{"success":true,"data":{"author":"Cosarinsky, Gaggion, Ferrante et al. (PhysioNet)","bucketAccessType":"public","bucketListable":true,"bucketName":"physionet-open","bucketPrefix":"chexmask-u","category":"Health & Biomedical","cloudProvider":"aws","createdAt":"2026-11-16T14:11:27.946Z","description":"CheXmask-U (PhysioNet v1.0.0, 2026) extends the CheXmask database with node-wise uncertainty for anatomical segmentation of chest radiographs (CXR). It covers images from five public sources, ChestX-ray8 (NIH), CheXpert, MIMIC-CXR-JPG, PadChest and VinDr-CXR, and for each image gives landmark-based contours of the left lung, right lung and heart predicted by the HybridGNet graph neural network. Uncertainty comes from Monte Carlo sampling of the HybridGNet latent space: 50 stochastic landmark predictions per image, summarised as per-landmark mean [x,y] positions and standard deviations. Pixel masks are filled from the mean contours and stored run-length encoded (RLE), with mask height and width, plus Reverse Classification Accuracy (RCA) quality scores (Dice RCA max and mean; the authors suggest Dice RCA mean >= 0.7). Uncertainty was validated against 255 doubly annotated radiographs, synthetic noise and occlusion, and out-of-distribution tests. Layout: OriginalResolution/ holds one large CSV per source (CheXpert_1/2, ChestX-Ray8, MIMIC-CXR-JPG_1..3, Padchest, VinDr-CXR, 0.9 to 4.4 GB each) in original image coordinates, Preprocessed/ holds masks at the standardised 1024x1024 resolution; 39.9 GB uncompressed, read with pandas. The images themselves are not included and must be obtained from each source under its own terms. Masks released under CC BY 4.0; cite the CheXmask-U PhysioNet record and Gaggion et al. (CheXmask, Scientific Data 2024).","displayTitle":"CheXmask-U: Chest X-ray Lung and Heart Segmentation Masks with Landmark Uncertainty","doi":"10.13026/6vn3-3j51","fuseMountBrowsable":true,"id":"cb70b193-96a3-50fc-aad7-40c748361ab7","license":"CC-BY-4.0","licenseEvidenceUrl":"https://physionet.org/content/chexmask-u/1.0.0/","objectCountEstimate":1,"probeStatus":"ok","referenceUrl":"https://physionet.org/content/chexmask-u/1.0.0/","region":"us-east-1","schemaCard":{"accessHints":[{"language":"python-anonymous","snippet":"import boto3\nfrom botocore import UNSIGNED\nfrom botocore.client import Config\ns3 = boto3.client('s3', region_name='us-east-1', config=Config(signature_version=UNSIGNED))\ns3.list_objects_v2(Bucket='physionet-open', Prefix='chexmask-u/', Delimiter='/')"},{"language":"python-local","snippet":"open('/mnt/archive/1.0.0/OriginalResolution/CheXpert_1.csv', 'rb')"}],"caveats":["Files are binary — use a domain-specific parser (e.g. metpy / pynexrad / pydicom) rather than text I/O."],"format":"binary","generatedAt":"2026-11-16T14:11:32.868Z","layout":{"example":"chexmask-u/1.0.0/OriginalResolution/CheXpert_1.csv","pattern":"{prefix}/{seg1}/{seg2}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"Path,Dice RCA (Mean),Dice RCA (Max),Landmarks (Mean),Landmarks (Std),Left Lung,Right Lung,Heart,Height,Width\ntrain/patient40255/study2/view1_frontal.jpg,0.8877833598970419,0.916566992898438,\"983,251,895,265,815,298,743,351,663,417,594,492,539,569,497,646,475,718,461,784,439,856,417,933,400,1000,384,1072,362,1152,351,1221,334,1290,323,1362,312,1442,279,1522,268,1599,282,1646,326,1588,387,1513,458,1464,552,1450,641,1450,724,1453,815,1464,884,1455,906,1375,911,1284,928,1201,956,1113,975,1027,980,942,980,853,99","encoding":"utf8","key":"chexmask-u/1.0.0/OriginalResolution/CheXpert_1.csv","truncatedAt":512},{"bytes":"Path,Dice RCA (Mean),Dice RCA (Max),Landmarks (Mean),Landmarks (Std),Left Lung,Right