Development and validation of open-source deep neural networks for comprehensive chest x-ray reading: a retrospective, multicentre study.

BACKGROUND: Artificial intelligence (AI) systems for automated chest x-ray interpretation hold promise for standardising reporting and reducing delays in health systems with shortages of trained radiologists. Yet, there are few freely accessible AI systems trained on large datasets for practitioners to use with their own data with a view to accelerating clinical deployment of AI systems in radiology. We aimed to contribute an AI system for comprehensive chest x-ray abnormality detection. METHODS: In this retrospective cohort study, we developed open-source neural networks, X-Raydar and X-Raydar-NLP, for classifying common chest x-ray findings from images and their free-text reports. Our networks were developed using data from six UK hospitals from three National Health Service (NHS) Trusts (University Hospitals Coventry and Warwickshire NHS Trust, University Hospitals Birmingham NHS Foundation Trust, and University Hospitals Leicester NHS Trust) collectively contributing 2 513 546 chest x-ray studies taken from a 13-year period (2006-19), which yielded 1 940 508 usable free-text radiological reports written by the contemporary assessing radiologist (collectively referred to as the "historic reporters") and 1 896 034 frontal images. Chest x-rays were labelled using a taxonomy of 37 findings by a custom-trained natural language processing (NLP) algorithm, X-Raydar-NLP, from the original free-text reports. X-Raydar-NLP was trained on 23 230 manually annotated reports and tested on 4551 reports from all hospitals. 1 694 921 labelled images from the training set and 89 238 from the validation set were then used to train a multi-label image classifier. Our algorithms were evaluated on three retrospective datasets: a set of exams sampled randomly from the full NHS dataset reported during clinical practice and annotated using NLP (n=103 328); a consensus set sampled from all six hospitals annotated by three expert radiologists (two independent annotators for each image and a third consultant to facilitate disagreement resolution) under research conditions (n=1427); and an independent dataset, MIMIC-CXR, consisting of NLP-annotated exams (n=252 374). FINDINGS: X-Raydar achieved a mean AUC of 0·919 (SD 0·039) on the auto-labelled set, 0·864 (0·102) on the consensus set, and 0·842 (0·074) on the MIMIC-CXR test, demonstrating similar performance to the historic clinical radiologist reporters, as assessed on the consensus set, for multiple clinically important findings, including pneumothorax, parenchymal opacification, and parenchymal mass or nodules. On the consensus set, X-Raydar outperformed historical reporter balanced accuracy with significance on 27 of 37 findings, was non-inferior on nine, and inferior on one finding, resulting in an average improvement of 13·3% (SD 13·1) to 0·763 (0·110), including a mean 5·6% (13·2) improvement in critical findings to 0·826 (0·119). INTERPRETATION: Our study shows that automated classification of chest x-rays under a comprehensive taxonomy can achieve performance levels similar to those of historical reporters and exhibit robust generalisation to external data. The open-sourced neural networks can serve as foundation models for further research and are freely available to the research community. FUNDING: Wellcome Trust.

Patient Reidentification from Chest Radiographs: An Interpretable Deep Metric Learning Approach and Its Applications.

Purpose: To train an explainable deep learning model for patient reidentification in chest radiograph datasets and assess changes in model-perceived patient identity as a marker for emerging radiologic abnormalities in longitudinal image sets. Materials and Methods: This retrospective study used a set of 1 094 537 frontal chest radiographs and free-text reports from 259 152 patients obtained from six hospitals between 2006 and 2019, with validation on the public ChestX-ray14, CheXpert, and MIMIC-CXR datasets. A deep learning model was trained for patient reidentification and assessed on patient identity confirmation, retrieval of patient images from a database based on a query image, and radiologic abnormality prediction in longitudinal image sets. The representation learned was incorporated into a generative adversarial network, allowing visual explanations of the relevant features. Performance was evaluated with sensitivity, specificity, F1 score, Precision at 1, R-Precision, and area under the receiver operating characteristic curve (AUC) for normal and abnormal prediction. Results: Patient reidentification was achieved with a mean F1 score of 0.996 ± 0.001 (2 SD) on the internal test set (26 152 patients) and F1 scores of 0.947-0.993 on the external test data. Database retrieval yielded a mean Precision at 1 score of 0.976 ± 0.005 at 299 × 299 resolution on the internal test set and Precision at 1 scores between 0.868 and 0.950 on the external datasets. Patient sex, age, weight, and other factors were identified as key model features. The model achieved an AUC of 0.73 ± 0.01 for abnormality prediction versus an AUC of 0.58 ± 0.01 for age prediction error. Conclusion: The image features used by a deep learning patient reidentification model for chest radiographs corresponded to intuitive human-interpretable characteristics, and changes in these identifying features over time may act as markers for an emerging abnormality.Keywords: Conventional Radiography, Thorax, Feature Detection, Supervised Learning, Convolutional Neural Network, Principal Component Analysis Supplemental material is available for this article. © RSNA, 2023See also the commentary by Raghu and Lu in this issue.

