LAMA: automated image analysis for the developmental phenotyping of mouse embryos.

Advanced 3D imaging modalities, such as micro-computed tomography (micro-CT), have been incorporated into the high-throughput embryo pipeline of the International Mouse Phenotyping Consortium (IMPC). This project generates large volumes of raw data that cannot be immediately exploited without significant resources of personnel and expertise. Thus, rapid automated annotation is crucial to ensure that 3D imaging data can be integrated with other multi-dimensional phenotyping data. We present an automated computational mouse embryo phenotyping pipeline that harnesses the large amount of wild-type control data available in the IMPC embryo pipeline in order to address issues of low mutant sample number as well as incomplete penetrance and variable expressivity. We also investigate the effect of developmental substage on automated phenotyping results. Designed primarily for developmental biologists, our software performs image pre-processing, registration, statistical analysis and segmentation of embryo images. We also present a novel anatomical E14.5 embryo atlas average and, using it with LAMA, show that we can uncover known and novel dysmorphology from two IMPC knockout lines.

Exploring watercolor illusion spreading between dissected stimulus parts.

The watercolor illusion (WCI) occurs when an achromatic region is surrounded by an outer contour and inner chromatic fringe, resulting in an apparent pale tint of the same hue as the fringe. The WCI both fills in and spreads out, with the previous literature suggesting it always spreads out in the absence of an enclosing border. We examined how global stimulus configuration affects this illusion by dissecting various WCI-inducing stimuli into parts. Specifically, would color spread out of the unenclosed ends of the disconnected parts? Participants provided WCI illusion magnitude ratings and shading data indicating perceived locations of color spreading for a variety of stimulus configurations. Instead of the WCI spreading modally into the spaces between the disconnected parts, we found a global reorganization of the stimuli occurred. The dissected WCI stimuli were perceived as either amodally completed behind a white illusory surface perceptually different than the physically identical background or, as empty space between separate objects depending in part on the distance between dissected parts. This study demonstrates the WCI does not always spread outside of unenclosed borders when the global interpretation interferes with spreading. These findings highlight the importance of global configuration and perceptual organization in the WCI.

Corrigendum: High-throughput discovery of novel developmental phenotypes.

This corrects the article DOI: 10.1038/nature19356.

Improvement in Multidisciplinary Provider Rounding (Surgical Rounds) in the Pediatric Cardiac ICU: An Application of Lean Methodology.

OBJECTIVES: Provider-only, combined surgical, and medical multidisciplinary rounds ("surgical rounds") are essential to achieve optimal outcomes in large pediatric cardiac ICUs. Lean methodology was applied with the aims of identifying areas of waste and nonvalue-added work within the surgical rounds process. Thereby, the goals were to improve rounding efficiency and reduce rounding duration while not sacrificing critical patient care discussion nor delaying bedside rounds or surgical start times. DESIGN: Single-center improvement science study with observational and interventional phases from February 2, 2021, to July 31, 2021. SETTING: Tertiary pediatric cardiac ICU. PARTICIPANTS: Cardiothoracic surgery and cardiac intensive care team members participating in daily "surgical" rounds. INTERVENTIONS: Implementation of technology automation, creation of work instructions, standardization of patient presentation content and order, provider training, and novel role assignment. MEASUREMENTS AND MAIN RESULTS: Sixty-one multidisciplinary rounds were observed (30 pre, 31 postintervention). During the preintervention period, identified inefficiencies included prolonged preparation time, redundant work, presentation variability and extraneous information, and frequent provider transitions. Application of targeted interventions resulted in a 26% decrease in indexed rounds duration (2.42 vs 1.8 min; p = 0.0003), 50% decrease in indexed rounds preparation time (0.53 vs 0.27 min; p < 0.0001), and 66% decrease in transition time between patients (0.09 vs 0.03 min; p < 0.0001). The number of presenting provider changes also decreased (9 vs 4; p < 0.0001). Indexed discussion duration did not change (1 vs 0.98 min; p = 0.08) nor did balancing measures (bedside rounds and surgical start times) change (8.5 vs 9 min; p = 0.89 and 38 vs 22 min; p = 0.09). CONCLUSIONS: Lean methodology can be effectively applied to multidisciplinary rounds in a joint cardiothoracic surgery/cardiac intensive care setting to decrease waste and inefficiency. Interventions resulted in decreased preparation time, transition time, presenting provider changes, total rounds duration indexed to patient census, and anecdotal improvements in provider satisfaction.

