UTHealth - Endometriosis MRI Dataset (UT-EndoMRI)

Introduction Magnetic Resonance Imaging (MRI) is widely recommended as a primary non-invasive diagnostic tool for endometriosis. Endometriomas affect 17–44% of women diagnosed with the condition. Accurate MRI-based ovary segmentation in endometriosis patients is essential for detecting endometriomas, guiding surgery, and predicting post-operative complications. However, ovary segmentation becomes challenging when the ovary is deformed or absent, often due to surgical resection, emphasizing the need for highly experienced clinicians. An automatic segmentation pipeline for pelvic MRI in endometriosis patients could greatly reduce the manual workload for clinicians and help standardize ovary segmentation. The UTHealth Endometriosis MRI Dataset (UT-EndoMRI) includes multi-sequence MRI scans and structural labels collected from two clinical institutions, Memorial Hermann Hospital System and Texas Children’s Hospital Pavilion for Women. The first dataset comprises MRI scans and labels from 51 patients collected before 2022, featuring T2-weighted and T1-weighted fat-suppressed MRI sequences. The uterus, ovaries, endometriomas, cysts, and cul-de-sac structures were manually segmented by three raters. The second dataset, collected in 2022, consists of MRI scans and labels from 82 endometriosis patients. These sequences include T1-weighted, T1-weighted fat suppression, T2-weighted, and T2-weighted fat suppression MRI. In this dataset, the uterus, ovaries, and endometriomas were manually contoured by a single rater. Using these datasets, we investigated interrater agreement and developed an automatic ovary segmentation pipeline, RAovSeg, for endometriosis. The study and the data sharing were approved by the Committee for the Protection of Human Subjects at UTHealth (protocol no. HSC-SBMI-22-0184). The UT-EndoMRI dataset is available for free use exclusively in non-commercial scientific research. Endometriosis MRI This dataset includes MRI scans and labels from two clinical institutions. The data from the first institution can be found in the ```D1_MHS/ ```directory, while the data from the second institution are located in the ```D2_TCPW/``` directory. Each subfolder contains MRI scans and corresponding labels from different raters. The naming conventions for the files are as follows: MRI scans:D[dataset ID]- [patient ID] _ [MRI sequence].nii.gz Anatomical structure labels:D[dataset ID]- [patient ID] _ [structure name] _ r[rater ID].nii.gz For the labels in the ```D2_TCPW/ ```directory, since they were generated by a single rater, there is no rater ID included in the file names. The abbreviations used for naming:T1: T1-weighted MRIT1FS: T1-weighted fat suppression MRIT2: T2-weighted MRIT2FS: T2-weighted fat suppression MRIov: ovaryut: uterusem: endometriomacy: cystcds: cul de sac For example, the file located at ```UT-EndoMRI/D1_MHS/D1-000/D1-000_T1FS.nii.gz```represents the T1-weighted fat suppression MRI for subject 000 in dataset 1. The file at ```UT-EndoMRI/D1_MHS/D1-000/D1-000_ ut_r1.nii.gz``` is the uterus segmentation manually contoured by rater 1 for subject 000 in dataset 1. The file at```UT-EndoMRI/ D2_TCPW/D2-006/D2-006_ cy.nii.gz``` is the cyst segmentation manually contoured for subject 006 in dataset 2. MRI sequences may be missing due to a lack of acquisition. The MR Scanner information for Dataset 1 is available in 'SiteScannerInfo.csv'. Train/Validation/Test Replication The data split for RAovSeg training, validation, and testing is provided as follows:- Training/validation subjects IDs: D2-000 – D2-007- Testing subjects IDs: D2-008 – D2-037All data in dataset 1, as well as other data in dataset 2, are not used in RAovSeg development. Data Acquisition This dataset was acquired at the Texas Medical Center, within the Memorial Hermann Hospital System and the Texas Children’s Hospital Pavilion for Women. The study and the data sharing were approved by the Committee for the Protection of Human Subjects at UTHealth (protocol no. HSC-SBMI-22-0184). User Agreement The UT-EndoMRI dataset is available for free use exclusively in non-commercial scientific research. Any publications resulting from its use must cite the following paper.Liang, X., Alpuing Radilla, L.A., Khalaj, K. et al. A Multi-Modal Pelvic MRI Dataset for Deep Learning-Based Pelvic Organ Segmentation in Endometriosis. Sci Data 12, 1292 (2025). https://doi.org/10.1038/s41597-025-05623-3 Funding This work has been supported by the Robert and Janice McNair Foundation. Research Team Here are the people behind this data acquisition effort:Xiaomin Liang, Linda A Alpuing Radilla, Kamand Khalaj, Haaniya Dawoodally, Chinmay Mokashi, Xiaoming Guan, Kirk E Roberts, Sunil A Sheth, Varaha S Tammisetti, Luca Giancardo Acknowledgements We would also like to acknowledge for their support: Memorial Hermann Hospital System and Texas Children’s Hospital Pavilion for Women.