IPH
SDH (ABC/2 ÷ 2)
TOTAL SAHV — 5-Space Cisternal Volume
6-Space Total (All + IPH/SDH)
IPH + SDH Combined
Click any image to enlarge. Images from Freeman & Föttinger ABC/2-derived SAHV method.
| Region | Definition |
|---|---|
| 1. Suprasellar cistern | Region around the cranial sella turcica, under the hypothalamus and between the uncus of the temporal lobes |
| 2. Perimesencephalic cistern | Region of the ambient cistern and crural cistern, extending in a crescent-shaped manner along the lateral aspect of the mesencephalon |
| 3. Prepontine cistern | Region extending ventrally over the most caudal segment to the most cranial part of pons |
| 4. Sylvian cisternal complex | Region of the Sylvian cistern containing the lateral projection of the suprasellar cistern (along the temporal pole), the cistern of lateral cerebral fossa and its occipital projection into the Sylvian fissure; anterior part sits between the fronto-orbital cortex and the temporal lobe |
| 5. Anterior interhemispheric cistern | Region cranioventral to suprasellar cistern, until leveled with frontal horn of lateral ventricle |
| 6. IPH | Intraparenchymal hemorrhage (IPH), or intracerebral hemorrhage (ICH) |
| 7. IVH / SDH / EDH | Intraventricular hemorrhage (IVH) is semi-quantitatively measured by modified Graeb, and/or subdural hematoma (SDH), or epidural hemorrhage (EDH) components are measured by a modified ABC/2 formula using maximal EDH/SDH thickness and length as A and B, and vertical dimensions as C (Freeman et al., Neurocrit Care 2008; PMID: 18392784) |
| 8. Gyral-sulcal SAH blood (not measured) | Generally too small to contribute significant volume; not included in the above 5-compartment SAHV ABC/2 model by Föttinger/Freeman et al. |
B = maximal thickness/width of SAH (mm)
C = number of vertical NCHCT slices with SAH × slice thickness (mm)
For bilateral structures (Perimesencephalic, Sylvian, IPH, SDH), right and left are measured independently and summed.
Enhanced SAH (eSAH) Score Calculator
Weighted score (0–5) combining Age, GCS, and SAHV. AUC = 0.89 (mRS), 0.88 (mortality). Sharma, Freeman et al., Sci Rep 2024.
Morgan TC, Dawson J, Spengler D, et al. Stroke. 2013;44(3):635-41. doi:10.1161/STROKEAHA.112.670653
Score range: 0–32. Higher scores indicate greater IVH burden.
| Ventricle | Score (0–4) | Expansion (0/1) |
|---|---|---|
| Left Lateral (frontal horn) | ||
| Right Lateral (frontal horn) | ||
| Left Lateral (body) | ||
| Right Lateral (body) | ||
| Left Lateral (temporal horn) | ||
| Right Lateral (temporal horn) | ||
| Third Ventricle | ||
| Fourth Ventricle |
HEADS-UP — intracranial hemorrhage detector, step 1 of the suite
Hemorrhage Evaluation and Detector System for Underserved Populations (Salman et al., Mayo Clin Proc Digit Health 2023). Drop a non-contrast head CT to detect hemorrhage and classify epidural, intraparenchymal, intraventricular, subarachnoid and subdural blood slice by slice, with the RSNA 2019 1st-place networks. It then sends the scan to the segmentation tab that suits the type found. Runs locally, including on a phone browser — images are never uploaded.
Open full screen ↗Research use only — not a medical device and not a triage system.
SAHVAI Web — AI hemorrhage segmentation in your browser
Drag and drop a non-contrast head CT (DICOM study, folder or zip, NIfTI or NRRD) to segment SAH, IPH and IVH and measure their volumes with the Hybrid 2D/3D UNet (Hu et al., NeuroImage 2023). The model runs locally on your GPU or CPU — images are never uploaded.
Open full screen ↗Research use only — not a medical device. Embedded below, CPU mode runs on one thread; open full screen for multi-threaded CPU (WebGPU is fast in both).
