An AI-based segmentation and analysis pipeline for high-field MR monitoring of cerebral organoids

Abstract

BackgroundCerebral organoids simulate the structure and function of the developing human brain in vitro, offering a large potential for personalized therapeutic strategies. The enormous growth of this research area over the past decade with its capability for clinical translation makes a non-invasive, automated analysis pipeline of organoids highly desirable. PurposeThis work presents the first application of magnetic resonance imaging (MRI) for the non-invasive quantification and quality assessment of cerebral organoids using an automated analysis tool. Three specific objectives are addressed, namely organoid segmentation to investigate organoid development over time, global cysticity classification, and local cyst segmentation. MethodsNine wildtype cerebral organoids were imaged over nine weeks using high-field 9.4T MRI including a 3D T2*-w and 2D diffusion tensor imaging (DTI) sequence. This dataset was used to train a deep learning-based 3D U-Net for organoid and local cyst segmentation. For global cysticity classification, we developed a new metric, compactness, to separate low- and high-quality organoids. ResultsThe 3D U-Net achieved a Dice score of 0.92{+/-}0.06 (mean {+/-} SD) for organoid segmentation in the T2*-w sequence. For global cysticity classification, compactness separated low- and high-quality organoids with high accuracy (ROC AUC 0.98). DTI showed that low-quality organoids have a significantly higher diffusion than high-quality organoids (p < .001). For local cyst segmentation in T2*-w, the 3D U-Net achieved a Dice score of 0.63{+/-}0.15 (mean {+/-} SD). ConclusionWe present a novel non-invasive approach to monitor and analyze cerebral organoids over time using high-field MRI and state-of-the-art tools for automated image analysis, offering a comparative pipeline for personalized medicine. We show that organoid growth can be monitored reliably over time and low- and high-quality organoids can be separated with high accuracy. Local cyst segmentation is feasible but could be further improved in the future.