Technology Research

The Mayo Clinic Radiology Informatics Lab develops educational resources, new technology and advanced workflows to drive discovery and innovation in radiology informatics.

Educational tools to learn and use AI in radiology
Deep learning and artificial intelligence (AI) are growing fields in radiology. Many new researchers need resources for best practices and tutorials for AI. The Radiology Informatics Lab is developing the necessary materials and resources.

Fast MRI reconstruction with superresolution
MRI is a widely used noninvasive form of medical imaging. It can help make clinical diagnoses and support research. But slow acquisition speed often limits these benefits. Reducing acquisition time would make the procedure easier on patients, reduce examination costs and decrease motion-induced artifacts.

The lab is developing new deep-learning models to achieve state-of-the-art quality acquisitions through both intraslice and interslice orientations. These models could be applied to clinical diagnoses and radiographic study.

MIDeL
MIDeL helps healthcare professionals and medical-imaging scientists use deep-learning technologies for medical images. It is an extensive electronic textbook that seamlessly combines rich textual content with practical code examples. This approach creates a learning environment that allows users to understand how to apply deep-learning techniques in medical imaging.

Learn more on the MIDeL GitHub page.

Noise reduction AI models for molecular breast imaging
The lab is developing AI models to help lower the radiation dose of molecular breast imaging and shorten scan time. Currently, shorter scan times reduce the signal-to-noise ratio in molecular breast imaging, making images harder to evaluate and making it easier to miss cancers. The research team is finding techniques to remove noise that can boost the signal-to-noise ratio, leading to more accurate diagnoses with less exposure to radiation.

Synthetic data generation to improve fairness and reduce bias
Synthetic image data may be able to generate viable images of people from underrepresented groups. This will help ensure that diagnostic AI models that are trained on research data plus synthetic data are more accurate and better represent all people.

Tabular medical data imputation
The Radiology Informatics Lab is developing a model to impute missing data in large medical tabular datasets.

Using topological data analysis in medical-imaging research
Topological data analysis is revolutionizing medical imaging, offering nuanced insights into complex data structures. The lab is developing topological data analysis resources for aspiring researchers, including specialized courses, practical software tools and community forums. These resources will contribute to a full understanding and application of topological data analysis and advance medical-imaging technologies and methodologies.

Workflow engine to improve data retrieval and curation in multimodal research studies
Dr. Erickson's lab has created a workflow engine and library of modules that allow researchers to specify whether humans are involved at fixed steps. The team also added new technology to help researchers arrange multimodal data into research cohorts to train AI models and describe major design decisions.