{ "@context": { "@language": "en", "@vocab": "https://schema.org/", "citeAs": "cr:citeAs", "column": "cr:column", "conformsTo": "dct:conformsTo", "cr": "http://mlcommons.org/croissant/", "rai": "http://mlcommons.org/croissant/RAI/", "data": { "@id": "cr:data", "@type": "@json" }, "dataType": { "@id": "cr:dataType", "@type": "@vocab" }, "dct": "http://purl.org/dc/terms/", "equivalentProperty": { "@id": "cr:equivalentProperty", "@type": "@vocab" }, "samplingRate": "cr:samplingRate", "examples": { "@id": "cr:examples", "@type": "@json" }, "extract": "cr:extract", "field": "cr:field", "fileProperty": "cr:fileProperty", "fileObject": "cr:fileObject", "fileSet": "cr:fileSet", "format": "cr:format", "includes": "cr:includes", "isLiveDataset": "cr:isLiveDataset", "jsonPath": "cr:jsonPath", "key": "cr:key", "md5": "cr:md5", "parentField": "cr:parentField", "path": "cr:path", "recordSet": "cr:recordSet", "references": "cr:references", "regex": "cr:regex", "repeated": "cr:repeated", "replace": "cr:replace", "sc": "https://schema.org/", "separator": "cr:separator", "source": "cr:source", "subField": "cr:subField", "transform": "cr:transform", "wd": "https://www.wikidata.org/wiki/", "prov": "http://www.w3.org/ns/prov#", "containedIn": "cr:containedIn", "value": { "@id": "cr:value", "@type": "@json" } }, "@type": "sc:Dataset", "conformsTo": "http://mlcommons.org/croissant/1.0", "version": "1.0.0", "datePublished": "2026-05-01", "dateModified": "2026-11-05", "isLiveDataset": false, "creator": [ { "@type": "sc:Person", "name": "Zhenghua Li", "affiliation": { "@type": "sc:Organization", "name": "Tsinghua University" } }, { "@type": "sc:Person", "name": "Lanyue Zhang", "affiliation": { "@type": "sc:Organization", "name": "Tsinghua University" } }, { "@type": "sc:Person", "name": "Zekang Yang", "affiliation": { "@type": "sc:Organization", "name": "Tsinghua University" } }, { "@type": "sc:Person", "name": "Kai Li", "affiliation": { "@type": "sc:Organization", "name": "Tsinghua University" } }, { "@type": "sc:Person", "name": "Yuwei Peng", "affiliation": { "@type": "sc:Organization", "name": "Tsinghua University" } }, { "@type": "sc:Person", "name": "Song-Hai Shi", "affiliation": { "@type": "sc:Organization", "name": "Tsinghua University" } }, { "@type": "sc:Person", "name": "Xiaolin Hu", "affiliation": { "@type": "sc:Organization", "name": "Tsinghua University" } } ], "publisher": { "@type": "sc:Organization", "name": "BrainEM authors", "url": "https://github.com/kwinderic/BrainEM" }, "citeAs": "Li, Zhenghua; Zhang, Lanyue; Yang, Zekang; Li, Kai; Peng, Yuwei; Shi, Song-Hai; Hu, Xiaolin (2026). BrainEM: A Large-Scale and Diverse Benchmark for EM Neuron Segmentation in Connectomics. NeurIPS 2026 Evaluations and Datasets Track.", "keywords": [ "electron microscopy", "connectomics", "neuron segmentation", "instance segmentation", "3D segmentation", "biomedical image analysis", "benchmark", "HDF5", "TIFF" ], "inLanguage": "en", "rai:dataUseCases": [ "Training and evaluating dense 3D neuron instance segmentation in electron microscopy volumes.", "Same-source evaluation with CREMI training and the held-out FlyWire crop; cross-source evaluation with a pooled training collection and held-out FIB25, AC3, Wafer4 and MC93 targets.", "Research on transfer across acquisition sources, species, imaging modalities and resolutions. No clinical diagnostic use is established." ], "rai:dataLimitations": [ "Source datasets differ in acquisition, resolution and annotation quality; pooled performance does not establish validity for every species or imaging setting.", "Small source regions and spatially correlated voxels limit independence; preserve the published source-level train/test exclusions to avoid leakage.", "Reference segmentations may contain residual errors. Results measure agreement with these references rather than error-free biological reconstruction.", "MC93 represents one animal, one brain region and one imaging session, without pathological conditions. Its 20-section axial extent limits spatial context. Partial proofreading cannot exclude residual annotation errors." ], "rai:dataBiases": [ "The collection is selected from available labeled EM sources, rather than sampled representatively across nervous systems. Species, brain regions and acquisition settings are unevenly represented.", "Different label-generation and proofreading procedures introduce source-dependent annotation