The diagnosis of certain spine pathologies, such as scoliosis, spondylolisthesis
and vertebral fractures, are part of the daily clinical routine. Very
frequently, MRI data are used to diagnose these kinds of pathologies in
order to avoid exposing patients to harmful radiation, like X-ray.
Developing a segmentation system for an array of vertebrae is complex,
so the method was first tested on brain tumors of types glioblastoma
multiforme and pituitary adenoma. A small triangular surface mesh at
the approximate center of the tumor is inflated towards the boundary
using balloon force, keeping it approximately star-shaped. The boundary
is implicitly binarized by the inflation rules, based on the minimum and
maximum intensity from the initialization step. After the segmentation is
finished, the tumor volume is calculated.
The spine segmentation system uses a bottom-up approach for detecting
vertebral bodies based on just one manual initialization. A subdivision
surface hierarchy is introduced as an efficient global-to-local smoothness
constraint, which can be thought of as an internal force. Together with
intensities, low-high (LH) values were initially used to ease boundary
finding, but the boundary estimation evolved into a multi-feature combiner.
The final system utilizes a Viola-Jones detector to determine centers
and approximate sizes of vertebral bodies. This gives the user a chance
to manually correct detections, enables parallel feature calculation and
segmentation, and is a basis for reliable diagnosis established at the end.
The system was evaluated on 26 lumbar datasets containing 234 reference
vertebrae. Vertebra detection has 7.1% false negatives and 1.3% false
positives. The average Dice coefficient to manual reference is 79.3% and
mean distance error is 1.77 mm. No severe case of the three addressed
illnesses was missed, and false alarms occurred rarely – 0% for scoliosis,
3.9% for spondylolisthesis and 2.6% for vertebral fractures.
The main advantages of this system are high speed, robust handling
of a large variety of routine clinical images, and simple and minimal user
interaction.