The authors report the creation of DataPlex, a large, labeled structural MRI collection focused on the lateral ventricle choroid plexus (ChP). DataPlex comprises 700 brain MR images in which the lateral ventricle ChP has been manually delineated and verified by expert human raters. The stated purpose is to provide a comprehensive training resource to improve the accuracy and generalizability of automated ChP segmentation methods.
The dataset addresses a recognized gap: prior work lacked a sufficiently large, diverse set of labeled ChP images to support robust machine learning development. Because ChP morphology varies substantially across individuals, training sets that capture wide variability in anatomy, scanner parameters, and clinical profiles are essential for creating reliable segmentation tools.
The choroid plexus is a highly vascularized structure located within the brain ventricles that contributes to multiple central nervous system functions. The authors highlight that ChP enlargement has been implicated in several neuropsychiatric and neurodegenerative conditions, including multiple sclerosis, major depression, schizophrenia, and Alzheimer’s disease. Accurate quantification of ChP volume and morphology is therefore relevant for both basic neuroscience and clinical research.
According to the authors, current automated segmentation tools have limitations in accuracy when applied to the ChP. Contributing factors include pronounced inter-individual variability in ChP shape and size, as well as differences in image acquisition parameters across studies. These limitations motivate the creation of a large, labeled dataset that can support development and validation of improved segmentation algorithms.
DataPlex consists of 700 structural MRI scans with manual labels of the lateral ventricle ChP. Each labeled image was manually delineated and reviewed by expert human raters, ensuring label quality at the subject level. The authors state that subject-level information will be available for each labeled image, although specific metadata fields were not detailed in the source.
The dataset is presented as comprehensive and intended for use in training and evaluating segmentation models targeting the lateral ventricle choroid plexus. The authors emphasize that the size and manual verification of labels are central strengths of DataPlex for fostering more reliable segmentation approaches.
The authors describe DataPlex as balanced and representative across a wide range of scanning parameters, demographic variables, and clinical profiles. This diversity is intended to improve the external validity of segmentation models trained on the dataset. The source text reports that the collection spans varied scanners and participant clinical statuses, but it does not provide detailed breakdowns, numeric distributions, or specific scanner models in the preprint abstract.
DataPlex is planned to be released as an open-source resource. The authors indicate that subject-level information will accompany each labeled image to support downstream analyses. The preprint positions DataPlex as a community resource to accelerate the creation and benchmarking of reliable segmentation tools for the lateral ventricle choroid plexus. The specific repository, licensing details beyond the preprint’s CC-BY 4.0 statement, and the timing or mechanism of dataset release were not specified in the source abstract.
The work was supported by the National Institute of Mental Health (grant MH131586). The authors declared no competing interests. The dataset description and accompanying manuscript are available as a bioRxiv preprint (doi: 10.64898/2026.10.08.757728) posted October 11, 2026.
By providing 700 manually labeled MR images of the lateral ventricle choroid plexus, DataPlex aims to enable development of segmentation models with greater accuracy and generalizability across scanners and clinical populations. Such tools could facilitate more reliable quantification of ChP metrics in studies of psychiatric and neurodegenerative disorders where ChP changes have been implicated.
Note: The source material is the bioRxiv preprint abstract and associated metadata. Detailed dataset statistics, label protocols, inter-rater reliability metrics, and access instructions were not reported in the provided source text.