Lung,Heart,Height,Width\ntrain/patient32603/study9/view1_frontal.jpg,0.8293813176986404,0.8815309409427052,\"1182,251,1096,262,1016,298,945,348,875,406,812,464,773,533,748,602,726,668,707,735,674,801,652,870,646,933,643,1011,627,1077,621,1135,610,1196,599,1259,591,1331,566,1395,552,1461,566,1489,610,1428,666,1342,737,1265,834,1237,936,1234,1038,1240,1135,1262,1193,1279,1207,1226,1201,1146,1215,1074,1268,991,1331,920,1370,845,","encoding":"utf8","key":"chexmask-u/1.0.0/OriginalResolution/CheXpert_2.csv","truncatedAt":512}],"sampleObjects":["chexmask-u/1.0.0/OriginalResolution/CheXpert_1.csv","chexmask-u/1.0.0/OriginalResolution/CheXpert_2.csv","chexmask-u/1.0.0/OriginalResolution/ChestX-Ray8.csv","chexmask-u/1.0.0/OriginalResolution/MIMIC-CXR-JPG_1.csv","chexmask-u/1.0.0/OriginalResolution/MIMIC-CXR-JPG_2.csv"],"sidecars":[{"bytes":"# CheXmask Database Data Dictionary\n\n## CSV File Structure\nEach dataset in the CheXmask Database is provided as a separate CSV file. Below are detailed descriptions of all fields/variables present in these files.\n\n### Image Identification\n**Field Name**: Image ID  \n**Description**: Reference identifier linking to original image in source dataset  \n**Type**: String  \n**Notes**: Format varies by source dataset (ChestX-ray8, CheXpert, MIMIC-CXR-JPG, Padchest, VinDr-CXR)  \n\n### Quality Metrics\n\n**Field Name**: Dice RCA (Max)  \n**Description**: Maximum Dice Similarity Coefficient from Reverse Classification Accuracy  \n**Type**: Float  \n**Range**: 0.0 to 1.0  \n**Units**: Dimensionless  \n**Notes**: Higher values indicate better segmentation quality  \n\n**Field Name**: Dice RCA (Mean)  \n**Description**: Mean Dice Similarity Coefficient from Reverse Classification Accuracy  \n**Type**: Float  \n**Range**: 0.0 to 1.0  \n**Units**: Dimensionless  \n**Notes**: Recommended threshold for use is >= 0.7\n\n### Anatomical Features\n\n**Field Name**: Landmarks (Mean)  \n**Description**: Mean set of points representing organ contours generated by HybridGNet model  \n**Type**: Array of coordinates  \n**Format**: JSON array of [x,y] coordinates  \n**Units**: Pixels  \n**Notes**: Same format as original Landmarks; represents mean positions of each landmark  \n\n**Field Name**: Landmarks (Std)  \n**Description**: Standard deviation for each landmark coordinate  \n**Type**: Array of floats  \n**Format**: JSON array of [x_std, y_std] values  \n**Units**: Pixels  \n**Notes**: Represents uncertainty in each coordinate; same structure as Landmarks  \n\n**Field Name**: Left Lung  \n**Description**: Segmentation mask for left lung  \n**Type**: String  \n**Format**: Run-length encoding (RLE)  \n**Notes**: Must be decoded using provided dimensions  \n\n**Field Name**: Right Lung  \n**Description**: Segmentation mask for right lung  \n**Type**: String  \n**Format**: Run-length encoding (RLE)  \n**Notes**: Must be decoded using provided dimensions  \n\n**Field Name**: Heart  \n**Description**: Segmentation mask for heart  \n**Type**: String  \n**Format**: Run-length encoding (RLE)  \n**Notes**: Must be decoded using provided dimensions  \n\n### Image Dimensions\n\n**Field Name**: Height  \n**Description**: Height of segmentation mask  \n**Type**: Integer  \n**Units**: Pixels  \n**Notes**: Required for decoding RLE masks  \n\n**Field Name**: Width  \n**Description**: Width of segmentation mask  \n**Type**: Integer  \n**Units**: Pixels  \n**Notes**: Required for decoding RLE masks  \n\n## Additional Notes\n\n1. **RLE Format**: Run-length encoding is used to compress the binary mask data. Each RLE string represents pairs of (start_position, run_length) for the mask. A decoding script is present in the GitHub repository.