SAHVAI 3D-4D — rotate the hemorrhage in 3D and track it over time
Drop a non-contrast head CT to see SAH, IPH and IVH in a rotatable 3D brain with the skull removed, alongside axial, coronal and sagittal slices. Drop 2–6 serial CTs of the same patient to chart SAHV, IPH, IVH and total blood volume by scan date (SAHVAI-4D; Wirtz et al., Stroke Vasc Interv Neurol 2025). Segmentation by the BrainHemoAI Hybrid 2D/3D UNet, run locally — images are never uploaded.
Open full screen ↗Research use only — not a medical device. Embedded below, CPU mode runs on one thread; open full screen for multi-threaded CPU (WebGPU is fast in both).
CLAIM aSAH — multiclass segmentation in your browser
The aneurysmal SAH nnU-Net of the CLAIM group, Charité Berlin (Kiewitz et al., Front Neurol 2024): SAH, IVH, ventricles, ICH, aneurysm and SDH. Weights CC BY-NC 4.0 (non-commercial). The model runs locally on your GPU or CPU — images are never uploaded.
Open full screen ↗Research and education use only — not a medical device. Embedded below, CPU mode runs on one thread; open full screen for multi-threaded CPU (WebGPU is fast in both).
BLAST-CT — traumatic brain injury lesions in your browser
The TBI lesion model of the University of Cambridge and Imperial College London (Monteiro et al., Lancet Digit Health 2020): intraparenchymal and extra-axial haemorrhage, oedema and IVH — for SAH after a fall or traumatic ICH. Apache-2.0. Runs locally; images are never uploaded.
Open full screen ↗Research and education use only — not a medical device. Embedded below, CPU mode runs on one thread; open full screen for multi-threaded CPU (WebGPU is fast in both).
DeepBleed — ICH segmentation in your browser
The open-source 3D V-Net for intracerebral hemorrhage from the MISTIE trial investigators at Johns Hopkins (Sharrock et al., Neuroinformatics 2021): ICH + IVH segmentation and volume, trained on MISTIE II and validated on MISTIE III. CC BY-NC 4.0 (non-commercial), with the permission of Dr Matthew Sharrock. Runs locally; images are never uploaded.
Open full screen ↗Research and education use only — not a medical device. Embedded below, CPU mode runs on one thread; open full screen for multi-threaded CPU (WebGPU is fast in both).
ICH Expansion Signs — CT predictors of hematoma expansion in your browser
Drop a head CT with a spontaneous ICH: the hemorrhage is segmented with SAHVAI, then the expansion signs are read inside and around the hematoma — blend, black hole, hypodensity, swirl, island, satellite, irregular shape, heterogeneous density and fluid level on non-contrast CT, spot-sign candidates on CTA — with the BAT, BRAIN and target expansion scores (the last needs the CTA spot sign, from a CTA run or the reader) and the image patches of the DeepMedX end-to-end expansion pipeline (Yu et al., npj Digit Med 2026). The spot, leakage and iodine signs are referenced with their definitions (Liu et al., GeroScience 2026). Runs locally; images are never uploaded.
Open full screen ↗Research and education use only — not a medical device. Signs are rule-based candidates for a reader to confirm. Embedded below, CPU mode runs on one thread; open full screen for multi-threaded CPU (WebGPU is fast in both).
Bedside score from five items available at the first CT; no images needed. Predicts clinically significant hematoma growth (≥ 6 mL absolute increase from baseline to 24 h). Derived in the INTERACT2 CT substudy (n = 964) and validated in INTERACT1 (n = 346; C-statistic 0.73, Hosmer–Lemeshow P = 0.82). To score a scan automatically, use the AI tool below, which fills in volume and IVH from the segmentation.
