Brain MRI Deep Learning and Bayesian Inference System Augments Radiology Resident Performance.

Automated quantitative and probabilistic medical image analysis has the potential to improve the accuracy and efficiency of the radiology workflow. We sought to determine whether AI systems for brain MRI diagnosis could be used as a clinical decision support tool to augment radiologist performance. We utilized previously developed AI systems that combine convolutional neural networks and expert-derived Bayesian networks to distinguish among 50 diagnostic entities on multimodal brain MRIs. We tested whether these systems could augment radiologist performance through an interactive clinical decision support tool known as Adaptive Radiology Interpretation and Education System (ARIES) in 194 test cases. Four radiology residents and three academic neuroradiologists viewed half of the cases unassisted and half with the results of the AI system displayed on ARIES. Diagnostic accuracy of radiologists for top diagnosis (TDx) and top three differential diagnosis (T3DDx) was compared with and without ARIES. Radiology resident performance was significantly better with ARIES for both TDx (55% vs 30%; P < .001) and T3DDx (79% vs 52%; P = 0.002), with the largest improvement for rare diseases (39% increase for T3DDx; P < 0.001). There was no significant difference between attending performance with and without ARIES for TDx (72% vs 69%; P = 0.48) or T3DDx (86% vs 89%; P = 0.39). These findings suggest that a hybrid deep learning and Bayesian inference clinical decision support system has the potential to augment diagnostic accuracy of non-specialists to approach the level of subspecialists for a large array of diseases on brain MRI.

Artificial Intelligence System Approaching Neuroradiologist-level Differential Diagnosis Accuracy at Brain MRI.

Background Although artificial intelligence (AI) shows promise across many aspects of radiology, the use of AI to create differential diagnoses for rare and common diseases at brain MRI has not been demonstrated. Purpose To evaluate an AI system for generation of differential diagnoses at brain MRI compared with radiologists. Materials and Methods This retrospective study tested performance of an AI system for probabilistic diagnosis in patients with 19 common and rare diagnoses at brain MRI acquired between January 2008 and January 2018. The AI system combines data-driven and domain-expertise methodologies, including deep learning and Bayesian networks. First, lesions were detected by using deep learning. Then, 18 quantitative imaging features were extracted by using atlas-based coregistration and segmentation. Third, these image features were combined with five clinical features by using Bayesian inference to develop probability-ranked differential diagnoses. Quantitative feature extraction algorithms and conditional probabilities were fine-tuned on a training set of 86 patients (mean age, 49 years ± 16 [standard deviation]; 53 women). Accuracy was compared with radiology residents, general radiologists, neuroradiology fellows, and academic neuroradiologists by using accuracy of top one, top two, and top three differential diagnoses in 92 independent test set patients (mean age, 47 years ± 18; 52 women). Results For accuracy of top three differential diagnoses, the AI system (91% correct) performed similarly to academic neuroradiologists (86% correct; P = .20), and better than radiology residents (56%; P < .001), general radiologists (57%; P < .001), and neuroradiology fellows (77%; P = .003). The performance of the AI system was not affected by disease prevalence (93% accuracy for common vs 85% for rare diseases; P = .26). Radiologists were more accurate at diagnosing common versus rare diagnoses (78% vs 47% across all radiologists; P < .001). Conclusion An artificial intelligence system for brain MRI approached overall top one, top two, and top three differential diagnoses accuracy of neuroradiologists and exceeded that of less-specialized radiologists. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Zaharchuk in this issue.

Comparison of γ-Aminobutyric Acid, Type A (GABAA), Receptor αßγ and αßδ Expression Using Flow Cytometry and Electrophysiology: EVIDENCE FOR ALTERNATIVE SUBUNIT STOICHIOMETRIES AND ARRANGEMENTS.

The subunit stoichiometry and arrangement of synaptic αßγ GABAA receptors are generally accepted as 2α:2ß:1γ with a ß-α-γ-ß-α counterclockwise configuration, respectively. Whether extrasynaptic αßδ receptors adopt the analogous ß-α-δ-ß-α subunit configuration remains controversial. Using flow cytometry, we evaluated expression levels of human recombinant γ2 and δ subunits when co-transfected with α1 and/or ß2 subunits in HEK293T cells. Nearly identical patterns of γ2 and δ subunit expression were observed as follows: both required co-transfection with α1 and ß2 subunits for maximal expression; both were incorporated into receptors primarily at the expense of ß2 subunits; and both yielded similar FRET profiles when probed for subunit adjacency, suggesting similar underlying subunit arrangements. However, because of a slower rate of δ subunit degradation, 10-fold less δ subunit cDNA was required to recapitulate γ2 subunit expression patterns and to eliminate the functional signature of α1ß2 receptors. Interestingly, titrating γ2 or δ subunit cDNA levels progressively altered GABA-evoked currents, revealing more than one kinetic profile for both αßγ and αßδ receptors. This raised the possibility of alternative receptor isoforms, a hypothesis confirmed using concatameric constructs for αßγ receptors. Taken together, our results suggest a limited cohort of alternative subunit arrangements in addition to canonical ß-α-γ/δ-ß-α receptors, including ß-α-γ/δ-α-α receptors at lower levels of γ2/δ expression and ß-α-γ/δ-α-γ/δ receptors at higher levels of expression. These findings provide important insight into the role of GABAA receptor subunit under- or overexpression in disease states such as genetic epilepsies.

