More Like this

1. Evaluation of federated learning variations for COVID-19 diagnosis using chest radiographs from 42 US and European hospitals

   https://doi.org/10.1093/jamia/ocac188  

   Peng, Le ; Luo, Gaoxiang ; Walker, Andrew ; Zaiman, Zachary ; Jones, Emma K. ; Gupta, Hemant ; Kersten, Kristopher ; Burns, John L. ; Harle, Christopher A. ; Magoc, Tanja ; et al ( October 2022 , Journal of the American Medical Informatics Association)

   Federated learning (FL) allows multiple distributed data holders to collaboratively learn a shared model without data sharing. However, individual health system data are heterogeneous. “Personalized” FL variations have been developed to counter data heterogeneity, but few have been evaluated using real-world healthcare data. The purpose of this study is to investigate the performance of a single-site versus a 3-client federated model using a previously described Coronavirus Disease 19 (COVID-19) diagnostic model. Additionally, to investigate the effect of system heterogeneity, we evaluate the performance of 4 FL variations.

   We leverage a FL healthcare collaborative including data from 5 international healthcare systems (US and Europe) encompassing 42 hospitals. We implemented a COVID-19 computer vision diagnosis system using the Federated Averaging (FedAvg) algorithm implemented on Clara Train SDK 4.0. To study the effect of data heterogeneity, training data was pooled from 3 systems locally and federation was simulated. We compared a centralized/pooled model, versus FedAvg, and 3 personalized FL variations (FedProx, FedBN, and FedAMP).

   We observed comparable model performance with respect to internal validation (local model: AUROC 0.94 vs FedAvg: 0.95, P = .5) and improved model generalizability with the FedAvg model (P < .05). When investigating the effects of model heterogeneity, we observed poor performance with FedAvg on internal validation as compared to personalized FL algorithms. FedAvg did have improved generalizability compared to personalized FL algorithms. On average, FedBN had the best rank performance on internal and external validation.

   FedAvg can significantly improve the generalization of the model compared to other personalization FL algorithms; however, at the cost of poor internal validity. Personalized FL may offer an opportunity to develop both internal and externally validated algorithms.

    

   more » « less
2. Federated End-to-End Unrolled Models for Magnetic Resonance Image Reconstruction

   https://doi.org/10.3390/bioengineering10030364  

   Levac, Brett R. ; Arvinte, Marius ; Tamir, Jonathan I. ( March 2023 , Bioengineering)

   Image reconstruction is the process of recovering an image from raw, under-sampled signal measurements, and is a critical step in diagnostic medical imaging, such as magnetic resonance imaging (MRI). Recently, data-driven methods have led to improved image quality in MRI reconstruction using a limited number of measurements, but these methods typically rely on the existence of a large, centralized database of fully sampled scans for training. In this work, we investigate federated learning for MRI reconstruction using end-to-end unrolled deep learning models as a means of training global models across multiple clients (data sites), while keeping individual scans local. We empirically identify a low-data regime across a large number of heterogeneous scans, where a small number of training samples per client are available and non-collaborative models lead to performance drops. In this regime, we investigate the performance of adaptive federated optimization algorithms as a function of client data distribution and communication budget. Experimental results show that adaptive optimization algorithms are well suited for the federated learning of unrolled models, even in a limited-data regime (50 slices per data site), and that client-sided personalization can improve reconstruction quality for clients that did not participate in training. 

   more » « less
3. Data heterogeneity in federated learning with Electronic Health Records: Case studies of risk prediction for acute kidney injury and sepsis diseases in critical care

   https://doi.org/10.1371/journal.pdig.0000117  

   Rajendran, Suraj ; Xu, Zhenxing ; Pan, Weishen ; Ghosh, Arnab ; Wang, Fei ( March 2023 , PLOS Digital Health)

   Frasch, Martin G. (Ed.)

   With the wider availability of healthcare data such as Electronic Health Records (EHR), more and more data-driven based approaches have been proposed to improve the quality-of-care delivery. Predictive modeling, which aims at building computational models for predicting clinical risk, is a popular research topic in healthcare analytics. However, concerns about privacy of healthcare data may hinder the development of effective predictive models that are generalizable because this often requires rich diverse data from multiple clinical institutions. Recently, federated learning (FL) has demonstrated promise in addressing this concern. However, data heterogeneity from different local participating sites may affect prediction performance of federated models. Due to acute kidney injury (AKI) and sepsis’ high prevalence among patients admitted to intensive care units (ICU), the early prediction of these conditions based on AI is an important topic in critical care medicine. In this study, we take AKI and sepsis onset risk prediction in ICU as two examples to explore the impact of data heterogeneity in the FL framework as well as compare performances across frameworks. We built predictive models based on local, pooled, and FL frameworks using EHR data across multiple hospitals. The local framework only used data from each site itself. The pooled framework combined data from all sites. In the FL framework, each local site did not have access to other sites’ data. A model was updated locally, and its parameters were shared to a central aggregator, which was used to update the federated model’s parameters and then subsequently, shared with each site. We found models built within a FL framework outperformed local counterparts. Then, we analyzed variable importance discrepancies across sites and frameworks. Finally, we explored potential sources of the heterogeneity within the EHR data. The different distributions of demographic profiles, medication use, and site information contributed to data heterogeneity. 

   more » « less
4. Detecting and Removing Sample Contamination in Phylogenomic Data: An Example and its Implications for Cicadidae Phylogeny (Insecta: Hemiptera)

   https://doi.org/10.1093/sysbio/syac043  

   Owen, Christopher L. ; Marshall, David C. ; Wade, Elizabeth J. ; Meister, Russ ; Goemans, Geert ; Kunte, Krushnamegh ; Moulds, Max ; Hill, Kathy ; Villet, M. ; Pham, Thai-Hong ; et al ( June 2022 , Systematic Biology)

