Data Info¶
Dataset collection and characteristics¶
Our dataset was collected from Dokuz Eylul University Radiology Department’s Picture Archiving and Communication System (PACS). 40 Abdominal CTA volumes consisting of 12-bit DICOM images with 512 x 512 resolution. Slice thickness vary from 2 to 3.2 mm with 90 slices per volume on average. Pixel spacing is around 0.68 x 0.68 mm. These CT scans belong to 22 female and 13 male potential liver donors. The patient ages ranged from 18 to 57, with an average of 37 years.
Challenges¶
The challenges of the dataset are mostly stem from variation in image quality and contrast levels. The reasons of these variations can be summarized as follows:
1. The miscalculation of timing: The time period between the injection of the contrast agent and the beginning of the scanning is miscalculated. Therefore, contrast agent is not in the desired location of the vessel during the scanning and enhancement of the vessels is not achieved. (See Fig 1a)
2. Artefacts: Artefacts and noise might further drop the quality of rendering. Hepatic and portal vessels may look connected, some parts of parenchyma may appear brighter and look like vessels, noise may obscure the appearance of vessels. (See Fig.1b
3. Interslice distance: High slice thickness alters the adequate appearance of vessels in 3D. (See Fig.1c)
4. Transient hepatic attenuation: Transient hepatic attenuations are areas of enhancement on CTA that occur as results of localized variations in the proportion of hepatic arterial and portal venous blood supply. (See Fig.1d)
5. Non-uniform attenuation: Non-uniform blood velocity into vessel branches complicates the vascular imaging. A short acquisition time is preferable in order to obtain an uniform opacification on vessels. (See Fig. 1e)
Besides the angiographic acquisition, computed tomography itself has general drawbacks and limitations. Physics-based artefacts, such as are Beam hardening, partial volume effect and undersampling may seriously degrade the imaging quality. Fig.1f presents a slice with beam hardening caused due to CT acquisition. Besides, patient-based artefacts result from such reasons as patient movements.
Fig. 1: Examples of quality losses and artifacts.
Training and Testing Data¶
The training data that contains CTA images, mask images for liver and ground truth will be shared with the participant before the challenge. The data will be available after online registration of participants and signing a letter of intent. The training data is selected to contain examples of the challenges mentioned in the previous section.
Remaining data will be the test dataset. Ground truth for this data will not be shared. The results need to be submitted as binary or labeled data.