Semi-automatic teeth segmentation in 3D models of dental casts using a hybrid methodology - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 362

Acknowledgments
This work was made possible thanks to the support of the Universidad Nacional de Colombia
Sede Manizales, and in particular the research groups on Perception and Intelligent Control
(PCI) and Soft and Hard Applied Computing (SHAC).
1 Introduction
The advent of new technologies has driven the development of increasingly sophisticated 3D
acquisition systems [ 1 ]. 3D models originated by using those systems, also referred as point
clouds, provide surface information, metrics, texture, etc. This 3D information is of great in-
terest in different fields in the industry, such as in-process inspection, accident reconstruction,
crime scene analysis, machine calibration, orthodontics, etc. [ 2 ] .
Digital dental models have proven to be helpful and important for experts in odontology
and orthodontics. They provide information precise enough to be used in diagnosis and pro-
gnosis [ 3 ] , this has caused a number of concerns to emerge using 3D dental models is becom-
ing a more common practice in the ield.
Colombian laws in consumer protection require dental cast records of patients to be pre-
served during a period no less than 10 years, and according to the Association of Orthodont-
ists [ 4 ] , it is recommended that study models are retained for at least 11 years or until the pa-
tient is 26 years old. This has caused a number of concerns to emerge, like limited storage ca-
pacity, model fragility, among others.
Therefore, it is important to use the 3D digital dental models in place of physical alginate
casts. Aiming to further strengthen the acceptance of such practice, 3D image processing and
digital measuring tools are finding their way into orthodontics.
Surface and object segmentation is an intermediate step in artificial vision that eases object
recognition and classification. A good segmentation is a key to facilitate, enhance, and achieve
further interpretation of the input data.
For that reason, in this work we present an exploratory analysis of current segmentation
techniques for point clouds applied to dental 3D models. This is a first step toward parameter
measurement [ 5 , 6 ], simulation of the movement of teeth to correct malocclusions [ 7 ] , planning
for dental and maxillofacial surgery [ 8 ], pose estimation [ 9 ], among others.
This chapter is organized as follows: Section 2 describes briefly the process of obtaining in-
tegrated 3D point clouds from plaster dental models. Section 3 introduces the segmentation
techniques applied to 3D digital dental models. Section 3.4 shows the analysis and results of
applying those segmentation techniques to point clouds of dental models. Finally, Section 3.5
presents the findings and discussion of the results of this work.
2 Dental Study Model
Experts in dental areas use diagnostic logs, which are kept in order to document the initial
condition of the patient and complement the information gathered during clinical examina-
tion. These records are commonly divided in three categories: dental models, photographs,
and radiographs [ 10 ] .
Dental models in dentistry are built primarily using alginate cast. They are important for dia-
gnosis and orthodontic treatment planning, as well as to detect anomalies of pose, size, and
shape of the teeth. Also, they are indispensable to assess the outcomes of treatment process
[ 11 - 13 ] .