Database Open Access
Computed Tomography Images for Intracranial Hemorrhage Detection and Segmentation
Published: Nov. 14, 2019. Version: 1.2.0 <View latest version>
When using this resource, please cite:
(show more options)
Hssayeni, M. (2019). Computed Tomography Images for Intracranial Hemorrhage Detection and Segmentation (version 1.2.0). PhysioNet. https://doi.org/10.13026/w8q8-ky94.
Hssayeni, M. D., Croock, M. S., Al-Ani, A., Al-khafaji, H. F., Yahya, Z. A., & Ghoraani, B. (2019). Intracranial Hemorrhage Segmentation Using Deep Convolutional Model. arXiv preprint arXiv:1910.08643.
Please include the standard citation for PhysioNet:
(show more options)
Goldberger, A., Amaral, L., Glass, L., Hausdorff, J., Ivanov, P. C., Mark, R., ... & Stanley, H. E. (2000). PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation [Online]. 101 (23), pp. e215–e220.
After traumatic brain injury (TBI), intracranial hemorrhage (ICH) may occur that could lead to death or disability if it is not accurately diagnosed and treated in a time-sensitive procedure. Currently, Computerized Tomography (CT) scans are examined by radiologists to diagnose intracranial hemorrhage to localize affected regions. In this work, we collected a dataset of 82 CT scans of patients with traumatic brain injury. Intracranial hemorrhage regions in these scans were delineated in each slice by two radiologists. The radiologists also annotated each CT slice for the presence of different types of intracranial hemorrhage and fracture.
Traumatic brain injury (TBI) is a major cause of death and disability in the United States, contributing to about 30% of all injury deaths as of 2013 . After accidents that involve TBI, extra-axial intracranial lesions may present, such as intracranial hemorrhage (ICH). An ICH is a critical medical lesion that is associated with a high rate of mortality . ICH is considered to be clinically dangerous because of its high risk of turning into a secondary brain insult that may lead to paralysis and death if it is not treated appropriately. Hemorrhages can be classified based upon their location in the brain: Intraventricular (IVH), Intraparenchymal (IPH), Subarachnoid (SAH), Epidural (EDH) and Subdural (SDH).
Computerized Tomography (CT) scans are commonly used in the emergency evaluation of patients with TBI to diagnose intracranial hemorrhage by capturing multiple layers of the brain . The availability of CT scans and their rapid acquisition time makes CT a preferred diagnostic tool over Magnetic Resonance Imaging (MRI) for initial hemorrhage assessment. CT scans generate a sequence of images using X-ray beams where brain tissues are captured with different intensities depending on the amount of X-ray absorbency of the tissue. CT scans are displayed using a windowing method, which converts Hounsfield units (HU) into grayscale values ([0, 255]) using two parameters: window level (WL) and window width (WW). Different windows allow different features of tissues to be displayed in a grayscale image (e.g., brain window, stroke window, or a bone window) . In CT scans using brain windows, hemorrhages appear as hyper intense regions with relatively undefined structure. CT images are examined by senior radiologists to determine whether a hemorrhage has occurred and if so, to detect the type and its region. However, this process can be lengthy, and subspecialty-trained neuroradiologists may not always be available to make an assessment.
Convolutional neural networks (CNN) have been shown to have excellent performance in automating multiple image classification and segmentation tasks . We hypothesized that deep learning algorithms have the potential to automate the diagnosing procedure of segmenting the ICH regions and detecting skull fracture, reducing the time taken for ICH diagnosis and potentially improving its accuracy. An automated ICH screen tool could be used to assist radiologists with less experience in detecting hemorrhage types, or when experts are not immediately available in the emergency room, which is especially prevalent in developing countries or remote areas.
A retrospective study was designed to collect head CT scans of subjects with TBI and it was approved by the research and ethics board in the Iraqi Ministry of Health, Babil Office (approval #1369). The inclusion criteria were any subject who was admitted to the hospital emergency unit with a TBI, and a CT scan was performed to him/her. CT scans were collected between February and August 2018 from Al Hilla Teaching Hospital, Iraq. Sensitive information for each patient was anonymized.
A dataset of 82 CT scans was collected, including 36 scans for patients diagnosed with intracranial hemorrhage with the following types: Intraventricular, Intraparenchymal, Subarachnoid, Epidural and Subdural. Each CT scan for each patient includes about 30 slices with 5 mm slice-thickness. The mean and std of patients' age were 27.8 and 19.5, respectively. 46 of the patients were males and 36 of them were females. Each slice of the non-contrast CT scans was by two radiologists who recorded hemorrhage types if hemorrhage occurred or if a fracture occurred. The radiologists also delineated the ICH regions in each slice. There was a consensus between the radiologists. Radiologists did not have access to clinical history of the patients.
