Search

ObjectivesBrain segmentation of infant magnetic resonance (MR) images is vitally important in studying developmental mental health and disease. The infant brain undergoes many changes throughout the first years of postnatal life, making tissue segmentation difficult for most existing algorithms. Here, we introduce a deep neural network BIBSNet (Baby andInfantBrainSegmentation NeuralNetwork), an open-source, community-driven model that relies on data augmentation and a large sample size of manually annotated images to facilitate the production of robust and generalizable brain segmentations.Experimental DesignIncluded in model training and testing were MR brain images on 84 participants with an age range of 0-8 months (median postmenstrual ages of 13.57 months). Using manually annotated real and synthetic segmentation images, the model was trained using a 10-fold cross-validation procedure. Testing occurred on MRI data processed with the DCAN labs infant-ABCD-BIDS processing pipeline using segmentations produced from gold standard manual annotation, joint-label fusion (JLF), and BIBSNet to assess model performance.Principal ObservationsUsing group analyses, results suggest that cortical metrics produced using BIBSNet segmentations outperforms JLF segmentations. Additionally, when analyzing individual differences, BIBSNet segmentations perform even better.ConclusionsBIBSNet segmentation shows marked improvement over JLF segmentations across all age groups analyzed. The BIBSNet model is 600x faster compared to JLF and can be easily included in other processing pipelines.

BackgroundMagnetic resonance spectroscopy (MRS) can non-invasively measure levels of endogenous metabolites in living tissue and is of great interest to neuroscience and clinical research. To this day, MRS data analysis workflows differ substantially between groups, frequently requiring many manual steps to be performed on individual datasets, e.g., data renaming/sorting, manual execution of analysis scripts, and manual assessment of success/failure. Manual analysis practices are a substantial barrier to wider uptake of MRS. They also increase the likelihood of human error and prevent deployment of MRS at large scale. Here, we demonstrate an end-to-end workflow for fully automated data uptake, processing, and quality review.New MethodThe proposed continuous automated MRS analysis workflow integrates several recent innovations in MRS data and file storage conventions. They are efficiently deployed by a directory monitoring service that automatically triggers the following steps upon arrival of a new raw MRS dataset in a project folder: (1) conversion from proprietary manufacturer file formats into the universal format NIfTI-MRS; (2) consistent file system organization according to the data accumulation logic standard BIDS-MRS; (3) executing a command-line executable of our open-source end-to-end analysis software Osprey; (4) e-mail delivery of a quality control summary report for all analysis steps.ResultsThe automated architecture successfully completed for a demonstration dataset. The only manual step required was to copy a raw data folder into a monitored directory.Comparison with Existing Method(s)The workflow presented here is the first implementation of a continuous automated MRS analysis ecosystem based on NIfTI-MRS and BIDS-MRS standards. Traditional MRS workflows are non-standardized, often require manual input, and frequently noy compatible with established neuroimaging workflows. This workflow should therefore facilitate integration of MRS into large-scale and multi-center studies.ConclusionsContinuous automated analysis of MRS data can reduce the burden of manual data analysis and quality control, particularly for non-expert users and multi-center or large-scale studies.