New Model Improves Comparison of MRIs Taken at Different Institutions

Magnetic resonance imaging (MRI) plays a crucial role in medicine by offering detailed views of the internal structures of the body and providing valuable insights into various pathologies. However, inconsistencies in image acquisition protocols across different institutions create significant challenges in ensuring reliable and consistent interpretation, especially in multi-center research. To address this issue, a new study has introduced a method for modifying MRI scans from different hospitals, making them more comparable and enabling more accurate and dependable comparisons.

Harmonizing MRI results is a fundamental issue in both research and healthcare quality. Each medical facility, clinic, or research institution uses its own set of imaging protocols, equipment, and parameters, leading to variations in contrast, brightness, and other characteristics of the images. This variability presents a major challenge in clinical research, particularly when data from multiple research centers are combined. In a collaborative study led by researchers from Université de Montreal (Montreal, Canada), a new harmonization technique was developed, which involves three main steps. The first step is to create a model that learns how MRI images in the source domain.

Image: Researchers have come up with a better way to compare MRI images taken at different institutions (Photo courtesy of Adobe Stock)

Once the distribution of the source domain is understood, the next goal is to "re-format" MRI scans from other institutions to remove variations caused by differences in machine settings or parameter choices, while still maintaining patient-specific differences. Finally, when the model is applied to new images (for example, from an unfamiliar machine), it must adapt to ensure that the newly processed images continue to follow the learned distribution. To validate their approach, the researchers tested the harmonization method on MRI brain scans from databases in the United States, as well as from a neonatal imaging consortium developed with Australian researchers. These data were used for two key tasks: first, to segment brain images in both adults and newborns to verify if brain structures remained consistent after harmonization, and second, to estimate the brain age of newborns.

The study, published in Medical Image Analysis, demonstrated that this new technique outperforms existing harmonization methods, showcasing its versatility across various tasks and population groups. Notably, the method was successfully validated on MRI scans of a newborn’s brain with lesions—something that all other available models fail to handle since they are typically trained on images of healthy brains. Looking ahead, the researchers plan to apply this approach on a larger scale in future collaborations and studies, which will help improve the comparison and analysis of research data and further enhance the accuracy and reliability of medical diagnoses.

“Thanks to this model, we can now interpret data from several thousands of families and children who are monitored at various hospitals – data that come from different scanners," said Dr. Gregory Lodygensky, a clinical professor at Université de Montreal. "The analysis of these large cohorts in children and adults was hampered by the major harmonization problem, which has now been resolved.”