Accurate whole-brain segmentation, or brain extraction, of magnetic resonance imaging (MRI) is a critical first step in most neuroimage analysis pipelines. The majority of brain extraction algorithms have been developed and evaluated for adult data and their validity for neonatal brain extraction, which presents age-specific challenges for this task, has not been established. We developed a novel method for brain extraction of multi-modal neonatal brain MR images, named ALFA (Accurate Learning with Few Atlases). The method uses a new sparsity-based atlas selection strategy that requires a very limited number of atlases ‘uniformly’ distributed in the low-dimensional data space, combined with a machine learning based label fusion technique. The performance of the method for brain extraction from multi-modal data of 50 newborns is evaluated and compared with results obtained using eleven publicly available brain extraction methods. ALFA outperformed the eleven compared methods providing robust and accurate brain extraction results across different modalities. As ALFA can learn from partially labelled datasets, it can be used to segment large-scale datasets efficiently. ALFA could also be applied to other imaging modalities and other stages across the life course.

Neuroimage analysis pipelines rely on parcellated atlases generated from healthy individuals to provide anatomic context to structural and diffusion MRI data. Atlases constructed using adult data introduce bias into studies of early brain development. We aimed to create a neonatal brain atlas of healthy subjects that can be applied to multi-modal MRI data. Structural and diffusion 3T MRI scans were acquired soon after birth from 33 typically developing neonates born at term (mean postmenstrual age at birth 39+5 weeks, range 37+2–41+6). An adult brain atlas (SRI24/TZO) was propagated to the neonatal data using temporal registration via childhood templates with dense temporal samples (NIH Pediatric Database), with the final atlas (Edinburgh Neonatal Atlas, ENA33) constructed using the Symmetric Group Normalization (SyGN) method. After this step, the computed final transformations were applied to T2-weighted data, and fractional anisotropy, mean diffusivity, and tissue segmentations to provide a multi-modal atlas with 107 anatomical regions; a symmetric version was also created to facilitate studies of laterality. Volumes of each region of interest were measured to provide reference data from normal subjects. Because this atlas is generated from step-wise propagation of adult labels through intermediate time points in childhood, it may serve as a useful starting point for modeling brain growth during development.

Deep convolutional neural networks (CNNs) have been intensively used for multi-class segmentation of data from different modalities and achieved state-of-the-art performances. However, a common problem when dealing with large, high resolution 3D data is that the volumes input into the deep CNNs has to be either cropped or downsampled due to limited memory capacity of computing devices. These operations lead to loss of resolution and increment of class imbalance in the input data batches, which can downgrade the performances of segmentation algorithms. Inspired by the architecture of image super-resolution CNN (SRCNN) and self-normalization network (SNN), we developed a two-stage modified Unet framework that simultaneously learns to detect a ROI within the full volume and to classify voxels without losing the original resolution. Experiments on a variety of multi-modal volumes demonstrated that, when trained with a simply weighted dice coefficients and our customized learning procedure, this framework shows better segmentation performances than state-of-the-art Deep CNNs with advanced similarity metrics.

Background Standard methods for quantifying positron emission tomography (PET) uptake in the aorta are time consuming and may not reflect overall vessel activity. We describe aortic microcalcification activity (AMA), a novel method for quantifying 18F-sodium fluoride (18F-NaF) uptake in the thoracic aorta. Methods Twenty patients underwent two hybrid 18F-NaF PET and computed tomography (CT) scans of the thoracic aorta less than three weeks apart. AMA, as well as maximum (TBRmax) and mean (TBRmean) tissue to background ratios, were calculated by two trained operators. Intra-observer repeatability, inter-observer repeatability and scan-rescan reproducibility were assessed. Each 18F-NaF quantification method was compared to validated cardiovascular risk scores. Results Aortic microcalcification activity demonstrated excellent intra-observer (intraclass correlation coefficient 0.98) and inter-observer (intraclass correlation coefficient 0.97) repeatability with very good scan-rescan reproducibility (intraclass correlation coefficient 0.86) which were similar to previously described TBRmean and TBRmax methods. AMA analysis was much quicker to perform than standard TBR assessment (3.4min versus 15.1min, P<0.0001). AMA was correlated with Framingham stroke risk scores and Framingham risk score for hard cononary heart disease. Conclusions AMA is a simple, rapid and reproducible method of quantifying global 18F-NaF uptake across the ascending aorta and aortic arch that correlates with cardiovascular risk scores.