Points → predicted probability (Figure 1 of the paper)
Wang X, Arima H, Al-Shahi Salman R, et al; INTERACT Investigators. Clinical prediction algorithm (BRAIN) to determine risk of hematoma growth in acute intracerebral hemorrhage. Stroke. 2015;46(2):376-381. doi:10.1161/STROKEAHA.114.006910. PMID: 25503550
Imaging-based “target expansion criteria” for patients first imaged 2–12 h after onset or last seen well, by analogy with target mismatch in ischemic stroke. Three items from NCCT and CTA give a 0–4 score; a score ≥ 2 meets the target expansion criteria. Outcome: hematoma expansion > 33% and/or > 6 mL from baseline to follow-up NCCT (24–72 h). Derived and internally validated in a retrospective multicenter cohort from Italy, Germany and China (n = 345 development, 147 validation; AUC 0.77 and 0.79).
Expansion rate and mean growth by score (Table 5 of the paper)
| Score | Development (n = 345) | Validation (n = 147) | ||
|---|---|---|---|---|
| HE, n/N (%) | Mean growth | HE, n/N (%) | Mean growth | |
| 0 | 13/132 (9.8%) | 0.9 mL | 3/51 (5.9%) | 4.3 mL |
| 1 | 21/123 (17.1%) | 3.9 mL | 10/55 (18.2%) | 6.7 mL |
| 2 | 28/58 (48.3%) | 11.5 mL | 12/24 (50.0%) | 11.2 mL |
| 3 | 17/24 (70.8%) | 13.0 mL | 3/6 (50.0%) | 19.5 mL |
| 4 | 7/8 (87.5%) | 26.9 mL | 8/11 (72.7%) | 21.2 mL |
| < 2 | 34/255 (13.3%) | 2.3 mL | 13/106 (12.3%) | 5.5 mL |
| ≥ 2 | 52/90 (57.8%) | 13.3 mL | 23/41 (56.1%) | 15.1 mL |
Score ≥ 2: sensitivity 0.61, specificity 0.85 (development); 0.64 and 0.84 (validation). Adjusted ORs: high uHG 3.44, NCCT hypodensities 2.51, CTA spot sign 8.97.
Morotti A, Pensato U, Nawabi J, et al. Intracerebral hemorrhage expansion in late presenters: proposed target expansion criteria. Neurology. 2026;107(9):e218675. doi:10.1212/WNL.0000000000218675. PMID: 42879724
SAH Swin UNETR — Valladolid blood segmentation in your browser
The Swin transformer blood-segmentation model of the Río Hortega University Hospital group in Valladolid (García-García, Cepeda et al., World Neurosurg 2024), from their open SAH mortality-prediction pipeline: SAH blood and its extensions, with volume. Academic non-commercial licence; hosted with the permission of Dr Santiago Cepeda. Runs locally; images are never uploaded.
Open full screen ↗Research and education use only — not a medical device. A large model (277 MB) that runs on the CPU: open full screen for multi-threaded speed (about 1–2 minutes per scan).
RHUH SAH dataset — compare a model with expert masks
Open aSAH CT scans with expert bleeding masks from Río Hortega University Hospital, Valladolid (Cepeda & García-García, Zenodo 2023, CC BY 4.0). Load a case and its mask, run SAHVAI (BrainHemoAI), CLAIM, BLAST-CT or the Valladolid SAH Swin UNETR, and see Dice and volumes. Data stay on your computer.
Open full screen ↗Research and education use only — not a medical device. Embedded below, CPU mode runs on one thread; open full screen for multi-threaded CPU (WebGPU is fast in both).
Research Dataset — your site's hemorrhage volumetric data
Every tool (and the ABC/2 calculator) can add a scan to a de-identified dataset kept in this browser. Review it, add outcomes, and export it for REDCap, NIH ScHARe, Parquet, FHIR, OMOP or Convex. Nothing is uploaded.
Open full screen ↗Collect and share under your institution's approvals. Research and education use only — not a medical device.
SAHVAI-MM (MultiModal) Calculator
Integrated outcome prediction combining eSAH, CHESS, NLR, and SAHV–DCI risk
Weighted score (0–5) combining Age, GCS, and SAHV. AUC = 0.89 (mRS), 0.88 (mortality). Auto-fills from Calculator tab if populated, or enter directly below.
Hydrocephalus risk after SAH. Predicts need for ventriculoperitoneal shunt.
Inflammatory biomarker predicting DCI. NLR >12.03 cutoff (AUC 0.805).