Abstract The black box nature of deep neural networks (DNNs) makes researchers and clinicians hesitant to rely on their findings. Saliency maps can enhance DNN explainability by suggesting the anatomic localization of relevant brain features. This study compares seven popular attribution-based saliency approaches to assign neuroanatomic interpretability to DNNs that estimate biological brain age (BA) from magnetic resonance imaging (MRI). Cognitively normal (CN) adults ( N = 13,394, 5,900 males; mean age: 65.82 ± 8.89 years) are included for DNN training, testing, validation, and saliency map generation to estimate BA. To study saliency robustness to the presence of anatomic deviations from normality, saliency maps are also generated for adults with mild traumatic brain injury (mTBI, $$N$$ N = 214, 135 males; mean age: 55.3 ± 9.9 years). We assess saliency methods’ capacities to capture known anatomic features of brain aging and compare them to a surrogate ground truth whose anatomic saliency is known a priori. Anatomic aging features are identified most reliably by the integrated gradients method, which outperforms all others through its ability to localize relevant anatomic features. Gradient Shapley additive explanations, input × gradient, and masked gradient perform less consistently but still highlight ubiquitous neuroanatomic features of aging (ventricle dilation, hippocampal atrophy, sulcal widening). Saliency methods involving gradient saliency, guided backpropagation, and guided gradient-weight class attribution mapping localize saliency outside the brain, which is undesirable. Our research suggests the relative tradeoffs of saliency methods to interpret DNN findings during BA estimation in typical aging and after mTBI.
Severe loss of excitatory synapses in key brain regions is thought to be one of the major mechanisms underlying stress-induced cognitive impairment. To date, however, the identity of the affected circuits remains elusive. Here we examined the effect of exposure to repeated multiple concurrent stressors (RMS) on the connectivity of the posterior parietal cortex (PPC) in adolescent male mice. We found that RMS led to layer-specific elimination of excitatory synapses with the most pronounced loss observed in deeper cortical layers. Quantitative analysis of cortical projections to the PPC revealed a significant loss of sensory and retrosplenial inputs to the PPC while contralateral and frontal projections were preserved. These results were confirmed by decreased synaptic strength from sensory, but not from contralateral, projections in stress-exposed animals. Functionally, RMS disrupted visuospatial working memory performance, implicating disrupted higher-order visual processing. These effects were not observed in mice subjected to restraint-only stress for an identical period of time. The PPC is considered to be a cortical hub for multisensory integration, working memory, and perceptual decision-making. Our data suggest that sensory information streams targeting the PPC may be impacted by recurring stress, likely contributing to stress-induced cognitive impairment. SIGNIFICANCE STATEMENT Repeated exposure to stress profoundly impairs cognitive functions like memory, attention, or decision-making. There is emerging evidence that stress not only impacts high-order regions of the brain, but may affect earlier stages of cognitive processing. Our work focuses on the posterior parietal cortex, a brain region supporting short-term memory, multisensory integration, and decision-making. We show evidence that repeated stress specifically damages sensory inputs to this region. This disruption of synaptic connectivity is linked to working memory impairment and is specific to repeated exposure to multiple stressors. Altogether, our data provide a potential alternative explanation to ailments previously attributed to downstream, cognitive brain structures.
Abstract In recent years, a growing number of spatial epigenome datasets have been generated, presenting rich opportunities for studying the regulation mechanisms in solid tissue sections. However, visual exploration of these datasets requires extensive computational processing of raw data, presenting a challenge for researchers without advanced computational skills to fully explore and analyze such datasets. Here we introduce AtlasXplore™, a web-based platform that enables scientists to interactively navigate a growing collection of spatial epigenome data using an expanding set of tools. Availability and implementation AtlasXplore is freely available at https://web.atlasxomics.com
Abstract Motivation In recent years, a growing number of spatial epigenome datasets have been generated, presenting rich opportunities for studying the regulation mechanisms in solid tissue sections. However, visual exploration of these datasets requires extensive computational processing of raw data, presenting a challenge for researchers without advanced computational skills to fully explore and analyze such datasets. Results Here, we introduce AtlasXplore, a web-based platform that enables scientists to interactively navigate a growing collection of spatial epigenome data using an expanding set of tools. Availability and implementation https://web.atlasxomics.com
Abstract Precise cortical brain localization presents an important challenge in the literature. Brain atlases provide data-guided parcellation based on functional and structural brain metrics, and each atlas has its own unique benefits for localization. We offer a parcellation guided by intracranial electroencephalography, a technique which has historically provided pioneering advances in our understanding of brain structure–function relationships. We used a consensus boundary mapping approach combining anatomical designations in Duvernoy’s Atlas of the Human Brain, a widely recognized textbook of human brain anatomy, with the anatomy of the MNI152 template and the magnetic resonance imaging scans of an epilepsy surgery cohort. The Yale Brain Atlas consists of 690 one-square centimeter parcels based around conserved anatomical features and each with a unique identifier to communicate anatomically unambiguous localization. We report on the methodology we used to create the Atlas along with the findings of a neuroimaging study assessing the accuracy and clinical usefulness of cortical localization using the Atlas. We also share our vision for the Atlas as a tool in the clinical and research neurosciences, where it may facilitate precise localization of data on the cortex, accurate description of anatomical locations, and modern data science approaches using standardized brain regions.
Abstract The black box nature of deep neural networks (DNNs) makes researchers and clinicians hesitant to rely on their findings. Saliency maps can enhance DNN explainability by suggesting the anatomic localization of relevant brain features. This study compares seven popular attribution-based saliency approaches to assign neuroanatomic interpretability to DNNs that estimate biological brain age (BA) from magnetic resonance imaging (MRI). Cognitively normal (CN) adults (N = 13,394, 5,900 males; mean age: 65.82 ± 8.89 years) are included for DNN training, testing, validation, and saliency map generation to estimate BA. To study saliency robustness to the presence of anatomic deviations from normality, saliency maps are also generated for adults with mild traumatic brain injury (mTBI, \(\:N\) = 214, 135 males; mean age: 55.3 ± 9.9 years). We assess saliency methods’ capacities to capture known anatomic features of brain aging and compare them to a surrogate ground truth whose anatomic saliency is known a priori. Anatomic aging features are identified most reliably by the integrated gradients method, which outperforms all others through its ability to localize relevant anatomic features. Gradient Shapley additive explanations, input × gradient, and masked gradient perform less consistently but still highlight ubiquitous neuroanatomic features of aging (ventricle dilation, hippocampal atrophy, sulcal widening). Saliency methods involving gradient saliency, guided backpropagation, and guided gradient-weight class attribution mapping localize saliency outside the brain, which is undesirable. Our research suggests the relative tradeoffs of saliency methods to interpret DNN findings during BA estimation in typical aging and after mTBI.
'%3e%3cpath fill='%23012169' d='M0 0v30h60V0z'/%3e%3cpath stroke='%23fff' stroke-width='6' d='m0 0 60 30m0-30L0 30'/%3e%3cpath stroke='%23C8102E' stroke-width='4' d='m0 0 60 30m0-30L0 30' clip-path='url(%23b)'/%3e%3cpath stroke='%23fff' stroke-width='10' d='M30 0v30M0 15h60'/%3e%3cpath stroke='%23C8102E' stroke-width='6' d='M30 0v30M0 15h60'/%3e%3c/g%3e%3c/svg%3e)