administration in 2018 and a research showed evidence that gadolinium can accumulate in various non-target tissues (including the brain, bone and kidneys).
Therefore, the researchers in the study we are talking about sought to demonstrate the feasibility of models and methods of “deep learning” to generate post-contrast T1 images using non-contrast magnetic resonance images in patients with primary brain tumor. The model used a 10% gadolinium dose to create contrast-enhanced brain images that could be used to predict full-dose images.
The data set included 335 patients from the Multimodal Brain Tumor Segmentation Challenge (BraTS) 2019 who were used to train the model, while a set of 125 BraTS 2019 patients was used as test data. The qualitative assessment was carried out by two radiologists and, according to their consensus classifications, the model was able to synthesize the enhancement by contrast with excellent, good and bad results in 49, 61 and 15 cases, respectively.
Although the proposed deep learning model is not ready for clinical translation, the study demonstrates the feasibility of using deep learning methods to synthesize gadolinium images without contrast.