Prediction of MYCN Gene Amplification in Pediatric Neuroblastomas: Development of a Deep Learning-Based Tool for Automatic Tumor Segmentation and Comparative Analysis of Computed Tomography-Based Radiomics Features Harmonization.

OBJECTIVE: MYCN oncogene amplification is closely linked to high-grade neuroblastoma with poor prognosis. Accurate quantification is essential for risk assessment, which guides clinical decision making and disease management. This study proposes an end-to-end deep-learning framework for automatic tumor segmentation of pediatric neuroblastomas and radiomics features-based classification of MYCN gene amplification. METHODS: Data from pretreatment contrast-enhanced computed tomography scans and MYCN status from 47 cases of pediatric neuroblastomas treated at a tertiary children's hospital from 2009 to 2020 were reviewed. Automated tumor segmentation and grading pipeline includes (1) a modified U-Net for tumor segmentation; (2) extraction of radiomic textural features; (3) feature-based ComBat harmonization for removal of variabilities across scanners; (4) feature selection using 2 approaches, namely, ( a ) an ensemble approach and ( b ) stepwise forward-and-backward selection method using logistic regression classifier; and (5) radiomics features-based classification of MYCN gene amplification using machine learning classifiers. RESULTS: Median train/test Dice score for modified U-Net was 0.728/0.680. The top 3 features from the ensemble approach were neighborhood gray-tone difference matrix (NGTDM) busyness, NGTDM strength, and gray-level run-length matrix (GLRLM) low gray-level run emphasis, whereas those from the stepwise approach were GLRLM low gray-level run emphasis, GLRLM high gray-level run emphasis, and NGTDM coarseness. The top-performing tumor classification algorithm achieved a weighted F1 score of 97%, an area under the receiver operating characteristic curve of 96.9%, an accuracy of 96.97%, and a negative predictive value of 100%. Harmonization-based tumor classification improved the accuracy by 2% to 3% for all classifiers. CONCLUSION: The proposed end-to-end framework achieved high accuracy for MYCN gene amplification status classification.

Neuroanatomical subtypes of schizophrenia and relationship with illness duration and deficit status.

BACKGROUND: The heterogeneity of schizophrenia (SCZ) regarding psychopathology, illness trajectory and their inter-relationships with underlying neural substrates remain incompletely understood. In a bid to reduce illness heterogeneity using neural substrates, our study aimed to replicate the findings of an earlier study by Chand et al. (2020). We employed brain structural measures for subtyping SCZ patients, and evaluate each subtype's relationship with clinical features such as illness duration, psychotic psychopathology, and additionally deficit status. METHODS: Overall, 240 subjects (160 SCZ patients, 80 healthy controls) were recruited for this study. The participants underwent brain structural magnetic resonance imaging scans and clinical rating using the Positive and Negative Syndrome Scale. Neuroanatomical subtypes of SCZ were identified using "Heterogeneity through discriminative analysis" (HYDRA), a clustering technique which accounted for relevant covariates and the inter-group normalized percentage changes in brain volume were also calculated. RESULTS: As replicated, two neuroanatomical subtypes (SG-1 and SG-2) were found amongst our patients with SCZ. The subtype SG-1 was associated with enlargements in the third and lateral ventricles, volume increase in the basal ganglia (putamen, caudate, pallidum), longer illness duration, and deficit status. The subtype SG-2 was associated with reductions of cortical and subcortical structures (hippocampus, thalamus, basal ganglia). CONCLUSIONS: These replicated findings have clinical implications in the early intervention, response monitoring, and prognostication of SCZ. Future studies may adopt a multi-modal neuroimaging approach to enhance insights into the neurobiological composition of relevant subtypes.

The lateral entorhinal cortex is a hub for local and global dysfunction in early Alzheimer's disease states.

Functional network activity alterations are one of the earliest hallmarks of Alzheimer's disease (AD), detected prior to amyloidosis and tauopathy. Better understanding the neuronal underpinnings of such network alterations could offer mechanistic insight into AD progression. Here, we examined a mouse model (3xTgAD mice) recapitulating this early AD stage. We found resting functional connectivity loss within ventral networks, including the entorhinal cortex, aligning with the spatial distribution of tauopathy reported in humans. Unexpectedly, in contrast to decreased connectivity at rest, 3xTgAD mice show enhanced fMRI signal within several projection areas following optogenetic activation of the entorhinal cortex. We corroborate this finding by demonstrating neuronal facilitation within ventral networks and synaptic hyperexcitability in projection targets. 3xTgAD mice, thus, reveal a dichotomic hypo-connected:resting versus hyper-responsive:active phenotype. This strong homotopy between the areas affected supports the translatability of this pathophysiological model to tau-related, early-AD deficits in humans.

