Detailed whole brain segmentation is an essential quantitative technique in medical image analysis, which provides a non-invasive way of measuring brain regions from a clinical acquired structural magnetic resonance imaging (MRI). Recently, deep convolution neural network (CNN) has been applied to whole brain segmentation. However, restricted by current GPU memory, 2D based methods, downsampling based 3D CNN methods, and patch-based high-resolution 3D CNN methods have been the de facto standard solutions. 3D patch-based high resolution methods typically yield superior performance among CNN approaches on detailed whole brain segmentation (>100 labels), however, whose performance are still commonly inferior compared with state-of-the-art multi-atlas segmentation methods (MAS) due to the following challenges: (1) a single network is typically used to learn both spatial and contextual information for the patches, (2) limited manually traced whole brain volumes are available (typically less than 50) for training a network. In this work, we propose the spatially localized atlas network tiles (SLANT) method to distribute multiple independent 3D fully convolutional networks (FCN) for high-resolution whole brain segmentation. To address the first challenge, multiple spatially distributed networks were used in the SLANT method, in which each network learned contextual information for a fixed

Skilled reading depends on recognizing words efficiently in isolation (word-level processing; WL) and extracting meaning from text (discourse-level processing; DL); deficiencies in either result in poor reading. FMRI has revealed consistent overlapping networks in word and passage reading, as well as unique regions for DL processing, however less is known about how WL and DL processes interact. Here we examined functional connectivity from seed regions derived from where BOLD signal overlapped during word and passage reading in 38 adolescents ranging in reading ability, hypothesizing that even though certain regions support word- and higher-level language, connectivity patterns from overlapping regions would be task modulated. Results indeed revealed that the left-lateralized semantic and working memory (WM) seed regions showed task-dependent functional connectivity patterns: during DL processes, semantic and WM nodes all correlated with the left angular gyrus, a region implicated in semantic memory/coherence building. In contrast, during WL, these nodes coordinated with a traditional WL area (left occipitotemporal region). Additionally, these WL and DL findings were modulated by decoding and comprehension abilities, respectively, with poorer abilities correlating with decreased connectivity. Findings indicate that key regions may uniquely contribute to multiple levels of reading; we speculate that these connectivity patterns may be especially salient for reading outcomes and intervention response.

BackgroundThere is a substantial literature on the neurobiology of reading and dyslexia. Differences are often described in terms of individual regions or individual cognitive processes. However, there is a growing appreciation that the brain areas subserving reading are nested within larger functional systems, and new network analysis methods may provide greater insight into how reading difficulty arises. Yet, relatively few studies have adopted a principled network-based approach (e.g., connectomics) to studying reading. In this study, we combine data from previous reading literature, connectomics studies, and original data to investigate the relationship between network architecture and reading.MethodsFirst, we detailed the distribution of reading-related areas across many resting-state networks using meta-analytic data from NeuroSynth. Then, we tested whether individual differences in modularity, the brain’s tendency to segregate into resting-state networks, are related to reading skill. Finally, we determined whether brain areas that function atypically in dyslexia, as identified by previous meta-analyses, tend to be concentrated in hub regions.ResultsWe found that most resting-state networks contributed to the reading network, including those subserving domain-general cognitive skills such as attention and executive function. There was also a positive relationship between the global modularity of an individual’s brain network and reading skill, with the visual, default mode and cingulo-opercular networks showing the highest correlations. Brain areas implicated in dyslexia were also significantly more likely to have a higher participation coefficient (connect to multiple resting-state networks) than other areas.ConclusionsThese results contribute to the growing literature on the relationship between reading and brain network architecture. They suggest that an efficient network organization, i.e., one in which brain areas form cohesive resting-state networks, is important for skilled reading, and that dyslexia can be characterized by abnormal functioning of hub regions that map information between multiple systems. Overall, use of a connectomics framework opens up new possibilities for investigating reading difficulty, especially its commonalities across other neurodevelopmental disorders.Electronic supplementary materialThe online version of this article (10.1186/s11689-018-9251-z) contains supplementary material, which is available to authorized users.

A primary challenge facing the development of interventions for dyslexia is identifying effective predictors of intervention response. While behavioral literature has identified core cognitive characteristics of response, the distinction of reading versus executive cognitive contributions to response profiles remains unclear, due in part to the difficulty of segregating these constructs using behavioral outputs. In the current study we used functional neuroimaging to piece apart the mechanisms of how/whether executive and reading network relationships are predictive of intervention response. We found that readers who are responsive to intervention have more typical pre-intervention functional interactions between executive and reading systems compared to nonresponsive readers. These findings suggest that intervention response in dyslexia is influenced not only by domain-specific reading regions, but also by contributions from intervening domain-general networks. Our results make a significant gain in identifying predictive bio-markers of outcomes in dyslexia, and have important implications for the development of personalized clinical interventions.