American Association for Cancer Research, Abstract 2344: Federated image quality assessment of prostate MRI scans in a multi-institutional setting

The reliability of machine and deep learning models on medical images can be compromised by image artifacts and variability in acquisition, particularly when validating performance across institutions with diverse protocols and scanner equipment, leading to significant numbers of outlier scans. To address this, a privacy-preserving outlier identification algorithm was developed using federated MR imaging quality evaluation on a distributed, multi-institutional cohort of prostate MRI scans. T2-weighted scans from one public collection (template) and three federated institutions were analyzed using the MRQy tool, implemented via Docker within the RhinoHealth Federated Computing Platform. This setup enabled remote extraction of nine metadata values and 15 quality measurements per scan while maintaining patient confidentiality. A rule-based classifier flagged scans as outliers if more than 60% of their quality attributes fell outside predefined bounds, derived from the template cohort. Among 630 MRI scans analyzed, the outlier rates varied significantly across institutions (3%, 19%, and 79%), with Institution 3 showing substantial intensity and resolution differences requiring extensive post-processing. These findings demonstrate the potential of federated image quality assessment for privacy-preserving curation of multi-institutional cohorts, enhancing the consistency of data for downstream machine learning tasks. Further studies will assess the impact of quality-based curation on federated machine learning applications.