Machine learning classification of schizophrenia patients and healthy controls using diverse neuroanatomical markers and Ensemble methods.

Schizophrenia is a major psychiatric disorder that imposes enormous clinical burden on patients and their caregivers. Determining classification biomarkers can complement clinical measures and improve understanding of the neural basis underlying schizophrenia. Using neuroanatomical features, several machine learning based investigations have attempted to classify schizophrenia from healthy controls but the range of neuroanatomical measures employed have been limited in range to date. In this study, we sought to classify schizophrenia and healthy control cohorts using a diverse set of neuroanatomical measures (cortical and subcortical volumes, cortical areas and thickness, cortical mean curvature) and adopted Ensemble methods for better performance. Additionally, we correlated such neuroanatomical features with Quality of Life (QoL) assessment scores within the schizophrenia cohort. With Ensemble methods and diverse neuroanatomical measures, we achieved classification accuracies ranging from 83 to 87%, sensitivities and specificities varying between 90-98% and 65-70% respectively. In addition to lower QoL scores within schizophrenia cohort, significant correlations were found between specific neuroanatomical measures and psychological health, social relationship subscale domains of QoL. Our results suggest the utility of inclusion of subcortical and cortical measures and Ensemble methods to achieve better classification performance and their potential impact of parsing out neurobiological correlates of quality of life in schizophrenia.

CAFT: a deep learning-based comprehensive abdominal fat analysis tool for large cohort studies.

BACKGROUND: There is increasing appreciation of the association of obesity beyond co-morbidities, such as cancers, Type 2 diabetes, hypertension, and stroke to also impact upon the muscle to give rise to sarcopenic obesity. Phenotypic knowledge of obesity is crucial for profiling and management of obesity, as different fat-subcutaneous adipose tissue depots (SAT) and visceral adipose tissue depots (VAT) have various degrees of influence on metabolic syndrome and morbidities. Manual segmentation is time consuming and laborious. Study focuses on the development of a deep learning-based, complete data processing pipeline for MRI-based fat analysis, for large cohort studies which include (1) data augmentation and preprocessing (2) model zoo (3) visualization dashboard, and (4) correction tool, for automated quantification of fat compartments SAT and VAT. METHODS: Our sample comprised 190 healthy community-dwelling older adults from the Geri-LABS study with mean age of 67.85 ± 7.90 years, BMI 23.75 ± 3.65 kg/m2, 132 (69.5%) female, and mainly Chinese ethnicity. 3D-modified Dixon T1-weighted gradient-echo MR images were acquired. Residual global aggregation-based 3D U-Net (RGA-U-Net) and standard 3D U-Net were trained to segment SAT, VAT, superficial and deep subcutaneous adipose tissue depots (SSAT and DSAT). Manual segmentation from 26 subjects was used as ground truth during training. Data augmentations, random bias, noise and ghosting were carried out to increase the number of training datasets to 130. Segmentation accuracy was evaluated using Dice and Hausdorff metrics. RESULTS: The accuracy of segmentation was SSAT:0.92, DSAT:0.88 and VAT:0.9. Average Hausdorff distance was less than 5 mm. Automated segmentation significantly correlated R2 > 0.99 (p < 0.001) with ground truth for all 3-fat compartments. Predicted volumes were within ± 1.96SD from Bland-Altman analysis. CONCLUSIONS: DL-based, comprehensive SSAT, DSAT, and VAT analysis tool showed high accuracy and reproducibility and provided a comprehensive fat compartment composition analysis and visualization in less than 10 s.

A triple-network organization for the mouse brain.

The triple-network model of psychopathology is a framework to explain the functional and structural neuroimaging phenotypes of psychiatric and neurological disorders. It describes the interactions within and between three distributed networks: the salience, default-mode, and central executive networks. These have been associated with brain disorder traits in patients. Homologous networks have been proposed in animal models, but their integration into a triple-network organization has not yet been determined. Using resting-state datasets, we demonstrate conserved spatio-temporal properties between triple-network elements in human, macaque, and mouse. The model predictions were also shown to apply in a mouse model for depression. To validate spatial homologies, we developed a data-driven approach to convert mouse brain maps into human standard coordinates. Finally, using high-resolution viral tracers in the mouse, we refined an anatomical model for these networks and validated this using optogenetics in mice and tractography in humans. Unexpectedly, we find serotonin involvement within the salience rather than the default-mode network. Our results support the existence of a triple-network system in the mouse that shares properties with that of humans along several dimensions, including a disease condition. Finally, we demonstrate a method to humanize mouse brain networks that opens doors to fully data-driven trans-species comparisons.

Gradients of functional connectivity in the mouse cortex reflect neocortical evolution.

