IRMA: Machine Learning for PET Brain Scan Harmonization

The IRMA method, introduced in a new paper, uses machine learning to harmonize 18F-FDG PET brain scans from different centers, reducing bias in diagnostic models.

The paper ‘IRMA: Machine learning-based harmonization of 18F-FDG PET brain scans in multi-center studies’ by Sofie Lövdal et al. has been published in the European Journal of Nuclear Medicine and Molecular Imaging.

This research introduces IRMA (Iterated Relevance Matrix Analysis), a method to mitigate bias in PET feature vectors from different medical centers. The study demonstrates that the center origin of healthy control brain images can be identified with high confidence, which can impact the accuracy of machine learning models in diagnosing neurodegenerative disorders.

The related code for IRMA is freely available at GitHub.

Tags: IRMA, PET brain scans, machine learning, bias mitigation, neurodegenerative disorders, multi-center studies