Developing a brain atlas through deep learning

Abstract

Neuroscientists have devoted significant effort into the creation of standard brain reference atlases for high-throughput registration of anatomical regions of interest. However, variability in brain size and form across individuals poses a significant challenge for such reference atlases. To overcome these limitations, we introduce a fully automated deep neural networkbased method (SeBRe) for registration through Segmenting Brain Regions of interest with minimal human supervision. We demonstrate the validity of our method on brain images from different mouse developmental time points, across a range of neuronal markers and imaging modalities. We further assess the performance of our method on images from MR-scanned human brains. Our registration method can accelerate brain-wide exploration of region-specific changes in brain development and, by simply segmenting brain regions of interest for highthroughput brain-wide analysis, provides an alternative to existing complex brain registration techniques.The mean AP scores achieved for brain sections in the original (P14) and the extended (P4, P14, P56) test datasets are 0.84 and 0.87, respectively. Performance of the network is optimal in most brain regions, including isocortex, hippocampus, basal ganglia, telencephalic vesicle and the midbrain. For segmenting the pre-thalamus, the performance of the network is limited due to a large variation in the structure and size of this region, as we move from lateral to medial images, across the sagittal plane.Performance of SeBRe on mouse brain images labelled with previously "unseen" markers and ages. After training and testing the performance of our network on P14 brain sections, we further checked the scalability of our method by testing it on various mouse brain images from different developmental time points (P4, P14, P28 and P56), for commonly used neuronal markers that the network had not been trained on (CaMKIIa, Nissl, GAD1 and VGAT) [11]. Ca 2+ /calmodulin-dependent protein kinase IIa (CaMKIIa) labels the largest population of

Approximate and Situated Causality in Deep Learning

“…It is beyond any doubt that DL is contributing to improving the knowledge in several areas, some of them very difficult to interpret because of the nature of obtained data, like neuroscience [39]. These advances are expanding the frontiers of verifiable knowledge beyond classic human standards.…”

DeepBrainSeg: Automated Brain Region Segmentation for Micro-Optical Images With a Convolutional Neural Network

“…While CNN methods can learn local and contextual features, they have difficulty utilizing global location features from the whole-brain range at high resolution, resulting in over-segmentation for other regions with similar local features. To locate brain structures, Iqbal et al (2019) segmented and classified the mouse brain into eight regions using Mask r-cnn (He et al, 2017), while the detected box has excessive redundancies for the region with complex shape. Chen et al (2019) combined a patchbased CNN and registration to segment the murine brainstem, whereas the accuracy of segmentation is easily affected by the effect of registration.…”

Bi-channel Image Registration and Deep-learning Segmentation (BIRDS) for efficient, versatile 3D mapping of mouse brain

“…Large projects which include the Mouse Brain Architecture project [4], the Allen Mouse Brain Connectivity Atlas [5] and the Mouse Connectome project map the mouse brain [6] in terms of cell types, long-range connectivity patterns and microcircuit connectivity. In addition to the large-scale collaborative, currently an increasing number of laboratories are also developing independent automated, or semi-automated framework for processing brain data obtained in specific projects [7][8][9][10][11][12]. With the improvement of experimental methods for dissection of connectivity and function, development of a standardized and automated computational pipeline to map, analyze, visualize and share brain data has become the major challenge to all such efforts [1,7].…”

“…Though some frameworks can register certain type of brain slab that contains complete coronal outlines in a way of slice by slice [7,[23][24][25], it remains very difficult to register small brain block without obvious anatomical outlines. As neural network has emerged as a technique of choice for image processing [26][27][28][29], deep learning-based brain mapping methods are also recently reported to directly provide segmentation/annotation of primary regions for 3D brain dataset [12,[30][31][32][33]. Such deep-learning-based segmentation networks are efficient in extracting pixel-level features, and thus are little dependent on the presence of global features, such as complete anatomical outlines, making it better suited for processing incomplete brain data, as compared to registration-based methods.…”