Dual-Stream Spatiotemporal Networks with Feature Sharing for Monitoring Animals in the Home Cage.

This paper presents a spatiotemporal deep learning approach for mouse behavioral classification in the home-cage. Using a series of dual-stream architectures with assorted modifications for optimal performance, we introduce a novel feature sharing approach that jointly processes the streams at regular intervals throughout the network. The dataset in focus is an annotated, publicly available dataset of a singly-housed mouse. We achieved even better classification accuracy by ensembling the best performing models; an Inception-based network and an attention-based network, both of which utilize this feature sharing attribute. Furthermore, we demonstrate through ablation studies that for all models, the feature sharing architectures consistently outperform the conventional dual-stream having standalone streams. In particular, the inception-based architectures showed higher feature sharing gains with their increase in accuracy anywhere between 6.59% and 15.19%. The best-performing models were also further evaluated on other mouse behavioral datasets.

High-throughput discovery of novel developmental phenotypes.

Approximately one-third of all mammalian genes are essential for life. Phenotypes resulting from knockouts of these genes in mice have provided tremendous insight into gene function and congenital disorders. As part of the International Mouse Phenotyping Consortium effort to generate and phenotypically characterize 5,000 knockout mouse lines, here we identify 410 lethal genes during the production of the first 1,751 unique gene knockouts. Using a standardized phenotyping platform that incorporates high-resolution 3D imaging, we identify phenotypes at multiple time points for previously uncharacterized genes and additional phenotypes for genes with previously reported mutant phenotypes. Unexpectedly, our analysis reveals that incomplete penetrance and variable expressivity are common even on a defined genetic background. In addition, we show that human disease genes are enriched for essential genes, thus providing a dataset that facilitates the prioritization and validation of mutations identified in clinical sequencing efforts.

Evaluation of a Deep Learning-Derived Quantitative Retinopathy of Prematurity Severity Scale.

PURPOSE: To evaluate the clinical usefulness of a quantitative deep learning-derived vascular severity score for retinopathy of prematurity (ROP) by assessing its correlation with clinical ROP diagnosis and by measuring clinician agreement in applying a novel scale. DESIGN: Analysis of existing database of posterior pole fundus images and corresponding ophthalmoscopic examinations using 2 methods of assigning a quantitative scale to vascular severity. PARTICIPANTS: Images were from clinical examinations of patients in the Imaging and Informatics in ROP Consortium. Four ophthalmologists and 1 study coordinator evaluated vascular severity on a scale from 1 to 9. METHODS: A quantitative vascular severity score (1-9) was applied to each image using a deep learning algorithm. A database of 499 images was developed for assessment of interobserver agreement. MAIN OUTCOME MEASURES: Distribution of deep learning-derived vascular severity scores with the clinical assessment of zone (I, II, or III), stage (0, 1, 2, or 3), and extent (<3 clock hours, 3-6 clock hours, and >6 clock hours) of stage 3 evaluated using multivariate linear regression and weighted κ values and Pearson correlation coefficients for interobserver agreement on a 1-to-9 vascular severity scale. RESULTS: For deep learning analysis, a total of 6344 clinical examinations were analyzed. A higher deep learning-derived vascular severity score was associated with more posterior disease, higher disease stage, and higher extent of stage 3 disease (P < 0.001 for all). For a given ROP stage, the vascular severity score was higher in zone I than zones II or III (P < 0.001). Multivariate regression found zone, stage, and extent all were associated independently with the severity score (P < 0.001 for all). For interobserver agreement, the mean ± standard deviation weighted κ value was 0.67 ± 0.06, and the Pearson correlation coefficient ± standard deviation was 0.88 ± 0.04 on the use of a 1-to-9 vascular severity scale. CONCLUSIONS: A vascular severity scale for ROP seems feasible for clinical adoption; corresponds with zone, stage, extent of stage 3, and plus disease; and facilitates the use of objective technology such as deep learning to improve the consistency of ROP diagnosis.