During data collection, syngo by Siemens Medical Solutions was first used to read the CT DICOM files and save two videos (avi format) using brain (level=40, width=120) and bone (level=700, width=3200) windows, respectively. Second, a custom tool was implemented in Matlab and used to read the avi files, record the radiologist annotations, delineate hemorrhage region and save it as white region in a black 650x650 image (jpg format). Gray-scale 650x650 images (jpg format) for each CT slice were also saved for both windows (brain and bone). Also, the 650x650 masks were mapped back to 512x512 to be used with the raw CT scans.
Files and folders in the dataset are:
patient_demographics.csvcontains the patient #, age and gender, the ICH subtypes if ICH was diagnosed, and a skull fracture if it was diagnosed for each CT scan.
hemorrhage_diagnosis.csvcontains the patient #, slice #, the ICH subtypes if ICH was diagnosed, and a skull fracture if it was diagnosed for each slice.
Patients_CT folder contains the CT scans in JPG format, organized as follows:
- Sub-folder names match the patients numbers in the patient demographics file
- Each sub-folder further contains:
- all the brain-window CT slices for each patient, numbers of CT slices started from 1.
- the segmentation of the slices that have hemorrhage (named #HGE_Seg)
- all the bone-window CT slices for each patient, numbers of CT slices started from 1.
Raw_ct_scans folder contains the CT scans in NIfTI format, organized as follows:
ct_scansfolder contains the NIfTI scans for the patients in the patient demographics file except subject # 59 to 65.
masksfolder contains the segmentation of the 512x512 slices that have ICH in each NIfTI CT scan. The sub-folders names match the patients numbers in the patient demographics file for the subjects with ICH.
hemorrhage_diagnosis_raw_ct.csvcontains the labels for each slice in the NIfTI CT scans. Note: the raw scans have more slices than the scans in
Patients_CTfolder, so use this annotation csv file with the raw scans.
Python 3.5+ code,
split_data.py, is provided. This code loads the gray scale images of the brain window, resizes them, and saves them into sub-folders. An environment file is provided,
ct_ich.yml, which specifies the virtual environment that was used to execute the code. The virtual environment can be recreated using conda as follows:
conda env create -f ct_ich.yml
This will create the
ct_ich virtual environment for running the code.
split_raw_data.py is provided to load the NIfTI CT scans, rotate them by 90 degrees counterclockwise to match them with the masks and window them using a brain window.
In this release, the authors added the deidentified CT scans in NIfTI format.
Conflicts of Interest
The authors declare no conflict of interest.
- Taylor, C. A., Bell, J. M., Breiding, M. J., & Xu, L. (2017). Traumatic brain injury–related emergency department visits, hospitalizations, and deaths—United States, 2007 and 2013. MMWR Surveillance Summaries, 66(9), 1.
- van Asch, C. J., Luitse, M. J., Rinkel, G. J., van der Tweel, I., Algra, A., & Klijn, C. J. (2010). Incidence, case fatality, and functional outcome of intracerebral haemorrhage over time, according to age, sex, and ethnic origin: a systematic review and meta-analysis. The Lancet Neurology, 9(2), 167-176.
- Currie, S., Saleem, N., Straiton, J. A., Macmullen-Price, J., Warren, D. J., & Craven, I. J. (2016). Imaging assessment of traumatic brain injury. Postgraduate medical journal, 92(1083), 41-50.
- Xue, Z., Antani, S., Long, L. R., Demner-Fushman, D., & Thoma, G. R. (2012). Window classification of brain CT images in biomedical articles. In AMIA Annual Symposium Proceedings (Vol. 2012, p. 1023). American Medical Informatics Association.
- Litjens, G., Kooi, T., Bejnordi, B. E., Setio, A. A. A., Ciompi, F., Ghafoorian, M., ... & Sánchez, C. I. (2017). A survey on deep learning in medical image analysis. Medical image analysis, 42, 60-88.
- Chilamkurthy, S., Ghosh, R., Tanamala, S., Biviji, M., Campeau, N. G., Venugopal, V. K., ... & Warier, P. (2018). Deep learning algorithms for detection of critical findings in head CT scans: a retrospective study. The Lancet, 392(10162), 2388-2396.
Anyone can access the files, as long as they conform to the terms of the specified license.
License (for files):
Creative Commons Attribution 4.0 International Public License