GABAA receptor biogenesis is impaired by the γ2 subunit febrile seizure-associated mutation, GABRG2(R177G).

A missense mutation in the GABAA receptor γ2L subunit, R177G, was reported in a family with complex febrile seizures (FS). To gain insight into the mechanistic basis for these genetic seizures, we explored how the R177G mutation altered the properties of recombinant α1ß2γ2L GABAA receptors expressed in HEK293T cells. Using a combination of electrophysiology, flow cytometry, and immunoblotting, we found that the R177G mutation decreased GABA-evoked whole-cell current amplitudes by decreasing cell surface expression of α1ß2γ2L receptors. This loss of receptor surface expression resulted from endoplasmic reticulum (ER) retention of mutant γ2L(R177G) subunits, which unlike wild-type γ2L subunits, were degraded by ER-associated degradation (ERAD). Interestingly, when compared to the condition of homozygous γ2L(R177G) subunit expression, disproportionately low levels of γ2L(R177G) subunits reached the cell surface with heterozygous expression, indicating that wild-type γ2L subunits possessed a competitive advantage over mutant γ2L(R177G) subunits for receptor assembly and/or forward trafficking. Inhibiting protein synthesis with cycloheximide demonstrated that the R177G mutation primarily decreased the stability of an intracellular pool of unassembled γ2L subunits, suggesting that the mutant γ2L(R177G) subunits competed poorly with wild-type γ2L subunits due to impaired subunit folding and/or oligomerization. Molecular modeling confirmed that the R177G mutation could disrupt intrasubunit salt bridges, thereby destabilizing secondary and tertiary structure of γ2L(R177G) subunits. These findings support an emerging body of literature implicating defects in GABAA receptor biogenesis in the pathogenesis of genetic epilepsies (GEs) and FS.

The effect of HSP-causing mutations in SPG3A and NIPA1 on the assembly, trafficking, and interaction between atlastin-1 and NIPA1.

Despite its genetic heterogeneity, hereditary spastic paraplegia (HSP) is characterized by similar clinical phenotypes, suggesting that a common biochemical pathway underlies its pathogenesis. In support of this hypothesis, we used a combination of immunoprecipitation, confocal microscopy, and flow cytometry to demonstrate that two HSP-associated proteins, atlastin-1 and NIPA1, are direct binding partners, and interestingly, that the endogenous expression and trafficking of these proteins is highly dependent upon their coexpression. In addition, we demonstrated that the cellular distribution of atlastin-1:NIPA1 complexes was dramatically altered by HSP-causing mutations, as missense mutations in atlastin-1 (R239C and R495W) and NIPA1 (T45R and G106R) caused protein sequestration in the Golgi complex (GC) and endoplasmic reticulum (ER), respectively. Moreover, we demonstrated that HSP-causing mutations in both atlastin-1 and NIPA1 reduced axonal and dendritic sprouting in cultured rat cortical neurons. Together, these findings support the hypothesis that NIPA1 and atlastin-1 are members of a common biochemical pathway that supports axonal maintenance, which may explain in part the characteristic degeneration of long spinal pathways observed in patients with HSP.

Targeting ligand-gated ion channels in neurology and psychiatry: is pharmacological promiscuity an obstacle or an opportunity?