A descriptive analysis of wheelchair repair registry data.

Wheeled mobility and seating (WMS) devices allow users to achieve greater mobility independence. Previous studies determined that 53% of wheelchair users required one or more repairs over a 6-month period; however, there are a limited number of studies that have evaluated types of repairs. The purpose of this study was to describe the types of manual wheelchair, power wheelchair, and scooter repairs within the Wheelchair Repair Registry (WRR) and examine the association between WMS devices and the frequency of repairs. A dataset of 4,645 devices distributed in the United States was collected from equipment suppliers who performed and logged community-based wheelchair repair services. The results demonstrated common repairs found across devices were within the wheels/tires/forks and batteries/cables categories. Device type was the most significant predictor of variance in the number of repairs. Customizable manual wheelchairs, tilt-in-space, Groups 2 & 3 power wheelchairs, and scooters were associated with higher number of repairs compared to non-customizable manual wheelchairs, pediatric, heavy-duty manual wheelchairs, and Group 4 power wheelchairs. The higher failure rate found in specific devices may be associated with a population of more active users, environment/conditions where equipment is used, time spent in equipment, additional features on device, or lower durability.

The ß2 integrin-kindlin-3 interaction is essential for T-cell homing but dispensable for T-cell activation in vivo.

Kindlin-3 is mutated in the rare genetic disorder, leukocyte adhesion deficiency type III, which is characterized by deficient integrin-mediated adhesion of leukocytes and platelets. However, the specific roles of kindlin-3-ß2-integrin interactions in T-cell adhesion and homing and immune responses in vivo remain unclear. Here, we show that the TTT motif in ß2 integrins controls kindlin-3 binding. Mutation of the kindlin-3 binding site in ß2 integrins caused a loss of firm adhesion of T cells under both static and shear flow conditions and a reduction of T-cell homing to lymph nodes in vivo. However, atomic force microscopy studies of integrin-ligand bonds revealed that initial ligand binding could still occur, and 2-dimensional T-cell migration was reduced but not abolished by the TTT/AAA mutation in the ß2 integrin. Importantly, dendritic cell-mediated T-cell activation in vivo was normal in TTT/AAA ß2 integrin knock-in mice. Our results reveal a selective role of the kindlin-3-integrin association for lymphocyte functions in vivo; the integrin-kindlin-3 interaction is particularly important in adhesion strengthening under shear flow, and for T-cell homing to lymph nodes, but dispensable for T cell activation which occurs in a shear-free environment.

Characterization of the SNAG and SLUG domains of Snail2 in the repression of E-cadherin and EMT induction: modulation by serine 4 phosphorylation.

Snail1 and Snail2, two highly related members of the Snail superfamily, are direct transcriptional repressors of E-cadherin and EMT inducers. Previous comparative gene profiling analyses have revealed important differences in the gene expression pattern regulated by Snail1 and Snail2, indicating functional differences between both factors. The molecular mechanism of Snail1-mediated repression has been elucidated to some extent, but very little is presently known on the repression mediated by Snail2. In the present work, we report on the characterization of Snail2 repression of E-cadherin and its regulation by phosphorylation. Both the N-terminal SNAG and the central SLUG domains of Snail2 are required for efficient repression of the E-cadherin promoter. The co-repressor NCoR interacts with Snail2 through the SNAG domain, while CtBP1 is recruited through the SLUG domain. Interestingly, the SNAG domain is absolutely required for EMT induction while the SLUG domain plays a negative modulation of Snail2 mediated EMT. Additionally, we identify here novel in vivo phosphorylation sites at serine 4 and serine 88 of Snail2 and demonstrate the functional implication of serine 4 in the regulation of Snail2-mediated repressor activity of E-cadherin and in Snail2 induction of EMT.

A systemic lupus erythematosus-associated R77H substitution in the CD11b chain of the Mac-1 integrin compromises leukocyte adhesion and phagocytosis.