A bioimage informatics platform for high-throughput embryo phenotyping.

High-throughput phenotyping is a cornerstone of numerous functional genomics projects. In recent years, imaging screens have become increasingly important in understanding gene-phenotype relationships in studies of cells, tissues and whole organisms. Three-dimensional (3D) imaging has risen to prominence in the field of developmental biology for its ability to capture whole embryo morphology and gene expression, as exemplified by the International Mouse Phenotyping Consortium (IMPC). Large volumes of image data are being acquired by multiple institutions around the world that encompass a range of modalities, proprietary software and metadata. To facilitate robust downstream analysis, images and metadata must be standardized to account for these differences. As an open scientific enterprise, making the data readily accessible is essential so that members of biomedical and clinical research communities can study the images for themselves without the need for highly specialized software or technical expertise. In this article, we present a platform of software tools that facilitate the upload, analysis and dissemination of 3D images for the IMPC. Over 750 reconstructions from 80 embryonic lethal and subviable lines have been captured to date, all of which are openly accessible at mousephenotype.org. Although designed for the IMPC, all software is available under an open-source licence for others to use and develop further. Ongoing developments aim to increase throughput and improve the analysis and dissemination of image data. Furthermore, we aim to ensure that images are searchable so that users can locate relevant images associated with genes, phenotypes or human diseases of interest.

Rapid Expansion of Human Epithelial Stem Cells Suitable for Airway Tissue Engineering.

RATIONALE: Stem cell-based tracheal replacement represents an emerging therapeutic option for patients with otherwise untreatable airway diseases including long-segment congenital tracheal stenosis and upper airway tumors. Clinical experience demonstrates that restoration of mucociliary clearance in the lungs after transplantation of tissue-engineered grafts is critical, with preclinical studies showing that seeding scaffolds with autologous mucosa improves regeneration. High epithelial cell-seeding densities are required in regenerative medicine, and existing techniques are inadequate to achieve coverage of clinically suitable grafts. OBJECTIVES: To define a scalable cell culture system to deliver airway epithelium to clinical grafts. METHODS: Human respiratory epithelial cells derived from endobronchial biopsies were cultured using a combination of mitotically inactivated fibroblasts and Rho-associated protein kinase (ROCK) inhibition using Y-27632 (3T3+Y). Cells were analyzed by immunofluorescence, quantitative polymerase chain reaction, and flow cytometry to assess airway stem cell marker expression. Karyotyping and multiplex ligation-dependent probe amplification were performed to assess cell safety. Differentiation capacity was tested in three-dimensional tracheospheres, organotypic cultures, air-liquid interface cultures, and an in vivo tracheal xenograft model. Ciliary function was assessed in air-liquid interface cultures. MEASUREMENTS AND MAIN RESULTS: 3T3-J2 feeder cells and ROCK inhibition allowed rapid expansion of airway basal cells. These cells were capable of multipotent differentiation in vitro, generating both ciliated and goblet cell lineages. Cilia were functional with normal beat frequency and pattern. Cultured cells repopulated tracheal scaffolds in a heterotopic transplantation xenograft model. CONCLUSIONS: Our method generates large numbers of functional airway basal epithelial cells with the efficiency demanded by clinical transplantation, suggesting its suitability for use in tracheal reconstruction.

Cell migration leads to spatially distinct but clonally related airway cancer precursors.

