Research and education use only
A research demonstration, not a medical device. This model is not validated or approved for clinical use here; do not use it for diagnosis, treatment or any other clinical decision. Segmentations can be wrong; always review them against the images.
Your images stay on this computer
Files are read and processed inside this browser tab on your GPU or CPU. No image data or results are uploaded. Only the page, the model files and the demo cases are downloaded, and the model is cached for later visits.
What you need
- A recent desktop Chrome or Edge (Firefox and Safari may work on the CPU).
- 8 GB RAM or more. WebGPU is fastest; on the CPU a scan takes one to several minutes.
- A non-contrast axial head CT in Hounsfield units.
DeepBleed MISTIE trial investigators · 3D V-Net for ICH
Intracerebral and intraventricular hemorrhage (one combined class) on non-contrast head CT, with volume: the first open-source 3D network for ICH segmentation, trained on MISTIE II and validated on MISTIE III (Dice 0.91).
Drop a head CT here
A zip of a DICOM study, a DICOM folder, loose DICOM files, a .nii / .nii.gz or a .nrrd volume. The brain is aligned to DeepBleed’s 1.5 mm, 128³ CT template, segmented there and mapped back to the scan.
Citation, licence and disclaimer
Model. Sharrock MF, Mould WA, Ali H, Hildreth M, Awad IA, Hanley DF, Muschelli J. 3D deep neural network segmentation of intracerebral hemorrhage: development and validation for clinical trials. Neuroinformatics 2021;19(3):403–415. doi:10.1007/s12021-020-09493-5. External validation: External validation and retraining of DeepBleed: the first open-source 3D deep learning network for the segmentation of spontaneous intracerebral and intraventricular hemorrhage. J Clin Med 2023 (median Dice 0.84 in 1,040 patients).
Code and weights. github.com/msharrock/deepbleed, MISTIE-2 weights linked from its README. Published under CC BY-NC-ND 4.0; this browser version is used under CC BY-NC 4.0 (attribution, non-commercial) with the permission of Dr Matthew Sharrock (November 2026), who agreed to drop the no-derivatives condition for it now that it runs in the browser and has been compared with the original. “This software is not intended for clinical use.” The V-Net weights are converted to ONNX in full precision without any change (a format conversion: same numbers, ONNX output matches TensorFlow on the same checkpoint with 0 voxels different); the 1.5 mm CT template is DeepBleed’s, unchanged. V-Net: Milletari F, Navab N, Ahmadi SA. 3DV 2016.
Demo data. Case 01 of: Hanley D, Mould WA, Ullman N, Muschelli J. Head CT scans associated with “Validated automatic brain extraction of head CT images”. Johns Hopkins Research Data Repository, V1, 2019. doi:10.7281/T1/CZDPSX (MISTIE II/III and CLEAR III scans), CC BY-NC 4.0, unchanged; see Muschelli J, et al. NeuroImage 2015;114:379–385. The BHSD and BLAST-CT demo cases and labels are those of the CLAIM and BLAST-CT pages.
How it runs here. As DeepBleed’s predict.py: brain-only image (outside 0) → rigid registration to the template (centres of mass, then mutual information, as ANTs) → linear resampling to 128³ at 1.5 mm → V-Net → mask > 0.5 → mapped back to the scan (linear interpolation of the mask ≥ 0.5, as ANTs’ genericLabel). It reproduces a quirk of the original that the network was trained with: the scan reaches ANTs turned 180° about the vertical axis. With the same brain mask, this port agrees with the original pipeline at Dice 0.93 (15.2 vs 15.1 mL) on the MISTIE demo; differences come from the registration and, without that mask, from the brain extraction.
Disclaimer. For research and education only. Not a medical device; not cleared or approved by the FDA, CE or any regulator; not validated for clinical use; must not be used for diagnosis, triage, treatment or any other clinical decision. Outputs can be wrong and must be reviewed against the images by a qualified reader. Provided "as is" without warranty of any kind. Images are processed locally in this browser and are not uploaded.