A robust and reproducible automated MRI pipeline for quantifying tissue outcomes after experimental stroke in...

The failure to translate promising preclinical stroke therapies into clinical success is a multi-faceted problem; however, a critical contributing factor is the lack of rigorous, reproducible preclinical outcome measures. While magnetic resonance imaging (MRI) offers a translational alternative to traditional histology, its use in large, multi-site trials is challenged by data heterogeneity and the need for scalable analysis. To address this, we developed and validated a fully automated, open-source image analysis pipeline for the Stroke Preclinical Assessment Network (SPAN), a six-center preclinical trial network. The pipeline processed T2-weighted and apparent diffusion coefficient (ADC) maps from over 2,000 mice and rats, incorporating steps for cross-site data harmonization, deep learning-based brain extraction, and rule-based segmentation to quantify infarct volume, brain swelling, and atrophy. The pipeline demonstrated high accuracy, as automated lesion volumes strongly correlated with manual expert tracing on both MRI (R = 0.96) and 2,3,5-triphenyl-tetrazolium chloride (TTC)-stained tissue (R = 0.86). The U-net model for brain extraction achieved a Dice score of 0.96, and our harmonization method successfully reduced inter-site variability in quantitative MRI parameters. This robust and reproducible pipeline provides a scalable framework for standardizing tissue outcome assessment, enhancing the rigor of multi-site preclinical studies.