3D volume segmentation and reconstruction. Supervised image classification and automated quantification of superparamagnetic iron oxide nanoparticles in histology slides for safety assessment

This article presents an automated image-processing workflow for quantitative assessment of SPION accumulation in tissue sections stained with Prussian blue for iron detection. We utilized supervised machine learning with manually labeled features used for training the classifier. Performance of the classifier was validated by 10-fold cross-validation of obtained data and by measuring Dice and Jaccard Similarity Coefficients between manually segmented image and automated segmentation. The proposed approach provides time and cost-effective solution for quantitative imaging analysis of SPION in tissue with a precision similar to that obtained via thresholding method for stain quantification. Furthermore, we exploited the classifiers to generate segmented 3D volumes from histological slides. This enabled visualization of particles which were obscured in original 3D histology stacks. Our approach offers a powerful tool for preclinical assessment of the precise tissue-specific SPION biodistribution, which could affect both their toxicity and their efficacy as nanocarriers for medicines.