BACKGROUND: The traditional emphasis on developing high specificity pharmaceuticals ("magic bullets") for the treatment of Neurological and Psychiatric disorders is being challenged by emerging pathophysiology concepts that view disease states as abnormal interactions within complex networks of molecular and cellular components. So-called network pharmacology focuses on modifying the behavior of entire systems rather than individual components, a therapeutic strategy that would ideally employ single pharmacological agents capable of interacting with multiple targets ("magic shotguns"). For this approach to be successful, however, a framework for understanding pharmacological "promiscuity"--the ability of individual agents to modulate multiple molecular targets--is needed. PRESENTATION OF THE HYPOTHESIS: Pharmacological promiscuity is more often the rule than the exception for drugs that target the central nervous system (CNS). We hypothesize that promiscuity is an important contributor to clinical efficacy. Modulation patterns of existing therapeutic agents may provide critical templates for future drug discovery in Neurology and Psychiatry. TESTING THE HYPOTHESIS: To demonstrate the extent of pharmacological promiscuity and develop a framework for guiding drug screening, we reviewed the ability of 170 therapeutic agents and endogenous molecules to directly modulate neurotransmitter receptors, a class of historically attractive therapeutic targets in Neurology and Psychiatry. The results are summarized in the form of 1) receptor-centric maps that illustrate the degree of promiscuity for GABA-, glycine-, serotonin-, and acetylcholine-gated ion channels, and 2) drug-centric maps that illustrated how characterization of promiscuity can guide drug development. IMPLICATIONS OF THE HYPOTHESIS: Developing promiscuity maps of approved neuro-pharmaceuticals will provide therapeutic class-based templates against which candidate compounds can be screened. Importantly, compounds previously rejected in traditional screens due to poor specificity could be reconsidered in this framework. Further testing will require high throughput assays to systematically characterize interactions between available CNS-active drugs and surface receptors, both ionotropic and metabotropic.

Subspecialty-Level Deep Gray Matter Differential Diagnoses with Deep Learning and Bayesian Networks on Clinical Brain MRI: A Pilot Study.

PURPOSE: To develop and validate a system that could perform automated diagnosis of common and rare neurologic diseases involving deep gray matter on clinical brain MRI studies. MATERIALS AND METHODS: In this retrospective study, multimodal brain MRI scans from 212 patients (mean age, 55 years ± 17 [standard deviation]; 113 women) with 35 neurologic diseases and normal brain MRI scans obtained between January 2008 and January 2018 were included (110 patients in the training set, 102 patients in the test set). MRI scans from 178 patients (mean age, 48 years ± 17; 106 women) were used to supplement training of the neural networks. Three-dimensional convolutional neural networks and atlas-based image processing were used for extraction of 11 imaging features. Expert-derived Bayesian networks incorporating domain knowledge were used for differential diagnosis generation. The performance of the artificial intelligence (AI) system was assessed by comparing diagnostic accuracy with that of radiologists of varying levels of specialization by using the generalized estimating equation with robust variance estimator for the top three differential diagnoses (T3DDx) and the correct top diagnosis (TDx), as well as with receiver operating characteristic analyses. RESULTS: In the held-out test set, the imaging pipeline detected 11 key features on brain MRI scans with 89% accuracy (sensitivity, 81%; specificity, 95%) relative to academic neuroradiologists. The Bayesian network, integrating imaging features with clinical information, had an accuracy of 85% for T3DDx and 64% for TDx, which was better than that of radiology residents (n = 4; 56% for T3DDx, 36% for TDx; P < .001 for both) and general radiologists (n = 2; 53% for T3DDx, 31% for TDx; P < .001 for both). The accuracy of the Bayesian network was better than that of neuroradiology fellows (n = 2) for T3DDx (72%; P = .003) but not for TDx (59%; P = .19) and was not different from that of academic neuroradiologists (n = 2; 84% T3DDx, 65% TDx; P > .09 for both). CONCLUSION: A hybrid AI system was developed that simultaneously provides a quantitative assessment of disease burden, explainable intermediate imaging features, and a probabilistic differential diagnosis that performed at the level of academic neuroradiologists. This type of approach has the potential to improve clinical decision making for common and rare diseases.Supplemental material is available for this article.© RSNA, 2020.

Hereditary spastic paraplegia-associated mutations in the NIPA1 gene and its Caenorhabditis elegans homolog trigger neural degeneration in vitro and in vivo through a gain-of-function mechanism.