BACKGROUND: Squamous cell carcinoma of the lung is a common cancer with 95% mortality at 5 years. These cancers arise from preinvasive lesions, which have a natural history of development progressing through increasing severity of dysplasia to carcinoma in situ (CIS), and in some cases, ending in transformation to invasive carcinoma. Synchronous preinvasive lesions identified at autopsy have been previously shown to be clonally related. METHODS: Using autofluorescence bronchoscopy that allows visual observation of preinvasive lesions within the upper airways, together with molecular profiling of biopsies using gene sequencing and loss-of-heterozygosity analysis from both preinvasive lesions and from intervening normal tissue, we have monitored individual lesions longitudinally and documented their visual, histological and molecular relationship. RESULTS: We demonstrate that rather than forming a contiguous field of abnormal tissue, clonal CIS lesions can develop at multiple anatomically discrete sites over time. Further, we demonstrate that patients with CIS in the trachea have invariably had previous lesions that have migrated proximally, and in one case, into the other lung over a period of 12 years. CONCLUSIONS: Molecular information from these unique biopsies provides for the first time evidence that field cancerisation of the upper airways can occur through cell migration rather than via local contiguous cellular expansion as previously thought. Our findings urge a clinical strategy of ablating high-grade premalignant airway lesions with subsequent attentive surveillance for recurrence in the bronchial tree.

The influence of "advancing" and "receding" colors on figure-ground perception under monocular and binocular viewing.

Research on figure-ground perception has consistently found that red images are more likely to be perceived as figure/nearer, yet the mechanisms behind this are not completely clear. The primary theories have pointed to optical chromatic aberrations or cortical mechanisms, such as the antagonistic interactions of the magno-/parvocellular (M/P) systems. Our study explored this color-biased figure-ground perception by examining the duration for which a region was perceived as figure under both binocular and monocular conditions, using all combinations of red, blue, green, and gray. In Experiment 1, we used figure-ground ambiguous Maltese crosses, composed of left- and right-tilting sectors of equal area. In Experiment 2, the crosses were figure-ground biased with size and orientation cues. Here, small sectors of cardinal orientations, likely perceived as figure, were contrasted with larger, obliquely oriented sectors, likely perceived as ground. Under monocular conditions, the results aligned with chromatic aberration predictions: red advanced and blue receded, regardless of size and orientation. However, under binocular conditions, the advancing effect of red continued, but the receding effect of blue was generally not observed. Notably, blue, along with red and green, was more frequently perceived as figure compared to gray. The results under binocular viewing are in line with the expectations of the antagonistic M/P system interactions theory, likely due to the collective input from both eyes, facilitating the anticipated effects. Our findings suggest that color-biased figure-ground perception may arise from the synergistic effect of antagonistic M/P system interactions and other optical and cortical mechanisms, together compensating for chromatic aberrations.

Examining Increment thresholds as a function of pedestal contrast under hypothetical parvo- and magnocellular-biased conditions.

Theoretically, the pulsed- and steady-pedestal paradigms are thought to track contrast-increment thresholds (ΔC) as a function of pedestal contrast (C) for the parvocellular (P) and magnocellular (M) systems, respectively, yielding linear ΔC versus C functions for the pulsed- and nonlinear functions for the steady-pedestal paradigm. A recent study utilizing these paradigms to isolate the P and M systems reported no evidence of the M system being suppressed by red light, contrary to previous physiological and psychophysical findings. Curious as to why this may have occurred, we examined how ΔC varies with C for the P and M systems using the pulsed- and steady-pedestal paradigms and stimuli biased towards the P or M systems based on their sensitivity to spatial frequency (SF) and color. We found no effect of color and little influence of SF. To explain this lack of color effects, we used a quantitative model of ΔC (as it changes with C) to obtain Csat and contrast-gain values. The contrast-gain values (i) contradicted the hypothesis that the steady-pedestal paradigm tracks the M-system response, and (ii) our obtained Csat values indicated strongly that both pulsed- and steady-pedestal paradigms track primarily the P-system response.