The Mac-1 integrin is expressed mainly on myeloid cells and binds several ligands, including members of the ICAM family and the complement factor iC3b. It is involved in essential immunological processes, such as leukocyte extravasation and phagocytosis. In addition, Mac-1 has been described to negatively regulate immune cell signaling. Recently, a single nucleotide polymorphism conferring an amino acid change in the Mac-1 integrin extracellular domain, R77H, was shown to be associated with systemic lupus erythematosus. Here, we demonstrate that the R77H-substituted Mac-1 can be expressed on the cell surface in transfected cells and can undergo conformational changes in response to integrin activation. The affinity of the integrin for ICAMs is only partially reduced, but cell adhesion to ICAM-1 and ICAM-2 is severely compromised, and Jß2.7 cells expressing R77H substituted integrins are deficient in adhesion to ICAM-1 under shear flow conditions. Importantly, cell adhesion to the complement factor iC3b is also diminished, and COS cells expressing R77H-substituted integrins display reduced iC3b-dependent phagocytosis. In addition, U937 cells expressing R77H-CD11b display increased IL-6 production as compared with WT-CD11b-expressing cells. These results suggest that the R77H substitution results in the deficiency of the mutated integrin to mediate cell adhesion to ligands such as ICAMs and iC3b. These deficiencies may ultimately lead to detrimental effects on the immune system and contribute to the development of systemic lupus erythematosus.

Phosphorylation of serine 11 and serine 92 as new positive regulators of human Snail1 function: potential involvement of casein kinase-2 and the cAMP-activated kinase protein kinase A.

Snail1 is a major factor for epithelial-mesenchymal transition (EMT), an important event in tumor metastasis and in other pathologies. Snail1 is tightly regulated at transcriptional and posttranscriptional levels. Control of Snail1 protein stability and nuclear export by GSK3beta phosphorylation is important for Snail1 functionality. Stabilization mechanisms independent of GSK3beta have also been reported, including interaction with LOXL2 or regulation of the COP9 signalosome by inflammatory signals. To get further insights into the role of Snail1 phosphorylation, we have performed an in-depth analysis of in vivo human Snail1 phosphorylation combined with mutational studies. We identify new phosphorylation sites at serines 11, 82, and 92 and confirmed previously suggested phosphorylations at serine 104 and 107. Serines 11 and 92 participate in the control of Snail1 stability and positively regulate Snail1 repressive function and its interaction with mSin3A corepressor. Furthermore, serines 11 and 92 are required for Snail1-mediated EMT and cell viability, respectively. PKA and CK2 have been characterized as the main kinases responsible for in vitro Snail1 phosphorylation at serine 11 and 92, respectively. These results highlight serines 11 and 92 as new players in Snail1 regulation and suggest the participation of CK2 and PKA in the modulation of Snail1 functionality.

Transient receptor potential-like channels are essential for calcium signaling and fluid transport in a Drosophila epithelium.

Calcium signaling is an important mediator of neuropeptide-stimulated fluid transport by Drosophila Malpighian (renal) tubules. We demonstrate the first epithelial role, in vivo, for members of the TRP family of calcium channels. RT-PCR revealed expression of trp, trpl, and trpgamma in tubules. Use of antipeptide polyclonal antibodies for TRP, TRPL, and TRPgamma showed expression of all three channels in type 1 (principal) cells in the tubule main segment. Neuropeptide (CAP(2b))-stimulated fluid transport rates were significantly reduced in tubules from the trpl(302) mutant and the trpl;trp double mutant, trpl(302);trp(343). However, a trp null, trp(343), had no impact on stimulated fluid transport. Measurement of cytosolic calcium concentrations ([Ca(2+)](i)) in tubule principal cells using an aequorin transgene in trp and trpl mutants showed a reduction in calcium responses in trpl(302). Western blotting of tubule preparations from trp and trpl mutants revealed a correlation between TRPL levels and CAP(2b)-stimulated fluid transport and calcium signaling. Rescue of trpl(302) with a trpl transgene under heat-shock control resulted in a stimulated fluid transport phenotype that was indistinguishable from wild-type tubules. Furthermore, restoration of normal stimulated rates of fluid transport by rescue of trpl(302) was not compromised by introduction of the trp null, trp(343). Thus, in an epithelial context, TRPL is sufficient for wild-type responses. Finally, a scaffolding component of the TRPL/TRP-signaling complex, INAD, is not expressed in tubules, suggesting that inaD is not essential for TRPL/TRP function in Drosophila tubules.

The dg2 (for) gene confers a renal phenotype in Drosophila by modulation of cGMP-specific phosphodiesterase.