We studied the consequences of expression of wild-type (WT) human NIPA1 and two mutant forms of NIPA1 with known HSP-associated mutations (T45R and G106R) on cultured rat cortical neurons and using equivalent substitutions in the Caenorhabditis elegans NIPA1 homolog CeNIPA. WT NIPA1 localized in transfected neuronal and non-neuronal cells to the Golgi complex, a subset of synaptic vesicles, to a subset of early endosomes, and plasma cell membrane. Mutant NIPA1 accumulated in the endoplasmic reticulum (ER) triggering ER stress and features of apoptotic cell death. Flow cytometric analysis of NIPA1 surface expression demonstrated relatively intact trafficking of mutant forms and only the T45R mutant exhibited modestly reduced patterns of surface expression without evidence for a dominant-negative effect. In vivo pan-neuronal expression of the WT C. elegans NIPA1 homolog (CeNIPA) was well tolerated, with no obvious impact on neuronal morphology or behavior. In striking contrast, expression of CeNIPA bearing HSP-associated mutations caused a progressive neural degeneration and a clear motor phenotype. Neuronal loss in these animals began at day 7 and by day 9 animals were completely paralyzed. These effects appeared to arise from activation of the apoptotic program triggered by unfolded protein response (UPR), as we observed marked modifications of motor and cellular phenotype when mutant NIPA1 was expressed in caspase (ced-3)- and UPR (xbp-1)-deficient backgrounds. We propose that HSP-associated mutations in NIPA1 lead to cellular and functional deficits through a gain-of-function mechanism supporting the ER accumulation of toxic NIPA1 proteins.

Context-dependent modulation of alphabetagamma and alphabetadelta GABA A receptors by penicillin: implications for phasic and tonic inhibition.

Penicillin, an open-channel blocker of GABA(A) receptors, was recently reported to inhibit phasic, but not tonic, currents in hippocampal neurons. To distinguish between isoform-specific and context-dependent modulation as possible explanations for this selectivity, the effects of penicillin were evaluated on recombinant GABA(A) receptors expressed in HEK293T cells. When co-applied with saturating GABA, penicillin decreased peak amplitude, induced rebound, and prolonged deactivation of currents evoked from both synaptic and extrasynaptic receptor isoforms. However, penicillin had isoform-specific effects on the extent of desensitization, reflecting its ability to differentially modulate peak (non-equilibrium) and residual (near-equilibrium) currents. This suggested that the context of activation could determine the apparent sensitivity of a given receptor isoform to penicillin. To test this hypothesis, we explored the ability of penicillin to modulate synaptic and extrasynaptic isoform currents that were activated under more physiologically relevant conditions. Interestingly, while currents evoked from synaptic isoforms under phasic conditions (transient activation by a saturating concentration of GABA) were substantially inhibited by penicillin, currents evoked from extrasynaptic isoforms under tonic conditions (prolonged application by a sub-saturating concentration of GABA) were minimally affected. We therefore concluded that the reported inability of penicillin to modulate tonic currents could not simply be attributed to insensitivity of extrasynaptic receptors, but rather, reflected an inability to modulate these receptors in their native context of activation.

Relation of melatonin to sleep architecture in children with autism.

Children with autism often suffer from sleep disturbances, and compared to age-matched controls, have decreased melatonin levels, as indicated by urine levels of the primary melatonin metabolite, 6-sulfatoxymelatonin (6-SM). We therefore investigated the relationship between 6-SM levels and sleep architecture in children with autism spectrum disorders (ASD). Twenty-three children, aged 4-10 years, completed two nights of polysomnography and one overnight urine collection for measurement of urinary 6-SM excretion rate. Parents completed the Children's Sleep Habits Questionnaire. We found that higher urinary 6-SM excretion rates were associated with increased N3 sleep, decreased N2 sleep, and decreased daytime sleepiness. The results warrant further examination to examine the effects of supplemental melatonin on sleep architecture and daytime sleepiness.

Benzodiazepine modulation of GABA(A) receptor opening frequency depends on activation context: a patch clamp and simulation study.

Benzodiazepines (BDZs) are GABA(A) receptor modulators with anxiolytic, hypnotic, and anticonvulsant properties. BDZs are understood to potentiate GABA(A) receptor function by increasing channel opening frequency, in contrast to barbiturates, which increase channel open duration. However, the in vitro evidence demonstrating increased opening frequency involved prolonged exposure to sub-saturating GABA concentrations, conditions most similar to those found in extrasynaptic areas. In contrast, synaptic GABA(A) receptors are transiently activated by high GABA concentrations. To determine if BDZ modulation of single-channel opening frequency would be different for BDZ-sensitive receptors activated under synaptic versus extrasynaptic conditions, a combination of patch clamp recording and kinetic modeling was used. Consistent with the original experimental findings, BDZs were found to increase receptor affinity for GABA by decreasing the unbinding rate. While this mechanism was predicted to increase opening frequency under extrasynaptic conditions, simulations predicted that the same mechanism under synaptic conditions would increase the number, but not the frequency, of single-channel openings. Thus, a single mechanism (slower GABA unbinding) can produce differential changes in opening frequency under synaptic versus extrasynaptic conditions. The functional impact of BDZs on GABA(A) receptors therefore depends upon the physiological context of receptor activation.