Further Examination of the Pulsed- and Steady-Pedestal Paradigms under Hypothetical Parvocellular- and Magnocellular-Biased Conditions.

The pulsed- and steady-pedestal paradigms were designed to track increment thresholds (ΔC) as a function of pedestal contrast (C) for the parvocellular (P) and magnocellular (M) systems, respectively. These paradigms produce contrasting results: linear relationships between ΔC and C are observed in the pulsed-pedestal paradigm, indicative of the P system's processing, while the steady-pedestal paradigm reveals nonlinear functions, characteristic of the M system's response. However, we recently found the P model fits better than the M model for both paradigms, using Gabor stimuli biased towards the M or P systems based on their sensitivity to color and spatial frequency. Here, we used two-square pedestals under green vs. red light in the lower-left vs. upper-right visual fields to bias processing towards the M vs. P system, respectively. Based on our previous findings, we predicted the following: (1) steeper ΔC vs. C functions with the pulsed than the steady pedestal due to different task demands; (2) lower ΔCs in the upper-right vs. lower-left quadrant due to its bias towards P-system processing there; (3) no effect of color, since both paradigms track the P-system; and, most importantly (4) contrast gain should not be higher for the steady than for the pulsed pedestal. In general, our predictions were confirmed, replicating our previous findings and providing further evidence questioning the general validity of using the pulsed- and steady-pedestal paradigms to differentiate the P and M systems.

Suitability of endobronchial ultrasound-guided transbronchial needle aspiration specimens for subtyping and genotyping of non-small cell lung cancer: a multicenter study of 774 patients.

RATIONALE: The current management of advanced non-small cell lung cancer (NSCLC) requires differentiation between squamous and nonsquamous subtypes as well as epidermal growth factor receptor (EGFR) mutation status. Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) is increasingly used for the diagnosis and staging of lung cancer. However, it is unclear whether cytology specimens obtained with EBUS-TBNA are suitable for the subclassification and genotyping of NSCLC. OBJECTIVES: To determine whether cytology specimens obtained from EBUS-TBNA in routine practice are suitable for phenotyping and genotyping of NSCLC. METHODS: Cytological diagnoses from EBUS-TBNA were recorded from 774 patients with known or suspected lung cancer across five centers in the United Kingdom between 2009 and 2011. MEASUREMENTS AND MAIN RESULTS: The proportion of patients with a final diagnosis by EBUS-TBNA in whom subtype was classified was 77% (95% confidence interval [CI], 73-80). The rate of NSCLC not otherwise specified (NSCLC-NOS) was significantly reduced in patients who underwent immunohistochemistry (adjusted odds ratio, 0.50; 95% CI, 0.28-0.82; P = 0.016). EGFR mutation analysis was possible in 107 (90%) of the 119 patients in whom mutation analysis was requested. The sensitivity, negative predictive value, and diagnostic accuracy of EBUS-TBNA in patients with NSCLC were 88% (95% CI, 86-91), 72% (95% CI, 66-77), and 91% (95% CI, 89-93), respectively. CONCLUSIONS: This large, multicenter, pragmatic study demonstrates that cytology samples obtained from EBUS-TBNA in routine practice are suitable for subtyping of NSCLC and EGFR mutation analysis and that the use of immunohistochemistry reduces the rate of NSCLC-NOS.

Regularities in vertical saccadic metrics: new insights, and future perspectives.

