Jason J. Corso
|
Lumbar Imaging: Intervertebral Disc Localization, Labeling, and Diagnosis
People: Raja Alomari, Jason Corso, Vipin Chaudhary, and Gurmeet Dhillon
According to the National Institute of Neurological
Disorders and Stroke (NINDS), back pain is the second
most common neurological ailment in the United States after
headache. Over 12 million Americans have some sort of Intervertebral
Disc Disease (IDD). Localization and labeling
of the vertebral column anatomical structures has thus been
a focus of recent studies, due to the high demand
of analysis of the vertebral column structures such as disc
size, disc shape, and water content percentage in discs. This
analysis is a core requirement for the diagnosis of the vertebral
column as a whole and for anatomical structures such as discs,
vertebrae and soft tissues.
Localization and Labeling
Accurate labeling of the backbone structures is a necessary step for
performing any type of analysis, diagnosis, or surgical planning. One key use
of labeling is the design of a computer aided diagnosis system for lumbar
area. In the clinical practice, the neuroradiologist reports the diagnosis at
each disc level. However, although accurate labeling of the backbone
structures is necessary, the backbone has wide variabilities including
degree of bending of the vertebral column, sizes, shapes, count (number) and
appearances of discs and vertebrae. In addition, existing abnormality
conditions such as vertebral fusion, degenerative disc diseases, spinal
infections, and spinal scoliosis add more variability.
We have developed a two-level probabilistic model for such disc
localization and labeling (illustration below). Whereas conventional
labeling approaches (e.g., our approach to brain tumor above) define all
models at the pixel level, our model integrates both pixel-level
information, such as appearance, and object-level information, such as
relative location and shape. Utilizing both levels of information adds
robustness to the ambiguous disc intensity signature and high structure
variation. Yet, we are able to do efficient (and convergent) localization and
labeling with generalized expectation-maximization. We have presented
accurate results, about 89% accuracy on 105 normal and abnormal cases (96%
when using normal alone and 87% when using abnormal alone).
Diagnosis:
Various diseases that affect the vertebral column are usually painful and
influence the patientŐs everyday life. We are concerned with the following
abnormalities: disc herniation, spinal stenosis, disc degeneration, disc
desiccation, and spinal infection. Automated detection and diagnosis of
these abnormalities holds great promise is in clinical practice, especially
with the increasing incidents of back problems outpacing the numbers of
radiologists.
We have investigated numerous approaches to diagnose the presence of an abnormality, the class of the abnormality, and quantification of the abnormality's parameters. Our results are described in full detail in the publications listed below.
Other Info:
Publications:
|