Understanding cortical organization is a fundamental goal of neuroscience that requires comparisons across species and modalities. Large-scale connectivity gradients have recently been introduced as a data-driven representation of the intrinsic organization of the cortex. We studied resting-state functional connectivity gradients in the mouse cortex and found robust spatial patterns across four data sets. The principal gradient of functional connectivity shows a striking overlap with an axis of neocortical evolution from two primordial origins. Additional gradients reflect sensory specialization and aspects of a sensory-to-transmodal hierarchy, and are associated with transcriptomic features. While some of these gradients strongly resemble observations in the human cortex, the overall pattern in the mouse cortex emphasizes the specialization of sensory areas over a global functional hierarchy.

Common functional networks in the mouse brain revealed by multi-centre resting-state fMRI analysis.

Preclinical applications of resting-state functional magnetic resonance imaging (rsfMRI) offer the possibility to non-invasively probe whole-brain network dynamics and to investigate the determinants of altered network signatures observed in human studies. Mouse rsfMRI has been increasingly adopted by numerous laboratories worldwide. Here we describe a multi-centre comparison of 17 mouse rsfMRI datasets via a common image processing and analysis pipeline. Despite prominent cross-laboratory differences in equipment and imaging procedures, we report the reproducible identification of several large-scale resting-state networks (RSN), including a mouse default-mode network, in the majority of datasets. A combination of factors was associated with enhanced reproducibility in functional connectivity parameter estimation, including animal handling procedures and equipment performance. RSN spatial specificity was enhanced in datasets acquired at higher field strength, with cryoprobes, in ventilated animals, and under medetomidine-isoflurane combination sedation. Our work describes a set of representative RSNs in the mouse brain and highlights key experimental parameters that can critically guide the design and analysis of future rodent rsfMRI investigations.

Evaluation of nuisance removal for functional MRI of rodent brain.

Functional MRI (fMRI) has become an important translational tool for studying brain activity and connectivity in animal models and humans. For accurate and reliable measurement of functional connectivity, nuisance removal strategies developed for human brain, such as regressing motion parameters, cerebrospinal fluid (CSF)/white matter-derived signals and the global signal, have been applied to rodent. However, due to the very different anatomy, with the majority of the rodent brain being gray matter, and experimental conditions, in which animals are anesthetized and head-fixed, these methods may not be suitable for rodent fMRI. In this study, we assessed various nuisance regression methods and the effects of motion correction on a large dataset of both task and resting fMRI of anesthetized rat brain. Sensitivity and specificity were assessed in the somatosensory pathway under forepaw stimulation and resting state. Reproducibility at various sample sizes was simulated by randomly subsampling the dataset. To overcome the difficulty in extracting nuisance from the brain, a method using principal components estimated from tissues outside the brain was evaluated. Our results showed that neither detrend, motion correction, motion regression nor CSF signal regression could improve specificity despite increasing temporal signal-to-noise ratios. Although global signal regression increased the specificity of task activation and functional connectivity, the sensitivity and connectivity strength was drastically reduced, likely due to its strong correlation with the cortical signal. Motion parameters also correlated with task activation and the global signal, indicating that motion correction detected intensity variations in the brain. The nuisance estimated from tissues outside the brain produced a moderate improvement in specificity. In conclusion, nuisance removal suitable for human fMRI may not be optimal for rodents. While further development is needed, estimating nuisance from tissues outside the brain may be an alternative.

Connectomic imaging reveals Huntington-related pathological and pharmaceutical effects in a mouse model.

Recent studies suggest that neurodegenerative diseases could affect brain structure and function in disease-specific network patterns; however, how spontaneous activity affects structural covariance network (SC) is not clear. We hypothesized that hyper-excitability in Huntington disease (HD) disrupts the coordinated structural and functional connectivity, and treatment with memantine helps to reduce excitotoxicity and normalize the connectivity. MRI was conducted to measure somatosensory activation, resting-state functional-connectivity (rsFC), SC, amplitude of low frequency fluctuation (ALFF) and ALFF covariance (ALFFC) in the YAC128 mouse model of HD. We found somatosensory activation was unchanged but the subcortical ALFF was increased in HD mice, indicating subcortical but not cortical hyperactivity. The reduced sensorimotor rsFC but spared hippocampal and default mode networks in the HD mice was consistent with the more pronounced impairment in motor function compared with cognitive performance. The disease suppressed SC globally and reduced ALFFC in the basal ganglia network as well as its anti-correlation with the default mode network. By comparing these connectivity measures, we found that the originally coupled rsFC-SC relationship was impaired whereas SC-ALFFC correlation was increased by HD, suggesting disease facilitated covariation of brain volume and activity amplitude but not neural synchrony. The comparison with mono-synaptic axonal projection supports the hypothesis that rsFC, but not SC or ALFFC, is highly dependent on structural connectivity under healthy conditions. Treatment with memantine had a strong effect on normalizing the SC and reducing ALFF while slightly increasing other connectivity measures and restoring the rsFC-SC coupling, which is consistent with its effect on alleviating hyper-excitability and improving the coordinated neural growth. These results indicate that HD affects the cerebral structure-function relationship which could be partially reverted by NMDA antagonism. These connectivity measures provide unique insights into pathological and pharmaceutical effects in brain circuitry, and could be translatable biomarkers for evaluating drug effect and refining its efficacy.