A conserved Cys-loop receptor aspartate residue in the M3-M4 cytoplasmic loop is required for GABAA receptor assembly.

Members of the Cys-loop superfamily of ligand-gated ion channels, which mediate fast synaptic transmission in the nervous system, are assembled as heteropentamers from a large repertoire of neuronal subunits. Although several motifs in subunit N-terminal domains are known to be important for subunit assembly, increasing evidence points toward a role for C-terminal domains. Using a combination of flow cytometry, patch clamp recording, endoglycosidase H digestion, brefeldin A treatment, and analytic centrifugation, we identified a highly conserved aspartate residue at the boundary of the M3-M4 loop and the M4 domain that was required for binary and ternary gamma-aminobutyric acid type A receptor surface expression. Mutation of this residue caused mutant and partnering subunits to be retained in the endoplasmic reticulum, reflecting impaired forward trafficking. Interestingly although mutant and partnering wild type subunits could be coimmunoprecipitated, analytic centrifugation studies demonstrated decreased formation of pentameric receptors, suggesting that this residue played an important role in later steps of subunit oligomerization. We thus conclude that C-terminal motifs are also important determinants of Cys-loop receptor assembly.

Enhanced macroscopic desensitization shapes the response of alpha4 subtype-containing GABAA receptors to synaptic and extrasynaptic GABA.

Up-regulation of the GABAA receptor alpha4 subunit subtype has been consistently shown in multiple animal models of chronic epilepsy. This isoform is expressed in both thalamus and hippocampus and is likely to play a significant role in regulating corticothalamic and hippocampal rhythms. However, little is known about its physiological properties, thus limiting understanding of the role of alpha4 subtype-containing GABAA receptors in normal and abnormal physiology. We used rapid GABA application to recombinant GABAA receptors expressed in HEK293T cells to compare the macroscopic kinetic properties of alpha4beta3gamma2L receptors to those of the more widely distributed alpha1beta3gamma2L receptors. These receptor currents had similar peak current amplitudes and GABA EC50 values. However, alpha4beta3gamma2L currents activated more slowly when exposed to submaximal GABA concentrations, had more fast desensitization (tau = 15-100 ms), and had less residual current during long GABA applications. In addition, alpha4beta3gamma2L currents deactivated more slowly than alpha1beta3gamma2L currents. Peak currents evoked by repetitive, brief GABA applications were more strongly attenuated for alpha4beta3gamma2L currents than alpha1beta3gamma2L currents. Moreover, the time required to recover from desensitization was prolonged in alpha4beta3gamma2L currents compared to alpha1beta3gamma2L currents. We also found that exposure to prolonged low levels of GABA, similar to those that might be present in the extrasynaptic space, greatly suppressed the response of alpha4beta3gamma2L currents to higher concentrations of GABA, while alpha1beta3gamma2L currents were less affected by exposure to low levels of GABA. Taken together, these data suggest that alpha4beta3gamma2L receptors have unique kinetic properties that limit the range of GABA applications to which they can respond maximally. While similar to alpha1beta3gamma2L receptors in their ability to respond to brief and low frequency synaptic inputs, alpha4beta3gamma2L receptors are less efficacious when exposed to prolonged tonic GABA or during repetitive stimulation, as may occur during learning and seizures.

Microscopic kinetic determinants of macroscopic currents: insights from coupling and uncoupling of GABAA receptor desensitization and deactivation.

The time course of inhibitory postsynaptic currents (IPSCs) reflects GABA(A) receptor deactivation, the process of current relaxation following transient activation. Fast desensitization has been demonstrated to prolong deactivation, and these processes have been described as being 'coupled'. However, the relationship between desensitization and deactivation remains poorly understood. We investigated the 'uncoupling' of GABA(A) receptor macroscopic desensitization and deactivation using experimental conditions that affected these two processes differently. Changing agonist affinity preferentially altered deactivation, changing agonist concentration preferentially altered macroscopic desensitization, and a pore domain mutation prolonged deactivation despite blocking fast desensitization. To gain insight into the mechanistic basis for coupling and uncoupling, simulations were used to systematically evaluate the interplay between agonist affinity, gating efficacy, and desensitized state stability in shaping macroscopic desensitization and deactivation. We found that the influence of individual kinetic transitions on macroscopic currents depended not only on model connectivity, but also on the relationship among transitions within a given model. In addition, changing single rate constants differentially affected macroscopic desensitization and deactivation, thus providing parsimonious kinetic explanations for experimentally observed uncoupling. Finally, these findings permitted development of an algorithmic framework for kinetic interpretation of experimental manipulations that alter macroscopic current properties.