Information Technology Reference
In-Depth Information
2 Background of Computer Aided Detection
Detection of spinal metastases on CT can be a challenge, as the spinal column is a
physically extended and complex structure. Lesion presentation may be subtle and
unexpected. Adding to this complexity, in the current clinical practice environment,
multiple high-throughput CT scanners can produce numerous patient studies in a
short time interval, each with thousands of images, restricting time for image
assessment. Numerous anatomic structures in each image must be assessed for
pathology, typically at multiple window/level settings, effectively increasing the
number of images to be reviewed geometrically.
Computer Aided Detection (CAD) researchers have been active in the past two
decades and have showed very promising results. CAD is a potential solution to the
challenge of rapid assessment of very large datasets. CAD is a software tool that can
detect, mark, and quantitatively assess potential pathologies for further scrutiny by a
radiologist. The CAD system should be topically focused with the ability to facil-
itate rapid and accurate assessment of important anatomy or high risk pathology
subsets of interest in the CT data, such as the spine. While the
final diagnosis is
made by the radiologist responsible for interpreting the examination, CAD can help
improve sensitivity in identifying lesions potentially overlooked by radiologists
in situations of data overload or fatigue.
Figure
2
is a block diagram of a typical CAD system. A typical CAD system has
two phases: training and testing. In the training phase, a set of training data is
rst
Clinical images
Image segmentation
Feature Extraction
Training images
Feature Selection
Detection rules
Classifier
Radiologist's diagnostic
knowledge
Preliminary detections
Radiologist's final decision
Training phase
Testing phase
