TBI-HemoCT data set
Description
The TBI-HemoCT dataset consists of more than 60,000 brain CT images describing 6 types of traumatic intracranial hemorrhages. The data was obtained at Shriman Superspecialty Hospital, Jalandhar, where TBI patients who underwent the non-contrast CT head scan were taken during a specified time. The original patient scans stored as DICOM format consist of each case having about 150-200 axial slices, depending upon the length of the scan and the scanner vials. In order to make the files machine-learning friendly as well as accessible to other image processing applications, the DICOM files were decoded into the PNG format with the help of a custom script written in Python. After conversion, all the PNG pictures were reconsidered and labeled individually by a trained radiologist and grouped into one of the following six groups relative to the existence and sort of hemorrhage shown in the picture.[3] 1. Any – Contains any form of visible hemorrhage (general abnormality indicator) 2. Epidural Hemorrhage (EDH) – Biconvex, lens-shaped bleeding located between the dura mater and the skull 3. Intraparenchymal Hemorrhage (IPH) – Bleeding within the brain tissue 4. Intraventricular Hemorrhage (IVH) – Bleeding into the brain’s ventricular system 5. Subarachnoid Hemorrhage (SAH) – Bleeding in the subarachnoid space surrounding the brain 6. Subdural Hemorrhage (SDH) – Crescent-shaped bleeding beneath the dura mater Total labeled PNG image is larger than 60,000 and each category folder has around 10000 CT slice images. Since it is a folder-based classification, the use of deep learning is performed easily with image classification, anomaly detection, segmentation, or hemorrhage type recognition.[4] The images retain original clinical characteristics such as grayscale intensity, anatomical orientation, and slice thickness, though no patient metadata or clinical notes are included to maintain full anonymity. The data do not contain any personal identifiers and comply with standard de-identification protocols. The scheme of the data set layout and some example of labeled images belonging to each category could be looked at in the supplementary figures (see Figure 1 and 2). The data has since been made publicly available through Gold Access library on Mendeley where all interested researchers can access, compare, and develop a diagnostic machine-based tool and clinical decision support system that is based on this resource.
Files
Institutions
- Lovely Professional University