Introduction: Asymmetries in processing by the healthy brain demonstrate regularities that facilitate the modeling of brain operations. The goal of the present study was to determine asymmetries in saccadic metrics during visual exploration, devoid of confounding clutter in the visual field. Methods: Twenty healthy adults searched for a small, low-contrast gaze-contingent target on a blank computer screen. The target was visible, only if eye fixation was within a 5 deg. by 5 deg. area of the target's location. Results: Replicating previously-reported asymmetries, repeated measures contrast analyses indicated that up-directed saccades were executed earlier, were smaller in amplitude, and had greater probability than down-directed saccades. Given that saccade velocities are confounded by saccade amplitudes, it was also useful to investigate saccade kinematics of visual exploration, as a function of vertical saccade direction. Saccade kinematics were modeled for each participant, as a square root relationship between average saccade velocity (i.e., average velocity between launching and landing of a saccade) and corresponding saccade amplitude (Velocity = S*[Saccade Amplitude]0.5). A comparison of the vertical scaling parameter (S) for up- and down-directed saccades showed that up-directed saccades tended to be slower than down-directed ones. Discussion: To motivate future research, an ecological theory of asymmetric pre-saccadic inhibition was presented to explain the collection of vertical saccadic regularities. For example, given that the theory proposes strong inhibition for the releasing of reflexive down-directed prosaccades (cued by an attracting peripheral target below eye fixation), and weak inhibition for the releasing of up-directed prosaccades (cued by an attracting peripheral target above eye fixation), a prediction for future studies is longer reaction times for vertical anti-saccade cues above eye fixation. Finally, the present study with healthy individuals demonstrates a rationale for further study of vertical saccades in psychiatric disorders, as bio-markers for brain pathology.

Association of Biomarker-Based Artificial Intelligence With Risk of Racial Bias in Retinal Images.

Importance: Although race is a social construct, it is associated with variations in skin and retinal pigmentation. Image-based medical artificial intelligence (AI) algorithms that use images of these organs have the potential to learn features associated with self-reported race (SRR), which increases the risk of racially biased performance in diagnostic tasks; understanding whether this information can be removed, without affecting the performance of AI algorithms, is critical in reducing the risk of racial bias in medical AI. Objective: To evaluate whether converting color fundus photographs to retinal vessel maps (RVMs) of infants screened for retinopathy of prematurity (ROP) removes the risk for racial bias. Design, Setting, and Participants: The retinal fundus images (RFIs) of neonates with parent-reported Black or White race were collected for this study. A u-net, a convolutional neural network (CNN) that provides precise segmentation for biomedical images, was used to segment the major arteries and veins in RFIs into grayscale RVMs, which were subsequently thresholded, binarized, and/or skeletonized. CNNs were trained with patients' SRR labels on color RFIs, raw RVMs, and thresholded, binarized, or skeletonized RVMs. Study data were analyzed from July 1 to September 28, 2021. Main Outcomes and Measures: Area under the precision-recall curve (AUC-PR) and area under the receiver operating characteristic curve (AUROC) at both the image and eye level for classification of SRR. Results: A total of 4095 RFIs were collected from 245 neonates with parent-reported Black (94 [38.4%]; mean [SD] age, 27.2 [2.3] weeks; 55 majority sex [58.5%]) or White (151 [61.6%]; mean [SD] age, 27.6 [2.3] weeks, 80 majority sex [53.0%]) race. CNNs inferred SRR from RFIs nearly perfectly (image-level AUC-PR, 0.999; 95% CI, 0.999-1.000; infant-level AUC-PR, 1.000; 95% CI, 0.999-1.000). Raw RVMs were nearly as informative as color RFIs (image-level AUC-PR, 0.938; 95% CI, 0.926-0.950; infant-level AUC-PR, 0.995; 95% CI, 0.992-0.998). Ultimately, CNNs were able to learn whether RFIs or RVMs were from Black or White infants regardless of whether images contained color, vessel segmentation brightness differences were nullified, or vessel segmentation widths were uniform. Conclusions and Relevance: Results of this diagnostic study suggest that it can be very challenging to remove information relevant to SRR from fundus photographs. As a result, AI algorithms trained on fundus photographs have the potential for biased performance in practice, even if based on biomarkers rather than raw images. Regardless of the methodology used for training AI, evaluating performance in relevant subpopulations is critical.

Bavituximab Decreases Immunosuppressive Myeloid-Derived Suppressor Cells in Newly Diagnosed Glioblastoma Patients.