Fluid transport in Drosophila melanogaster tubules is regulated by guanosine 3',5'-cyclic monophosphate (cGMP) signalling. Here we compare the functional effects on tubules of different alleles of the dg2 (foraging or for) gene encoding a cGMP-dependent protein kinase (cGK), and show that the fors allele confers an epithelial phenotype. This manifests itself as hypersensitivity of epithelial fluid transport to the nitridergic neuropeptide, capa-1, which acts through nitric oxide and cGMP. However, there was no significant difference in tubule cGK activity between fors and forR adults. Nonetheless, fors tubules contained higher levels of cGMP-specific phosphodiesterase (cG-PDE) activity compared to forR. This increase in cGMP-PDE activity sufficed to decrease cGMP content in fors tubules compared to forR. Challenge of tubules with capa-1 increases cGMP content in both fors and forR tubules, although the increase from resting cGMP levels is greater in fors tubules. Capa-1 stimulation of tubules reveals a potent inhibition of cG-PDE in both lines, although this is greater in fors; and is sufficient to explain the hypersensitive transport phenotype observed. Thus, polymorphisms at the dg2 locus do indeed confer a cGMP-dependent transport phenotype, but this can best be ascribed to an indirect modulation of cG-PDE activity, and thence cGMP homeostasis, rather than a direct effect on cGK levels.

Analysis of Drosophila cGMP-dependent protein kinases and assessment of their in vivo roles by targeted expression in a renal transporting epithelium.

cGMP-dependent protein kinase (cGK) forms encoded by the dg2 (for) gene are implicated in behavior and epithelial transport in Drosophila melanogaster. Here, we provide the first biochemical characterization and cellular localization of cGKs encoded by the major transcripts of dg2: dg2P1 and dg2P2. cGMP stimulates kinase activity of DG2P1 (EC(50): 0.13 +/- 0.039 microm) and DG2P2 (EC(50): 0.32 +/- 0.14 microm) in Malpighian tubule and S2 cell extracts. DG2P1 and DG2P2 are magnesium-requiring enzymes and were inhibited by 10 and 100 microm of a cGK inhibitor, 8-(4-chlorophenylthio)guanosine-3',5'-cyclic monophosphorothioate, Rp isomer; whereas DG1, the cGK encoded by the D. melanogaster dg1 gene, was unaffected. DG2P1 and DG2P2 were localized in the plasma membrane in S2 cells, whereas DG1 was localized in the cytosol. The D. melanogaster fluid-transporting Malpighian tubule was used as an organotypic model to analyze cGK localization and function in vivo. Targeted expression of DG2P2, DG2P1, and DG1 in tubule cells via the UAS/GAL4 system in transgenic flies revealed differential localization of all three cGKs in vivo: DG2P2 expression at the apical membrane; DG2P1 expression at both the apical and basolateral membranes; and DG1 expression at the basolateral membrane and in the cytosol. Transgenic tubules for all three cGKs displayed enhanced cGK activity compared with wild-type tubules. The physiological impact of targeted expression of individual cGKs in tubule principal cells was assessed by measuring basal and stimulated rates of fluid transport. DG1 expression greatly enhanced fluid transport by the tubule in response to exogenous cGMP, whereas DG2P2 expression significantly increased fluid transport in response to the nitridergic neuropeptide, capa-1. Thus, dg2-encoded proteins are bona fide cGKs, which have differential roles in epithelial fluid transport, as assessed by in vivo studies. Furthermore, a novel epithelial role is suggested for DG1, which is considerably more responsive to cGMP than to capa-1 stimulation.

Interactions between epithelial nitric oxide signaling and phosphodiesterase activity in Drosophila.

Signaling by nitric oxide (NO) and guanosine 3',5'-cyclic monophosphate (cGMP) modulates fluid transport in Drosophila melanogaster. Expression of an inducible transgene encoding Drosophila NO synthase (dNOS) increases both NOS activity in Malpighian (renal) tubules and DNOS protein in both type I (principal) and type II (stellate) cells. However, cGMP content is increased only in principal cells. DNOS overexpression results in elevated basal rates of fluid transport in the presence of the phosphodiesterase (PDE) inhibitor, Zaprinast. Direct assay of tubule cGMP-hydrolyzing phosphodiesterase (cG-PDE) activity in wild-type and dNOS transgenic lines shows that cG-PDE activity is Zaprinast sensitive and is elevated upon dNOS induction. Zaprinast treatment increases cGMP content in tubules, particularly at the apical regions of principal cells, suggesting localization of Zaprinast-sensitive cG-PDE to these areas. Potential cross talk between activated NO/cGMP and calcium signaling was assessed in vivo with a targeted aequorin transgene. Activated DNOS signaling alone does not modify either neuropeptide (CAP2b)- or cGMP-induced increases in cytosolic calcium levels. However, in the presence of Zaprinast, both CAP2b-and cGMP-stimulated calcium levels are potentiated upon DNOS overexpression. Use of the calcium channel blocker, verapamil, abolishes the Zaprinast-induced transport phenotype in dNOS-overexpressing tubules. Molecular genetic intervention in the NO/cGMP signaling pathway has uncovered a pivotal role for cell-specific cG-PDE in regulating the poise of the fluid transporting Malpighian tubule via direct effects on intracellular cGMP concentration and localization and via interactions with calcium signaling mechanisms.