Boston College researchers have developed an artificial intelligence (AI)-assisted method designed to enhance the clarity of functional magnetic resonance imaging (fMRI) data. This method, detailed in a recent report in Nature Methods, aims to mitigate image distortions, referred to as "noise," caused by factors such as patient movement and cardiac activity.
fMRI is a widely utilized noninvasive technique in neuroscience, with numerous studies published annually. A known challenge in fMRI research involves the presence of noise within MRI data, which can obscure brain responses.
DeepCor Method
Associate Professor of Psychology Stefano Anzellotti, the senior author of the paper, stated that improving noise removal could facilitate new discoveries in brain research. The new method, named DeepCor, was developed by Anzellotti, post-doctoral researcher Aidas Aglinskas, and then-undergraduate student Yu Zhu. DeepCor employs generative AI to enhance noise reduction capabilities.
Anzellotti indicated that the method achieved an improvement of over 200 percent compared to previous approaches.
Performance Metrics
DeepCor demonstrated superior performance against other state-of-the-art denoising techniques across various simulated datasets. In evaluations using real fMRI data, DeepCor exhibited a 215 percent improvement over CompCor, a commonly used method, in removing noise associated with face responses. It also showed a 339 percent improvement in clarifying realistic synthetic data designed to mimic actual fMRI datasets.
AI Mechanism
Anzellotti explained that the AI within DeepCor learns to differentiate between unique patterns found in brain regions containing neurons and those in regions without neurons, such as the ventricles. By identifying and removing common noise patterns affecting both types of regions, the method enhances the distinctiveness of neuronal patterns.
Future Implications
The researchers noted that the extent of improvement achieved, exceeding 200 percent, surpassed initial expectations. Future work will focus on making DeepCor accessible to a broader research community and applying it to denoise existing large public datasets to provide clearer data for the field.