\n","encoding":"utf8","key":"chexmask-u/1.0.0/DATA_DICTIONARY.md","name":"DATA_DICTIONARY.md","size":2778,"truncated":false},{"bytes":"Creative Commons Attribution 4.0 International Public License\r\n\r\nBy exercising the Licensed Rights (defined below), You accept and agree\r\nto be bound by the terms and conditions of this Creative Commons\r\nAttribution 4.0 International Public License (\"Public License\"). To the\r\nextent this Public License may be interpreted as a contract, You are\r\ngranted the Licensed Rights in consideration of Your acceptance of\r\nthese terms and conditions, and the Licensor grants You such rights in\r\nconsideration of benefits the Licensor receives from making the\r\nLicensed Material available under these terms and conditions.\r\n\r\n\r\nSection 1 -- Definitions.\r\n\r\n  a. Adapted Material means material subject to Copyright and Similar\r\n     Rights that is derived from or based upon the Licensed Material\r\n     and in which the Licensed Material is translated, altered,\r\n     arranged, transformed, or otherwise modified in a manner requiring\r\n     permission under the Copyright and Similar Rights held by the\r\n     Licensor. For purposes of this Public License, where the Licensed\r\n     Material is a musical work, performance, or sound recording,\r\n     Adapted Material is always produced where the Licensed Material is\r\n     synched in timed relation with a moving image.\r\n\r\n  b. Adapter's License means the license You apply to Your Copyright\r\n     and Similar Rights in Your contributions to Adapted Material in\r\n     accordance with the terms and conditions of this Public License.\r\n\r\n  c. Copyright and Similar Rights means copyright and/or similar rights\r\n     closely related to copyright including, without limitation,\r\n     performance, broadcast, sound recording, and Sui Generis Database\r\n     Rights, without regard to how the rights are labeled or\r\n     categorized. For purposes of this Public License, the rights\r\n     specified in Section 2(b)(1)-(2) are not Copyright and Similar\r\n     Rights.\r\n\r\n  d. Effective Technological Measures means those measures that, in the\r\n     absence of proper authority, may not be circumvented under laws\r\n     fulfilling obligations under Article 11 of the WIPO Copyright\r\n     Treaty adopted on December 20, 1996, and/or similar international\r\n     agreements.\r\n\r\n  e. Exceptions and Limitations means fair use, fair dealing, and/or\r\n     any other exception or limitation to Copyright and Similar Rights\r\n     that applies to Your use of the Licensed Material.\r\n\r\n  f. Licensed Material means the artistic or literary work, database,\r\n     or other material to which the Licensor applied this Public\r\n     License.\r\n\r\n  g. Licensed Rights means the rights granted to You subject to the\r\n     terms and conditions of this Public License, which are limited to\r\n     all Copyright and Similar Rights that apply to Your use of the\r\n     Licensed Material and that the Licensor has authority to license.\r\n\r\n  h. Licensor means the individual(s) or entity(ies) granting rights\r\n     under this Public License.\r\n\r\n  i. Share means to provide material to the public by any means or\r\n     process that requires permission under the Licensed Rights, such\r\n     as reproduction, public display, public performance, distribution,\r\n     dissemination, communication, or importation, and to make material\r\n     available to the public including in ways that members of the\r\n     public may access the material from a place and at a time\r\n     individually chosen by them.\r\n\r\n  j. Sui Generis Database Rights means rights other than copyright\r\n     resulting from Directive 96/9/EC of the European Parliament and of\r\n     the Council of 11 March 1996 on the legal protection of databases,\r\n     as amended and/or succeeded, as well as other essentially\r\n     equivalent rights anywhere in the world.