biases." ], "rai:personalSensitiveInformation": "MC93 contains mouse brain EM imagery and neuron instance labels, with no human subjects or personal identifiers.", "rai:dataSocialImpact": "The benchmark supports more reproducible comparisons and development of neuron segmentation tools for connectomics. Errors and domain shifts can distort reconstructed connectivity. The data do not establish clinical validity; avoid unsupported biological or clinical generalizations and follow source-specific use restrictions.", "rai:hasSyntheticData": false, "rai:dataReleaseMaintenancePlan": "Data are publicly hosted at https://huggingface.co/datasets/kwinderic/BrainEM. Code and issue reporting are available at https://github.com/kwinderic/BrainEM. Corrections can be tracked through repository revisions; this metadata pins the current data revision for reproducibility.", "rai:dataPreprocessingProtocol": [ "HDF5 files: chunked to (16, 256, 256), gzip level 4, byte-shuffle filter; uint64 segment-id volumes downcast to uint32 only when max(id) < 2^32.", "TIFF files: kept as multi-page TIFF stacks; segmentation float32-with-integer-values converted to uint16 losslessly.", "Verification: per-file SHA-256 / np.array_equal round-trip check before replacing the original." ], "name": "BrainEM — MC93", "description": "An in-house mouse V1 ATUM-SEM volume with dense manual neuron instance labels, released as a Cross-Source test target. Array shape (z, y, x): (20, 3960, 3990). Part of the 16-subset BrainEM collection.", "url": "https://huggingface.co/datasets/kwinderic/BrainEM/tree/main/MC93", "identifier": "https://huggingface.co/datasets/kwinderic/BrainEM/tree/7f5a0784577f6e7c9ac6ac0d21a47fdd3614f35d/MC93", "license": "https://creativecommons.org/licenses/by/4.0/", "isPartOf": { "@type": "sc:Dataset", "name": "BrainEM", "url": "https://huggingface.co/datasets/kwinderic/BrainEM" }, "distribution": [ { "@type": "cr:FileObject", "@id": "mc93/9_3_78_97_Labels.tif", "name": "9_3_78_97_Labels.tif", "description": "Neuron instance labels.", "contentUrl": "https://huggingface.co/datasets/kwinderic/BrainEM/resolve/7f5a0784577f6e7c9ac6ac0d21a47fdd3614f35d/MC93/9_3_78_97_Labels.tif", "encodingFormat": "image/tiff", "contentSize": "316040320 B", "sha256": "53ee79beee896d148cbc36e2067cf8c53d1aae216ae3f70d8d5c5564d8db04a0", "license": "https://creativecommons.org/licenses/by/4.0/" }, { "@type": "cr:FileObject", "@id": "mc93/9_3_78_97_Stack.tif", "name": "9_3_78_97_Stack.tif", "description": "Raw electron microscopy image volume.", "contentUrl": "https://huggingface.co/datasets/kwinderic/BrainEM/resolve/7f5a0784577f6e7c9ac6ac0d21a47fdd3614f35d/MC93/9_3_78_97_Stack.tif", "encodingFormat": "image/tiff", "contentSize": "316040320 B", "sha256": "1c784e01502eb36f92dd309cb4248fce271b43574de94b044079649c467b8442", "license": "https://creativecommons.org/licenses/by/4.0/" } ], "recordSet": [ { "@type": "cr:RecordSet", "@id": "mc93/volume_pairs", "name": "mc93_volume_pairs", "description": "One record per raw EM volume and matching dense neuron instance-label volume. Inline records provide paired download locations, checksums and file formats. Volume contents remain in the referenced HDF5/TIFF files: read HDF5 with h5py and all TIFF pages with tifffile. This manifest does not claim native voxel-array decoding by a Croissant reader. Label IDs identify instances within a volume; background is 0. The filename partition is distinct from benchmark track roles.", "key": { "@id": "mc93/volume_pairs/pair_id" }, "field": [ { "@type": "cr:Field", "@id": "mc93/volume_pairs/pair_id", "name": "pair_id", "description": "Identifier of the image/label volume pair.", "dataType": "sc:Text" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/image_path", "name": "image_path", "description": "Path of the raw EM volume in the dataset repository.", "dataType": "sc:Text" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/label_path", "name": "label_path", "description": "Path of the dense neuron instance-label volume in the dataset repository.", "dataType": "sc:Text" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/image_url", "name": "image_url", "description": "Direct download URL of the raw volume, pinned to the dataset revision.", "dataType": "sc:URL" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/label_url", "name": "label_url", "description": "Direct download URL of the instance-label volume, pinned to the