Regulation of feeding by somatostatin neurons in the tuberal nucleus.

The tuberal nucleus (TN) is a surprisingly understudied brain region. We found that somatostatin (SST) neurons in the TN, which is known to exhibit pathological or cytological changes in human neurodegenerative diseases, play a crucial role in regulating feeding in mice. GABAergic tuberal SST (TNSST) neurons were activated by hunger and by the hunger hormone, ghrelin. Activation of TNSST neurons promoted feeding, whereas inhibition reduced it via projections to the paraventricular nucleus and bed nucleus of the stria terminalis. Ablation of TNSST neurons reduced body weight gain and food intake. These findings reveal a previously unknown mechanism of feeding regulation that operates through orexigenic TNSST neurons, providing a new perspective for understanding appetite changes.

GABAergic effect on resting-state functional connectivity: Dynamics under pharmacological antagonism.

Resting state functional connectivity MRI measures synchronous activity among brain regions although the mechanisms governing the temporally coherent BOLD signals remain unclear. Recent studies suggest that γ-amino butyric acid (GABA) levels are correlated with functional connectivity. To understand whether changes in GABA transmission alter functional connectivity, we modulated the GABAergic activity by a GABAA receptor antagonist, bicuculline. Resting and evoked electrophysiology and BOLD signals were measured in isoflurane-anesthetized rats under infusion of low-dose bicuculline or vehicle individually. Both somatosensory BOLD activations and evoked potentials induced by forepaw stimulation were increased significantly under bicuculline compared to vehicle, indicating increased excitability. Gradually elevated resting BOLD correlation within and between the somatosensory and visual cortices, as well as between somatosensory and caudate putamen but not within subcortical areas were found with the infusion of bicuculline. Increased cerebral blood flow was observed throughout the cortical and subcortical areas where the receptor density is high, but it didn't correlate with BOLD connectivity except in the primary somatosensory cortex. Furthermore, resting EEG coherence in the alpha and beta bands exhibited consistent change with the BOLD correlation. The increased cortico-cortical and cortico-striatal connectivity without dependence on the receptor distribution indicate that the functional connectivity may be mediated by long-range projection via the cortical and striatal GABAergic inter-neurons. Our results indicate an important role of the GABAergic system on neural and hemodynamic oscillations, which further supports the neuronal basis of functional connectivity MRI and its correlation with neurotransmission.

Dependence of BOLD signal fluctuation on arterial blood CO2 and O2: Implication for resting-state functional connectivity.

Blood oxygenation level dependent (BOLD) functional MRI signal is known to be modulated by the CO2 level. Typically only end-tidal CO2, rather than the arterial partial pressure of CO2 (paCO2), was measured while the arterial partial pressure of O2 (paO2) level was not controlled due to free breathing, making their contribution not separable. Especially, the influences of paO2 and paCO2 on resting-state functional connectivity are not well studied. In this study, we investigated the relationship between paCO2 and resting as well as stimulus-evoked BOLD signals under hyperoxic and hypercapnic manipulation with tight control of arterial paO2. Rats under isoflurane anesthesia were subjected to six inspired gas conditions: 47% O2 in air (Normal), adding 1%, 2% or 5% CO2, carbogen (95% O2/5% CO2), and 100% O2. Somatosensory BOLD activation was significantly increased under 100% O2, while reduced with increased paCO2 levels. However, while resting BOLD connectivity pattern expanded and bilateral correlation increased under 100% O2, the correlation coefficient between the left and right somatosensory cortex was generally not dependent on paCO2 or paO2. Interestingly, the correlation in 0.04-0.07Hz range significantly increased with CO2 levels. Intracortical electrophysiological recordings showed a similar trend as the BOLD but the neurovascular coupling varied. The results suggest that paO2 and paCO2 together rather than paCO2 alone alter the BOLD signal. The response is not purely vascular in nature but has strong neuronal origins. This should be taken into consideration when designing calibrated BOLD experiment and interpreting functional connectivity data especially in aging, under drug, or neurological disorders.