   Contamination of a genetic sample with DNA from one or more nontarget species is a continuing concern of molecular phylogenetic studies, both Sanger sequencing studies and next-generation sequencing studies. We developed an automated pipeline for identifying and excluding likely cross-contaminated loci based on the detection of bimodal distributions of patristic distances across gene trees. When contamination occurs between samples within a data set, a comparison between a contaminated sample and its contaminant taxon will yield bimodal distributions with one peak close to zero patristic distance. This new method does not rely on a priori knowledge of taxon relatedness nor does it determine the causes(s) of the contamination. Exclusion of putatively contaminated loci from a data set generated for the insect family Cicadidae showed that these sequences were affecting some topological patterns and branch supports, although the effects were sometimes subtle, with some contamination-influenced relationships exhibiting strong bootstrap support. Long tip branches and outlier values for one anchored phylogenomic pipeline statistic (AvgNHomologs) were correlated with the presence of contamination. While the anchored hybrid enrichment markers used here, which target hemipteroid taxa, proved effective in resolving deep and shallow level Cicadidae relationships in aggregate, individual markers contained inadequate phylogenetic signal, in part probably due to short length. The cleaned data set, consisting of 429 loci, from 90 genera representing 44 of 56 current Cicadidae tribes, supported three of the four sampled Cicadidae subfamilies in concatenated-matrix maximum likelihood (ML) and multispecies coalescent-based species tree analyses, with the fourth subfamily weakly supported in the ML trees. No well-supported patterns from previous family-level Sanger sequencing studies of Cicadidae phylogeny were contradicted. One taxon (Aragualna plenalinea) did not fall with its current subfamily in the genetic tree, and this genus and its tribe Aragualnini is reclassified to Tibicininae following morphological re-examination. Only subtle differences were observed in trees after the removal of loci for which divergent base frequencies were detected. Greater success may be achieved by increased taxon sampling and developing a probe set targeting a more recent common ancestor and longer loci. Searches for contamination are an essential step in phylogenomic analyses of all kinds and our pipeline is an effective solution. [Auchenorrhyncha; base-composition bias; Cicadidae; Cicadoidea; Hemiptera; phylogenetic conflict.]

    

   more » « less
5. NEON Tree Crowns Dataset

   https://doi.org/10.5281/zenodo.3765871  

   Weinstein, Ben ; Marconi, Sergio ; Zare, Alina ; Bohlman, Stephanie ; Graves, Sarah ; Singh, Aditya ; White, Ethan ( January 2020 , Zenodo)

   Abstract

   The NeonTreeCrowns dataset is a set of individual level crown estimates for 100 million trees at 37 geographic sites across the United States surveyed by the National Ecological Observation Network’s Airborne Observation Platform. Each rectangular bounding box crown prediction includes height, crown area, and spatial location. 

   How can I see the data?

   A web server to look through predictions is available through idtrees.org

   Dataset Organization

   The shapefiles.zip contains 11,000 shapefiles, each corresponding to a 1km^2 RGB tile from NEON (ID: DP3.30010.001). For example "2019_SOAP_4_302000_4100000_image.shp" are the predictions from "2019_SOAP_4_302000_4100000_image.tif" available from the NEON data portal: https://data.neonscience.org/data-products/explore?search=camera. NEON's file convention refers to the year of data collection (2019), the four letter site code (SOAP), the sampling event (4), and the utm coordinate of the top left corner (302000_4100000). For NEON site abbreviations and utm zones see https://www.neonscience.org/field-sites/field-sites-map. 

   The predictions are also available as a single csv for each file. All available tiles for that site and year are combined into one large site. These data are not projected, but contain the utm coordinates for each bounding box (left, bottom, right, top). For both file types the following fields are available:

   Height: The crown height measured in meters. Crown height is defined as the 99th quartile of all canopy height pixels from a LiDAR height model (ID: DP3.30015.001)

   Area: The crown area in m² of the rectangular bounding box.

   Label: All data in this release are "Tree".

   Score: The confidence score from the DeepForest deep learning algorithm. The score ranges from 0 (low confidence) to 1 (high confidence)

   How were predictions made?

   The DeepForest algorithm is available as a python package: https://deepforest.readthedocs.io/. Predictions were overlaid on the LiDAR-derived canopy height model. Predictions with heights less than 3m were removed.

   How were predictions validated?

   Please see

   Weinstein, B. G., Marconi, S., Bohlman, S. A., Zare, A., & White, E. P. (2020). Cross-site learning in deep learning RGB tree crown detection. Ecological Informatics, 56, 101061.

   Weinstein, B., Marconi, S., Aubry-Kientz, M., Vincent, G., Senyondo, H., & White, E. (2020). DeepForest: A Python package for RGB deep learning tree crown delineation. bioRxiv.

   Weinstein, Ben G., et al. "Individual tree-crown detection in RGB imagery using semi-supervised deep learning neural networks." Remote Sensing 11.11 (2019): 1309.

   Were any sites removed?

   Several sites were removed due to poor NEON data quality. GRSM and PUUM both had lower quality RGB data that made them unsuitable for prediction. NEON surveys are updated annually and we expect future flights to correct these errors. We removed the GUIL puerto rico site due to its very steep topography and poor sunangle during data collection. The DeepForest algorithm responded poorly to predicting crowns in intensely shaded areas where there was very little sun penetration. We are happy to make these data are available upon request.

   # Contact

   We welcome questions, ideas and general inquiries. The data can be used for many applications and we look forward to hearing from you. Contact ben.weinstein@weecology.org. 

   Gordon and Betty Moore Foundation: GBMF4563 

   more » « less