dataset revision.", "dataType": "sc:URL" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/image_sha256", "name": "image_sha256", "description": "SHA-256 of the raw-volume file bytes.", "dataType": "sc:Text" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/label_sha256", "name": "label_sha256", "description": "SHA-256 of the instance-label file bytes.", "dataType": "sc:Text" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/image_bytes", "name": "image_bytes", "description": "Size of the raw-volume file in bytes.", "dataType": "sc:Integer" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/label_bytes", "name": "label_bytes", "description": "Size of the instance-label file in bytes.", "dataType": "sc:Integer" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/encoding_format", "name": "encoding_format", "description": "MIME type of both files.", "dataType": "sc:Text" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/axes", "name": "axes", "description": "Axis order when read as a volume array.", "dataType": "sc:Text" }, { "@type": "cr:Field", "@id": "mc93/volume_pairs/partition", "name": "partition", "description": "Subdivision encoded in the filename; this is not a universal benchmark track assignment. See the paper/code for track roles.", "dataType": "sc:Text" } ], "data": [ { "mc93/volume_pairs/pair_id": "MC93/9_3_78_97_Stack", "mc93/volume_pairs/image_path": "MC93/9_3_78_97_Stack.tif", "mc93/volume_pairs/label_path": "MC93/9_3_78_97_Labels.tif", "mc93/volume_pairs/image_url": "https://huggingface.co/datasets/kwinderic/BrainEM/resolve/7f5a0784577f6e7c9ac6ac0d21a47fdd3614f35d/MC93/9_3_78_97_Stack.tif", "mc93/volume_pairs/label_url": "https://huggingface.co/datasets/kwinderic/BrainEM/resolve/7f5a0784577f6e7c9ac6ac0d21a47fdd3614f35d/MC93/9_3_78_97_Labels.tif", "mc93/volume_pairs/image_sha256": "1c784e01502eb36f92dd309cb4248fce271b43574de94b044079649c467b8442", "mc93/volume_pairs/label_sha256": "53ee79beee896d148cbc36e2067cf8c53d1aae216ae3f70d8d5c5564d8db04a0", "mc93/volume_pairs/image_bytes": 316040320, "mc93/volume_pairs/label_bytes": 316040320, "mc93/volume_pairs/encoding_format": "image/tiff", "mc93/volume_pairs/axes": "z,y,x", "mc93/volume_pairs/partition": "full_or_unsplit" } ] } ], "rai:dataCollection": "MC93 was acquired by the authors from C57BL/6 mouse primary visual cortex (V1). Preparation used transcardial perfusion fixation and rOTO heavy-metal staining. ATUM sections were imaged on a SEMNova-1000 at 2 kV, with 4 nm XY sampling and 50 nm section spacing. Tiles of 12000 x 12000 pixels, with 12.5% overlap, formed a 23 x 11 grid. The released volume has array shape (z, y, x) = (20, 3960, 3990). Acquisition took place from September to December 2022 under IACUC protocol 22-SSH1.", "rai:dataCollectionType": "Original EM acquisition and manual neuron instance annotation.", "rai:dataAnnotationProtocol": "A single annotator with PhD-level neuroscience training performed dense tracing and instance segmentation using TrakEM2 in Fiji/ImageJ, following an internal protocol. Annotation took approximately four weeks. A senior expert proofread approximately 5-10% of the volume for major errors. No independent double annotation, adjudication or formal inter-annotator agreement measurement was performed.", "rai:dataAnnotationPlatform": "TrakEM2 in Fiji/ImageJ.", "prov:wasGeneratedBy": [ { "@type": "prov:Activity", "prov:label": "MC93 acquisition", "prov:description": "MC93 was acquired by the authors from C57BL/6 mouse primary visual cortex (V1). Preparation used transcardial perfusion fixation and rOTO heavy-metal staining. ATUM sections were imaged on a SEMNova-1000 at 2 kV, with 4 nm XY sampling and 50 nm section spacing. Tiles of 12000 x 12000 pixels, with 12.5% overlap, formed a 23 x 11 grid. The released volume has array shape (z, y, x) = (20, 3960, 3990). Acquisition took place from September to December 2022 under IACUC protocol 22-SSH1." }, { "@type": "prov:Activity", "prov:label": "MC93 annotation and quality control", "prov:description": "A single annotator with PhD-level neuroscience training performed dense tracing and instance segmentation using TrakEM2 in Fiji/ImageJ, following an internal protocol. Annotation took approximately four weeks. A senior expert proofread approximately 5-10% of the volume for major errors. No independent double annotation, adjudication or formal inter-annotator agreement measurement was performed." } ] }