Unsupervised caries detection in non-standardized bitewing dental X-Rays.
Abstract
In recent years dental image processing has become a useful tool in aiding healthcare
professionals diagnose patients by reducing some of the problems inherent with dental
radiographs. Despite advances in the eld, accurate diagnoses of dental caries using
Comptuer-aided Diagnosis (CAD) tools are still problematic due to the non-uniform
nature of dental X-rays. The reason as to why accurate diagnoses are problematic is
in part due to exisiting systems utilizing a supervised learning model for their diagnostic
algorithms. Using this approach results in a detection system which is trained to
identify caries under speci c conditions. When the input images vary greatly from the
training set, these systems have a tendency to misdiagnose patients or miss possible
caries altogether.
A method for the segmentation of teeth in periapical X-Rays is presented in this dissertation
as well as a method for the detection of caries across a variety of non-uniform X-ray
images using an unsupervised learning model. The diagnostic method proposed in this
dissertation uses an assessment protocol similar to how dentists evaluate the presence
of caries. Using this assessment protocol results in caries being evaluated relative to the
image itself and not evaluated relative to a set of identi ers obtained from a learning
model. The viability of an unsupervised learning model, and its relative e ectiveness
of accurately diagnosing dental caries when compared to current systems, is indicated
by the results detailed in this dissertation. The proposed model achieved a 96% correct
diagnostic which proved competitive with existing models.