PURPOSE: We evaluated the efficacy of bavituximab-a mAb with anti-angiogenic and immunomodulatory properties-in newly diagnosed patients with glioblastoma (GBM) who also received radiotherapy and temozolomide. Perfusion MRI and myeloid-related gene transcription and inflammatory infiltrates in pre-and post-treatment tumor specimens were studied to evaluate on-target effects (NCT03139916). PATIENTS AND METHODS: Thirty-three adults with IDH--wild-type GBM received 6 weeks of concurrent chemoradiotherapy, followed by 6 cycles of temozolomide (C1-C6). Bavituximab was given weekly, starting week 1 of chemoradiotherapy, for at least 18 weeks. The primary endpoint was proportion of patients alive at 12 months (OS-12). The null hypothesis would be rejected if OS-12 was ≥72%. Relative cerebral blood flow (rCBF) and vascular permeability (Ktrans) were calculated from perfusion MRIs. Peripheral blood mononuclear cells and tumor tissue were analyzed pre-treatment and at disease progression using RNA transcriptomics and multispectral immunofluorescence for myeloid-derived suppressor cells (MDSC) and macrophages. RESULTS: The study met its primary endpoint with an OS-12 of 73% (95% confidence interval, 59%-90%). Decreased pre-C1 rCBF (HR, 4.63; P = 0.029) and increased pre-C1 Ktrans were associated with improved overall survival (HR, 0.09; P = 0.005). Pre-treatment overexpression of myeloid-related genes in tumor tissue was associated with longer survival. Post-treatment tumor specimens contained fewer immunosuppressive MDSCs (P = 0.01). CONCLUSIONS: Bavituximab has activity in newly diagnosed GBM and resulted in on-target depletion of intratumoral immunosuppressive MDSCs. Elevated pre-treatment expression of myeloid-related transcripts in GBM may predict response to bavituximab.

Federated Learning for Multicenter Collaboration in Ophthalmology: Improving Classification Performance in Retinopathy of Prematurity.

OBJECTIVE: To compare the performance of deep learning classifiers for the diagnosis of plus disease in retinopathy of prematurity (ROP) trained using 2 methods for developing models on multi-institutional data sets: centralizing data versus federated learning (FL) in which no data leave each institution. DESIGN: Evaluation of a diagnostic test or technology. SUBJECTS: Deep learning models were trained, validated, and tested on 5255 wide-angle retinal images in the neonatal intensive care units of 7 institutions as part of the Imaging and Informatics in ROP study. All images were labeled for the presence of plus, preplus, or no plus disease with a clinical label and a reference standard diagnosis (RSD) determined by 3 image-based ROP graders and the clinical diagnosis. METHODS: We compared the area under the receiver operating characteristic curve (AUROC) for models developed on multi-institutional data, using a central approach initially, followed by FL, and compared locally trained models with both approaches. We compared the model performance (κ) with the label agreement (between clinical and RSD), data set size, and number of plus disease cases in each training cohort using the Spearman correlation coefficient (CC). MAIN OUTCOME MEASURES: Model performance using AUROC and linearly weighted κ. RESULTS: Four settings of experiment were used: FL trained on RSD against central trained on RSD, FL trained on clinical labels against central trained on clinical labels, FL trained on RSD against central trained on clinical labels, and FL trained on clinical labels against central trained on RSD (P = 0.046, P = 0.126, P = 0.224, and P = 0.0173, respectively). Four of the 7 (57%) models trained on local institutional data performed inferiorly to the FL models. The model performance for local models was positively correlated with the label agreement (between clinical and RSD labels, CC = 0.389, P = 0.387), total number of plus cases (CC = 0.759, P = 0.047), and overall training set size (CC = 0.924, P = 0.002). CONCLUSIONS: We found that a trained FL model performs comparably to a centralized model, confirming that FL may provide an effective, more feasible solution for interinstitutional learning. Smaller institutions benefit more from collaboration than larger institutions, showing the potential of FL for addressing disparities in resource access.

