Accelerating your AI Success
Explore
For data curation services, we cooperate with medical professionals of various experience. We have teams, varying from interns to seasoned medical specialists, of 50+ people in total, with possibility to ramp up a team upon request. Apart from radiologists, we have access to a range of medical specialists such as dermatologists, ophthalmologists, cardiologists, etc.
Bounding boxes are the most common type of datasets training. They are used to segment the area with some particular pathology.
Main advantages of using bounding boxes are high speed of annotation and thus, a lower cost.
medDARE’s team has been working with bounding boxes on multiple projects before, including lung CTs annotation, brain MRIs annotation, X-Ray spine annotation of different vertebrae.
3D segmentation is the most complex type of data training. It requires annotation of objects on all three dimentions of the DICOM file.
This type of annotation is very time consuming, and based on the number of organs or pathologies which needs to be annotated, it can take up to 8-10 hours for annotation of one dataset.
Usually, 3D segmentation is done in 3D Slicer.
2D segmentation is used when there’s a need to contour the exact shape in each slice of a DICOM file. These segments can be extrapolated to different slices.
2D segmentation is more time-consuming than the bounding boxes segmentation but gives more precise results. This type of data training is often used for small pathologies, for example, in brain MRI or for the annotation of vascular system.