\r\n\r\n  k. You means the individual or entity exercising the Licensed Rights\r\n     under this Public License. Your has a corresponding meaning.\r\n\r\n\r\nSection 2 -- Scope.\r\n\r\n  a. License grant.\r\n\r\n       1. Subject to the terms and conditions of this Public License,\r\n          the Licensor hereby grants You a worldwide, royalty-free,\r\n          non-sublicensable, non-exclus","encoding":"utf8","key":"chexmask-u/1.0.0/LICENSE.txt","name":"LICENSE.txt","size":14842,"truncated":true},{"bytes":"# CheXmask-U Database v1.0.0\n\nA large-scale collection of anatomical segmentation masks for chest X-rays with node-wise uncertainty estimates.\n\n## Overview\n\nCheXmask-U provides anatomical segmentation masks derived from chest radiographs across multiple public databases:\n- ChestX-ray8\n- Chexpert\n- MIMIC-CXR-JPG\n- Padchest\n- VinDr-CXR\n\nEach image includes **landmark-based uncertainty estimates** generated using the HybridGNet framework. The dataset also includes quality metrics based on Reverse Classification Accuracy (RCA).\n\n## Dataset Structure\n\nThe dataset consists of CSV files for each source database. Each CSV contains:\n\n| Column Name | Description |\n|------------|-------------|\n| Image ID | Reference to original image in source dataset |\n| Dice RCA (Max) | Maximum Dice Similarity Coefficient for RCA |\n| Dice RCA (Mean) | Mean Dice Similarity Coefficient for RCA |\n| Landmarks (Mean) | Mean positions of anatomical landmarks across stochastic predictions |\n| Landmarks (Std) | Standard deviation of landmarks positions, representing node-wise uncertainty |\n| Left Lung | Left lung segmentation mask in RLE format |\n| Right Lung | Right lung segmentation mask in RLE format |\n| Heart | Heart segmentation mask in RLE format |\n| Height | Height of segmentation mask |\n| Width | Width of segmentation mask |\n\n## Data Processing\n\nAll images were processed to maintain consistent quality:\n1. Images were preprocessed to 1024x1024 resolution\n2. HybridGNet model was applied to generate landmark predictions\n3. **N=50 stochastic landmark predictions per image were sampled from the latent space**\n4. Landmarks mean and standard deviation were computed\n5. Pixel-level masks were derived from mean landmarks\n6. RCA scores were calculated for quality assessment\n\n## Usage Guidelines\n\n1. **Source Images**: Users must obtain source images from the original databases and comply with their access requirements, including any ethical training or courses.\n2. **Quality Threshold**: For analysis, use only segmentation masks with Dice RCA (Mean) ≥ 0.7\n3. **Uncertainty Information**: The `Landmarks (Std)` column provides node-wise uncertainty; users may consider the average landmark standard deviation per image as an additional criterion for filtering uncertain predictions.\n4. **Resolution**: Pre-processed masks (1024x1024) are included to ensure consistent resolution across datasets.\n\n\n","encoding":"utf8","key":"chexmask-u/1.0.0/README.md","name":"README.md","size":2396,"truncated":false}],"sourceObjectKey":"chexmask-u/1.0.0/OriginalResolution/CheXpert_1.csv","starterCell":"with open('/mnt/archive/1.0.0/OriginalResolution/CheXpert_1.csv', 'rb') as fh:\n    head = fh.read(1024)\nprint(len(head), head[:64])","starterSamplePath":"1.0.0/OriginalResolution/CheXpert_1.csv","totalObjects":"1 top-level prefixes"},"summary":"CheXmask-U (PhysioNet v1.0.0, 2026) extends the CheXmask database with node-wise uncertainty for anatomical segmentation of chest radiographs (CXR).","title":"CheXmask-U: Chest X-ray Lung and Heart Segmentation Masks with Landmark Uncertainty","updatedAt":"2026-11-16T16:18:20.635Z","visibility":"public","websiteUrl":"https://github.com/mcosarinsky/CheXmask-U"}}