Federated Learning for Multicenter Collaboration in Ophthalmology: Implications for Clinical Diagnosis and Disease Epidemiology.

OBJECTIVE: To utilize a deep learning (DL) model trained via federated learning (FL), a method of collaborative training without sharing patient data, to delineate institutional differences in clinician diagnostic paradigms and disease epidemiology in retinopathy of prematurity (ROP). DESIGN: Evaluation of a diagnostic test or technology. SUBJECTS AND CONTROLS: We included 5245 patients with wide-angle retinal imaging from the neonatal intensive care units of 7 institutions as part of the Imaging and Informatics in ROP study. Images were labeled with the clinical diagnoses of plus disease (plus, preplus, no plus), which were documented in the chart, and a reference standard diagnosis was determined by 3 image-based ROP graders and the clinical diagnosis. METHODS: Demographics (birth weight, gestational age) and clinical diagnoses for all eye examinations were recorded from each institution. Using an FL approach, a DL model for plus disease classification was trained using only the clinical labels. The 3 class probabilities were then converted into a vascular severity score (VSS) for each eye examination, as well as an "institutional VSS," in which the average of the VSS values assigned to patients' higher severity ("worse") eyes at each examination was calculated for each institution. MAIN OUTCOME MEASURES: We compared demographics, clinical diagnoses of plus disease, and institutional VSSs between institutions using the McNemar-Bowker test, 2-proportion Z test, and 1-way analysis of variance with post hoc analysis by the Tukey-Kramer test. Single regression analysis was performed to explore the relationship between demographics and VSSs. RESULTS: We found that the proportion of patients diagnosed with preplus disease varied significantly between institutions (P < 0.001). Using the DL-derived VSS trained on the data from all institutions using FL, we observed differences in the institutional VSS and the level of vascular severity diagnosed as no plus (P < 0.001) across institutions. A significant, inverse relationship between the institutional VSS and mean gestational age was found (P = 0.049, adjusted R2 = 0.49). CONCLUSIONS: A DL-derived ROP VSS developed without sharing data between institutions using FL identified differences in the clinical diagnoses of plus disease and overall levels of ROP severity between institutions. Federated learning may represent a method to standardize clinical diagnoses and provide objective measurements of disease for image-based diseases.

Utility of endobronchial ultrasound-guided transbronchial needle aspiration in patients with tuberculous intrathoracic lymphadenopathy: a multicentre study.

BACKGROUND: Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) has emerged as an important tool for the diagnosis and staging of lung cancer but its role in the diagnosis of tuberculous intrathoracic lymphadenopathy has not been established. The aim of this study was to describe the diagnostic utility of EBUS-TBNA in patients with intrathoracic lymphadenopathy due to tuberculosis (TB). METHODS: 156 consecutive patients with isolated intrathoracic TB lymphadenitis were studied across four centres over a 2-year period. Only patients with a confirmed diagnosis or unequivocal clinical and radiological response to antituberculous treatment during follow-up for a minimum of 6 months were included. All patients underwent routine clinical assessment and a CT scan prior to EBUS-TBNA. Demographic data, HIV status, pathological findings and microbiological results were recorded. RESULTS: EBUS-TBNA was diagnostic of TB in 146 patients (94%; 95% CI 88% to 97%). Pathological findings were consistent with TB in 134 patients (86%). Microbiological investigations yielded a positive culture of TB in 74 patients (47%) with a median time to positive culture of 16 days (range 3-84) and identified eight drug-resistant cases (5%). Ten patients (6%) did not have a specific diagnosis following EBUS; four underwent mediastinoscopy which confirmed the diagnosis of TB while six responded to empirical antituberculous therapy. There was one complication requiring an inpatient admission. CONCLUSIONS: EBUS-TBNA is a safe and effective first-line investigation in patients with tuberculous intrathoracic lymphadenopathy.